Fix crashes found hunting the last open fuzzing record (#8524)

* specialize: check the member descriptor's type before caching its slot offset

The LOAD_ATTR/STORE_ATTR specializations cached the slot offset of any member
descriptor found on the owner's type and then guarded the specialized
instruction on the type version alone, while descr_get()/descr_set() check on
every access that the instance belongs to the type the descriptor was defined
for. A descriptor taken from a wider class and bound to a narrower one read
past the instance's slot array once the cache warmed up:

    class Big:
        __slots__ = ("a0", "a1", "a2", "a3", "a4", "a5", "a6", "a7")
    class Narrow:
        __slots__ = ("z",)
    Narrow.x = Big.__dict__["a7"]
    o = Narrow()
    for _ in range(1000):
        try: o.x
        except TypeError: pass
    # index out of bounds: the len is 1 but the index is 7 (object/core.rs)

A class with no slots at all reached the ext_ref().unwrap() on the same line.

Assisted-by: Claude

* socket: reserve recv()'s buffer fallibly

recv() and recvfrom() handed the caller's bufsize straight to
Vec::with_capacity, so an unreachable size aborted the process through
handle_alloc_error before any syscall was made:

    socket.socket().recv(2**62)
    # memory allocation of 4611686018427387904 bytes failed -> SIGABRT

try_reserve_exact reports MemoryError instead, which is what CPython raises.

Assisted-by: Claude

* types: count the __call__ and __get__ slot dispatches as recursion

Both wrappers re-enter Python without pushing a frame, so nothing counted the
nesting when the special method named the object it was looked up on:

    class C: pass
    c = C(); C.__call__ = c
    c()                                    # native stack overflow, SIGSEGV

    class D: pass
    d = D(); D.__get__ = d; D.x = d
    d.x                                    # the same, through descr_get

with_recursion around the two dispatches raises RecursionError instead, the
way Py_EnterRecursiveCall bounds a tp_call dispatch. It costs about 5% on a
__call__ dispatch and 3% on a __get__ dispatch through these wrappers.

Assisted-by: Claude

* typevar: show a ParamSpecArgs origin by its repr

ParamSpecArgs and ParamSpecKwargs fell back to a Rust `{:?}` of __origin__ when
it had no __name__. That walks the object graph natively through Debug for
PyInner, where no recursion guard sits, so a single repr() of a deeply nested
chain overflowed the native stack:

    a = object()
    for _ in range(30000):
        a = typing.ParamSpecArgs(a)
    repr(a)                                # SIGSEGV

The origin is formatted with its repr now, which is guarded, and a ParamSpec
origin is recognized by its type rather than by carrying a __name__.

Assisted-by: Claude

* Do not hold a lock across a call back into Python

Three places kept a lock while running code that can reach the same object, so
a callback that touched it wedged the process:

    _asyncio.future_add_to_awaited_by(fut, waiter)   # waiter.__hash__ adds again
    select.select(elements, [], [], 0)               # fileno() clears `elements`
    select.poll().poll(1000)                         # SIGALRM handler registers

The future's awaited-by field is read and written under its lock but the set is
built outside it, the list extraction re-reads the list on each step the way
map_iterable_object() does, and poll() waits on a copy of its descriptors. All
three ran forever before and now finish the way they do on CPython.

Assisted-by: Claude

* Validate memoryview.cast() arguments and export negative strides correctly

cast() accepted any struct format and any shape element. A zero-size
format ('0s') and a 0 in the shape both reached a division by zero;
cast() now takes only a native single character format, optionally
'@'-prefixed, and shape elements that are ints greater than zero.

A view with a negative stride starts at its last item, so the bytes it
exported began there and its own offsets walked off the front of them.
Such a view now exports the whole underlying buffer with `start` folded
into the descriptor's offsets, and zip_eq() hands over a whole run only
when both sides are contiguous in the last dimension.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* Charge native recursion to the stack, not to the frame limit

with_recursion() checked the limit sys.setrecursionlimit() sets and
incremented the same counter that pushing a frame does, so a guard on a
native dispatch spent what Python code had left to call with, and did so
where sys._getframe() cannot see it: test.support.get_recursion_available()
reported frames that were no longer there. Py_EnterRecursiveCall bounds
the native stack instead, which is a separate budget, and the C stack
check with_recursion already performs is that bound.

The snippets pinning the guarded paths nest deep enough to reach the
stack rather than the frame limit.

Assisted-by: Claude

* Report a size that cannot be allocated instead of aborting on it

A size taken from Python went straight into an infallible allocation in
several places, so the process aborted through handle_alloc_error before
any exception could be raised:

- str/bytes/bytearray center(), ljust(), rjust() and zfill() reserved
  the padded result for the caller's width
- expandtabs() built its runs of spaces from a tabsize of any width;
  the argument is a C int, and a wider one does not fit
- Buffered{Reader,Writer,Random} allocated buffer_size, and read(),
  read1() and FileIO.read() their read size
- bytes(n) and bytearray(n) allocated n
- pbkdf2_hmac() allocated the derived key length, which is a C int

Each of these now reports MemoryError, or OverflowError where the
argument does not fit the type it is declared with.

new_zeroed_bytes() leaves the zeroing to the allocator, so a large
request costs the pages that are written to rather than all of them.

Assisted-by: Claude

* marshal: answer allow_code where a code object is written or read

allow_code was answered by walking the whole result a second time, with
no depth counter and no record of what it had already seen, so a value
that referred back to itself or nested deeply enough ran off the native
stack. w_object() and r_object() answer it where the code object is,
inside the walk that already bounds its depth and resolves references.

A container length is read the way r_long() reads one: it is signed, so
a length with the top bit set is out of range rather than four billion
items to reserve room for.

load() no longer holds a borrow of the buffer read() returned across the
seek() it makes afterwards.

Assisted-by: Claude

* Do not lock an object while running code that can reach it

Several places held a lock or a borrow of an object across a call back
into Python, so a callback that touched the same object waited on a lock
its own caller was holding:

- memoryview slice assignment read a source overlapping the destination,
  and __setitem__ converted the value while holding the write borrow
- BytesIO.readinto() read into a buffer viewing the same BytesIO
- array.__setitem__ converted the value under the array's write lock,
  and mmap.write() read a source viewing the same map
- bytearray.join() and bytearray.__mod__ drove Python with the
  bytearray borrowed
- array and bytearray answered "is this resizable" after taking the
  write lock, though an export is exactly a borrow someone else holds

A TextIOWrapper cookie now has to name a position inside what was
decoded in characters as well as in bytes; only the byte offset was
checked, and the character count is what read() and tell() index with.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* Stop asserting a pbkdf2 message that depends on the width of a C long

The snippet asserted "key length is too great.", which pbkdf2_hmac()
only reaches once the length has been converted; where a C long is
narrower than the length asked for, the conversion fails first and says
so instead. Both are OverflowError, which is what the case is about.

test_support.test_get_recursion_depth passes now that a native recursion
guard no longer spends frames get_recursion_depth() cannot see.

Assisted-by: Claude

* Publish and read the same pointer for a thread's top frame

set_current_frame() casts the `Py<FrameObject>` it publishes straight to
`*mut FrameObject`, so ThreadSlot::top_frame holds the object's base.
sys._current_frames() read it back through Py::from_payload_ptr(), which
subtracts the payload offset from what it is given. The reference it took
therefore incremented, and later decremented, a word 48 bytes ahead of
the frame -- inside the object allocated before it, whose OnceLock state
word sits exactly there for two frames adjacent in the size class. The
neighbour then read an initialized-looking cold pointer that had never
been written and locked whatever the uninitialized word addressed, so
the thread that owned it crashed rather than the one that read.

The slot now holds `*mut Py<FrameObject>`, which is what both sides mean.

A thread parked in a call has no FrameObject for its topmost frame, so
top_frame is null there and the reader takes the materialize path
instead: test_sys.test_current_frames never reaches the branch. The
snippet takes _current_frames() against threads that are running.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* Decide stop-the-world parking under the thread registry lock

do_suspend() published SUSPENDED first and only then re-read `requested`,
restoring itself to ATTACHED if the stop had ended in the meantime. That
made a thread the second writer able to leave SUSPENDED, so a stop whose
completion check had already observed the thread parked could be undone
behind the requester's back:

  worker    CAS ATTACHED -> SUSPENDED
  requester all_non_requester_suspended() -> true, world_stopped = true
  requester start_the_world(): requested = false, then walks the registry
  worker    reads requested == false, stores ATTACHED

With the store landing inside that walk the debug assertion in
start_the_world fires; with the walk already past the slot, a following
stop force-parks the thread DETACHED -> SUSPENDED, counts it as stopped,
and the store then puts it back to ATTACHED with the world declared
stopped and the thread running bytecode.

`requested` is set in init_thread_countdown() and cleared in
start_the_world() with the registry held, and start_the_world() keeps
holding it while releasing every SUSPENDED thread. Taking the registry
around the check and the transition therefore makes the two orders the
only ones possible: park before that release pass and be woken by it, or
find the request already withdrawn and stay ATTACHED. The requester is
left as the only writer that takes a thread out of SUSPENDED, and the
self-restore is gone.

suspend_if_needed() takes the VirtualMachine to reach the registry.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* Keep an atexit callback alive while it is being compared

atexit.unregister() releases the callback list around each __eq__ call and
identified the entry it had compared by the address of its Box. __eq__ can
call atexit._clear(), which drops that Box, and atexit.register(), whose
new Box lands on the freed allocation; the identity search then matched
the freshly registered callback and removed it.

  atexit.register(a); atexit.register(b); atexit.register(c)
  # __eq__ runs _clear() then register(d), returns True
  atexit.unregister(probe)

left no callbacks registered where CPython leaves d.

Entries are Arc-shared now, so unregister() holds the one it is comparing
and matches it with Arc::ptr_eq: an address cannot be reused while the
comparison that named it is still running.

Assisted-by: Claude

* Hold atexit entries in PyRc rather than Arc

PyObjectRef is Send and Sync only under the threading feature, so an Arc
over a callback entry trips clippy::arc_with_non_send_sync in builds
without it, such as the wasm package. PyRc is Arc there and Rc otherwise.

Assisted-by: Claude

* Do not hold a buffer's storage while waiting for a peer

FileIO.readinto() and socket.recv_into()/recvfrom_into() took the target
buffer's write borrow and kept it for the whole call, including the wait
for data a pipe, socket or terminal may never deliver. What CPython holds
across that wait is the export, which only forbids resizing; the borrow is
a lock every other thread touching the same object waits on, so

    threading.Thread(target=lambda: sock.recv_into(buf)).start()
    len(buf)

did not answer until the peer sent. A thread parked on that lock is
ATTACHED and never reaches a safepoint, so gc.collect() in a third thread
waited for the peer as well: one incidental read of the buffer stopped the
world from being stopped at all.

The wait now runs against storage of its own and the bytes are copied over
once they arrive, with the export held throughout so the target still
cannot be resized meanwhile. A seekable file answers from itself rather
than from a peer, so FileIO.readinto() writes into the target directly
there and the buffered read path is unchanged.

Assisted-by: Claude

* Do not hold a buffer's storage while waiting to hand it over

socket.send()/sendall()/sendto()/sendmsg() and FileIO.write() kept the
source buffer's read borrow for the whole call, including the wait for a
peer that may never make room. That borrow is a lock every other thread
writing to the same object waits on, so

    threading.Thread(target=lambda: sock.sendall(buf)).start()
    buf[0] = 1

did not return until the peer read; and a thread parked there is ATTACHED
and never reaches a safepoint, so gc.collect() in a third thread waited
for the peer too -- the same wedge readinto() had on the receiving side.

ArgBytesLike::borrow_buf_unlocked() answers with bytes that survive the
borrow being dropped. An immutable object hands out a plain reference and
locks nothing, so those are sent where they lie and bytes and memoryviews
over them cost nothing; only bytes reached through a lock are copied out
first. The export is held throughout either way, so the source still
cannot be resized while it is being sent.

The regression snippet covers both directions now and is renamed for it.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* Check select()'s descriptor limit while the sequence is walked

seq2set() collected the whole sequence and compared the result's length
against FD_SETSIZE afterwards. Selectable::try_from_object() calls
fileno(), which runs Python and can append to the list being walked, and
the walk re-reads the list on every step, so the collection had no end to
reach and the comparison was never made. seq2set in Modules/selectmodule.c
checks the count per element instead.

stdlib_select.py gains a fileno() that appends to its own list, and
releases its sockets with close() rather than by dropping the name.

Assisted-by: Claude

* Bound native recursion where the C stack cannot be measured

check_c_stack_overflow() answers no unconditionally under miri and on
musl, where the stack pointer is not read. Since 9c9905aff that check is
all with_recursion() does, so every guard placed on native recursion --
__call__ and __get__ dispatch among them -- was a no-op on those targets
and the nesting ran until the stack ran out.

with_recursion() now counts its own depth on those targets and refuses
past NATIVE_RECURSION_LIMIT_UNMEASURED. The count is separate from the
frame limit sys.setrecursionlimit() sets, and compiles away where the
stack pointer can be read.

Assisted-by: Claude

* Report the failure to allocate pbkdf2's key and Take.readinto's scratch

Both buffers are sized from an argument -- pbkdf2_hmac's dklen accepts up
to i32::MAX, and readinto's from the length of the destination -- and were
built with vec![0u8; n], which aborts the process on allocation failure.
new_zeroed_bytes() raises MemoryError instead.

Assisted-by: Claude

* Read the frozen-code tuple length as a signed length

The '(' branches in read_marshal_str_vec() and read_marshal_const_tuple()
took the length with read_u32() as usize, so a value with the top bit set
read as four billion items rather than as out of range. read_len() is what
every other length in this file goes through, and it reinterprets as i32.

Both readers serve deserialize_code(), which reads only the frozen modules
baked in at build time, so this changes no reachable behavior;
marshal.loads() already went through read_len().

Assisted-by: Claude

* Make the snippets from the fuzzer sweep assert what they check

builtin_str.py expanded a tab to 2**31-1 columns, allocating 2 GiB to
observe that the width is accepted; a string with no tab observes the same
acceptance without laying anything out.

stdlib_array.py caught the refusal of frombytes() on its own exported
buffer and passed silently when nothing was raised. stdlib_hashlib.py
used a bare `assert False` as its failure branch. Both now say so through
the same shapes the other snippets use.

stdlib_socket.py also accepts OverflowError from recv() with a size that
does not fit the platform's C int.

stdlib_threading_current_frames.py indexed the frame chain for "f123"
without first asserting it is there.

Assisted-by: Claude

* Ask for a marshal container's room instead of assuming it

A flagged tuple or list is published in the reference table before its
children are read, and the placeholder was built with vec![none; len].
The length is the input's to choose and read_len() lets it reach
i32::MAX, so marshal.loads(b"\xa8\xff\xff\xff\x7f") -- five bytes -- asks
for 17 GB of element slots and aborts the process where the allocator
cannot serve it. r_object() allocates the container up front too, but
PyTuple_New() reports what it cannot get.

The elements are now reserved with try_reserve_exact() and a refusal is
raised as MemoryError through the decoder's pending-error channel.
PyTuple::new_marshal_placeholder() held nothing but that allocation and
is gone; the caller builds the elements and uses new_ref().

Assisted-by: Claude

* Compare the blocking-buffer snippet against something

measure() asserted len(buf) == len(buf), which holds whatever the length
is; it now takes the length the caller expects. The pipe case compared
the drained total against the whole source, while an unbuffered write()
reports only what it transferred and a signal can cut that short; it now
compares against what write() returned.

Assisted-by: Claude

* Refuse array.frombytes() a source whose items are not bytes

frombytes() reads its argument as bytes, but accepted any contiguous
buffer, so array("i").frombytes(memoryview(array("d", [1.0]))) appended a
double's bytes read as ints instead of raising. array_array_frombytes_impl
requires an itemsize of 1; ArgBytesLike now reports the itemsize so the
same check can be made here.

The BufferError that a resize meets while the array is exported also names
the array rather than repeating bytearray's wording.

stdlib_array.py's frombytes-of-itself case used typecode "i", where the
new check answers before the resize guard is reached; it uses "b" so the
guard is what refuses, and asserts the wider case separately.

Assisted-by: Claude

* Tell apart the ways marshal data can be bad

A type byte no reader knows, a back reference that names nothing, and
TYPE_NULL all came out as a bare "bad marshal data" ValueError. r_object()
answers the first two with "bad marshal data (unknown type code)" and
"bad marshal data (invalid reference)", and read_object() answers the
third with a TypeError, "NULL object in marshal data for object", since
TYPE_NULL stands for no object rather than for a value.

MarshalError gains the three cases and deserialize_value() maps them.
The container-specific wording r_object() uses for a NULL read inside a
tuple or list is not reproduced; the exception type is.

Assisted-by: Claude

* Hash the collector's tables by address rather than by SipHash

A collection keys three sets and two maps by object address, and it visits
every tracked object and every edge between them, so the hashing is a
per-edge cost. Those tables used the default RandomState, whose SipHash
buys resistance against a caller choosing colliding keys -- and nothing
chooses these keys: they are addresses this process handed out into tables
that live and die inside one collection. A profile of gc.collect() over a
423k-object heap spent 45% of its samples in SipHash.

They now hash with a splitmix64 finalizer. The shifts matter: a table
picks its bucket from the low bits and an address arrives with those bits
zeroed by alignment, so a plain multiply leaves every object in a handful
of buckets and is slower than SipHash was.

The reachability walk also copied each object's referent vector out of the
map it was cached in, a second pass over every edge; it reads them in
place, and reference subtraction hands its vector to the map instead of
cloning it.

Measured over 423k live objects: 0.93s to 0.15s. Over 843k dead ones:
3.00s to 0.79s. extra_tests/snippets/stdlib_threading_gc_import.py, whose
collector thread calls gc.collect() in a loop, ran anywhere from 2.7s to
28s and now runs in 3.1-3.5s: a collection that takes longer leaves more
garbage for the next one to walk, so the cost fed back on itself.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* Keep the collection's candidates and their counts in one table

A collection built a set of candidates and, beside it, a map from the same
addresses to their reference counts. Both were probed for every edge in
the heap -- membership from the set, the count from the map -- so each
edge paid to hash the same address twice, and each candidate paid to be
inserted twice. The map alone answers both questions.

The candidates also keep a walkable order now, which the reference
subtraction pass needs since it writes the counts while reading the
candidates, and which the unreachable set is built from instead of a
set difference.

Over the 423k-object heap measured in the previous commit: 0.15s to
0.13s live, and 0.79s to 0.49s dead.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* gc: collect referents into one buffer instead of a vector per object

Step 3 allocated a `Vec` for every tracked object to hold its referents
and kept them all in a map until step 4 read them back. The referents now
go into a single growing buffer, with the map holding each object's range
into it.

Adds `PyObject::gc_extend_referent_ptrs`, which appends to a caller's
buffer; `gc_get_referent_ptrs` calls it with a fresh one.

Assisted-by: Claude

* memoryview and struct: match the checks and errors of the reference

memoryview:
- `cast()` accepted a source and destination that are both item types,
  which reinterprets the items rather than re-dividing the bytes; one side
  now has to be a byte format.
- A cast to `shape=()` returned without checking that the buffer holds
  exactly the one item that shape describes.
- `hash()` hashes the bytes, so it now raises ValueError for a view whose
  items are not bytes, rather than returning a hash that disagrees with the
  value the view compares equal to.
- `tobytes()` takes the `order` argument, with 'F' walking a
  multidimensional view down its columns; `BufferDescriptor` gained
  `for_each_segment_fortran` for that walk.

struct:
- A value the format has no room for reported "argument out of range"
  instead of naming the format and its range. The format character is now
  passed to the packing functions to report it.
- `Struct.__new__` no longer reads the format; `__init__` does, so
  `__init__` can be called again and a subclass can pass the format up.
  Methods raise RuntimeError until it has run, and `Struct` is a base type.

Removes the expectedFailure from test_Struct_reinitialization and
test_struct_subclass_instantiation.

Assisted-by: Claude
Assisted-by: Codex:GPT-5

* io: decide the readinto path by file type, not by seekability

FileIO.readinto wrote straight into the caller's buffer, holding its write
borrow, when the fd was seekable; otherwise it read aside into scratch and
copied. Seekability stood in for "this read answers without waiting on a
peer", which a pipe on Windows breaks: lseek on one succeeds, so the pipe
took the borrow-holding path and every other thread touching that bytearray
waited for the peer.

host_io::reads_without_waiting answers it directly -- seekability elsewhere,
GetFileType() == FILE_TYPE_DISK on Windows.

The regression snippet times each operation separately, so a failure names
the one that waited; it asserts the transfer is still in flight before
checking the export; and the socket case fills the connection until it
refuses rather than assuming a size that outruns it, which SO_SNDBUF on an
already-connected pair does not settle.

Assisted-by: Claude

Lib/test/test_struct.py

@@ -601,7 +601,6 @@ class StructTest(ComplexesAreIdenticalMixin, unittest.TestCase):
                           'spam and eggs')
         self.assertRaises(struct.error, struct.unpack_from, '14s42', store, 0)
 
-    @unittest.expectedFailure  # TODO: RUSTPYTHON; AssertionError: '>h' != '>hh'
     def test_Struct_reinitialization(self):
         # Issue 9422: there was a memory leak when reinitializing a
         # Struct instance.  This test can be used to detect the leak
@@ -826,7 +825,6 @@ class StructTest(ComplexesAreIdenticalMixin, unittest.TestCase):
         test_error_propagation('N')
         test_error_propagation('n')
 
-    @unittest.expectedFailure  # TODO: RUSTPYTHON
     def test_struct_subclass_instantiation(self):
         # Regression test for https://github.com/python/cpython/issues/112358
         class MyStruct(struct.Struct):

Lib/test/test_support.py

@@ -631,7 +631,6 @@ class TestSupport(unittest.TestCase):
         else:
             self.assertTrue(support.has_strftime_extensions)
 
-    @unittest.expectedFailure  # TODO: RUSTPYTHON; - _testinternalcapi module not available
     def test_get_recursion_depth(self):
         # test support.get_recursion_depth()
         code = textwrap.dedent("""

crates/common/src/borrow.rs

@@ -34,6 +34,17 @@ impl_from!('a, T, BorrowedValue<'a, T>,
 );
 
 impl<'a, T: ?Sized> BorrowedValue<'a, T> {
+    /// Whether reaching the value holds a lock that other threads wait on.
+    ///
+    /// An immutable object hands out a plain reference and answers `false`;
+    /// one whose storage can change hands out a guard. A caller about to wait
+    /// for something unrelated -- a peer, a file, a signal -- can use this to
+    /// decide whether it may keep the borrow for the duration.
+    #[must_use]
+    pub const fn is_locked(&self) -> bool {
+        !matches!(self, Self::Ref(_))
+    }
+
     pub fn map<U: ?Sized, F>(s: Self, f: F) -> BorrowedValue<'a, U>
     where
         F: FnOnce(&T) -> &U,

crates/common/src/str.rs

@@ -592,20 +592,21 @@ pub fn codepoint_range_end(s: &Wtf8, n_chars: usize) -> Option<usize> {
 }
 
 #[must_use]
-pub fn zfill(bytes: &[u8], width: usize) -> Vec<u8> {
+/// Returns `None` for a width whose result cannot be allocated.
+pub fn zfill(bytes: &[u8], width: usize) -> Option<Vec<u8>> {
     if width <= bytes.len() {
-        bytes.to_vec()
-    } else {
-        let (sign, s) = match bytes.first() {
-            Some(_sign @ (b'+' | b'-')) => (unsafe { bytes.get_unchecked(..1) }, &bytes[1..]),
-            _ => (&b""[..], bytes),
-        };
-        let mut filled = Vec::new();
-        filled.extend_from_slice(sign);
-        filled.extend(core::iter::repeat_n(b'0', width - bytes.len()));
-        filled.extend_from_slice(s);
-        filled
+        return Some(bytes.to_vec());
     }
+    let (sign, s) = match bytes.first() {
+        Some(_sign @ (b'+' | b'-')) => (unsafe { bytes.get_unchecked(..1) }, &bytes[1..]),
+        _ => (&b""[..], bytes),
+    };
+    let mut filled = Vec::new();
+    filled.try_reserve_exact(width).ok()?;
+    filled.extend_from_slice(sign);
+    filled.extend(core::iter::repeat_n(b'0', width - bytes.len()));
+    filled.extend_from_slice(s);
+    Some(filled)
 }
 
 /// Convert a string to ascii compatible, escaping unicode-s into escape

crates/compiler-core/src/marshal.rs

@@ -19,6 +19,14 @@ pub enum MarshalError {
     InvalidLocation,
     /// Bad type marker
     BadType,
+    /// A type marker no reader knows
+    UnknownType,
+    /// A back reference that names nothing
+    InvalidRef,
+    /// A marker that stands for no object at all
+    NullObject,
+    /// A container length that is negative or does not fit, named by what it counts
+    BadSize(&'static str),
 }
 
 impl core::fmt::Display for MarshalError {
@@ -29,6 +37,10 @@ impl core::fmt::Display for MarshalError {
             Self::InvalidUtf8 => f.write_str("invalid utf8"),
             Self::InvalidLocation => f.write_str("invalid source location"),
             Self::BadType => f.write_str("bad type marker"),
+            Self::UnknownType => f.write_str("unknown type code"),
+            Self::InvalidRef => f.write_str("invalid reference"),
+            Self::NullObject => f.write_str("NULL object in marshal data for object"),
+            Self::BadSize(what) => write!(f, "{what} size out of range"),
         }
     }
 }
@@ -111,7 +123,7 @@ impl TryFrom<u8> for Type {
             b'A' => Self::AsciiInterned,
             b'z' => Self::ShortAscii,
             b'Z' => Self::ShortAsciiInterned,
-            _ => return Err(MarshalError::BadType),
+            _ => return Err(MarshalError::UnknownType),
         })
     }
 }
@@ -146,6 +158,13 @@ pub trait Read {
     fn read_u64(&mut self) -> Result<u64> {
         Ok(u64::from_le_bytes(*self.read_array()?))
     }
+
+    /// A length, read the way `r_long` reads one: it is signed, so a value
+    /// with the top bit set is out of range rather than four billion items.
+    fn read_len(&mut self, what: &'static str) -> Result<usize> {
+        let len = self.read_u32()? as i32;
+        usize::try_from(len).map_err(|_| MarshalError::BadSize(what))
+    }
 }
 
 pub(crate) trait ReadBorrowed<'a>: Read {
@@ -305,7 +324,7 @@ fn reserve_ref_slot<T>(has_flag: bool, refs: &mut Vec<Option<T>>) -> Option<usiz
 fn resolve_ref<T: Clone>(idx: usize, refs: &[Option<T>]) -> Result<T> {
     refs.get(idx)
         .and_then(|v| v.clone())
-        .ok_or(MarshalError::InvalidBytecode)
+        .ok_or(MarshalError::InvalidRef)
 }
 
 /// Read a marshal bytes object (TYPE_STRING = b's'), resolving TYPE_REF
@@ -408,7 +427,7 @@ fn read_marshal_str_vec<R: Read, Bag: ConstantBag>(
     }
 
     let n = match type_byte {
-        b'(' => rdr.read_u32()? as usize,
+        b'(' => rdr.read_len("tuple")?,
         b')' => rdr.read_u8()? as usize,
         _ => return Err(MarshalError::BadType),
     };
@@ -471,7 +490,7 @@ fn read_marshal_const_tuple<R: Read, Bag: ConstantBag>(
     }
 
     let n = match type_byte {
-        b'(' => rdr.read_u32()? as usize,
+        b'(' => rdr.read_len("tuple")?,
         b')' => rdr.read_u8()? as usize,
         _ => return Err(MarshalError::BadType),
     };
@@ -553,7 +572,7 @@ pub trait MarshalBag: Copy {
     fn make_code(
         &self,
         code: CodeObject<<Self::ConstantBag as ConstantBag>::Constant>,
-    ) -> Self::Value;
+    ) -> Result<Self::Value>;
 
     /// Construct a runtime code object while retaining the exact values read
     /// from ``co_consts``. Compiler bags ignore this second channel; runtime
@@ -563,7 +582,7 @@ pub trait MarshalBag: Copy {
         &self,
         code: CodeObject<<Self::ConstantBag as ConstantBag>::Constant>,
         _constants: Vec<Self::Value>,
-    ) -> Self::Value {
+    ) -> Result<Self::Value> {
         self.make_code(code)
     }
 
@@ -583,8 +602,12 @@ pub trait MarshalBag: Copy {
     /// Install partially-built containers in the marshal reference table
     /// before reading their children, as CPython's `r_object()` does.
     /// Runtime bags can opt in; constant bags retain collect-then-construct.
-    fn make_tuple_placeholder(&self, _len: usize) -> Option<Self::Value> {
-        None
+    ///
+    /// `len` comes straight from the input and is only bounded by what a
+    /// length can hold, so a bag that opts in reports the room it cannot get
+    /// rather than taking it for granted.
+    fn make_tuple_placeholder(&self, _len: usize) -> Result<Option<Self::Value>> {
+        Ok(None)
     }
 
     fn set_tuple_item(
@@ -596,8 +619,8 @@ pub trait MarshalBag: Copy {
         Err(MarshalError::BadType)
     }
 
-    fn make_list_placeholder(&self, _len: usize) -> Option<Self::Value> {
-        None
+    fn make_list_placeholder(&self, _len: usize) -> Result<Option<Self::Value>> {
+        Ok(None)
     }
 
     fn set_list_item(&self, _list: &Self::Value, _index: usize, _value: Self::Value) -> Result<()> {
@@ -725,8 +748,8 @@ impl<Bag: ConstantBag> MarshalBag for Bag {
     fn make_code(
         &self,
         code: CodeObject<<Self::ConstantBag as ConstantBag>::Constant>,
-    ) -> Self::Value {
-        self.make_code(code)
+    ) -> Result<Self::Value> {
+        Ok(self.make_code(code))
     }
 
     fn make_stop_iter(&self) -> Result<Self::Value> {
@@ -830,10 +853,7 @@ fn deserialize_value_after_header<R: Read, Bag: MarshalBag>(
     // TYPE_REF: return previously stored object
     if type_code == Type::Ref as u8 {
         let idx = rdr.read_u32()? as usize;
-        return refs
-            .get(idx)
-            .and_then(|v| v.clone())
-            .ok_or(MarshalError::InvalidBytecode);
+        return resolve_ref(idx, refs);
     }
 
     // Reserve ref slot before reading (matches write order)
@@ -986,7 +1006,7 @@ fn deserialize_code_value_inner<R: Read, Bag: MarshalBag>(
         linetable,
         exceptiontable,
     };
-    Ok(bag.make_code_with_constants(code, constant_values))
+    bag.make_code_with_constants(code, constant_values)
 }
 
 fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
@@ -1033,13 +1053,13 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
             bag.make_complex(value)
         }
         Type::Ascii | Type::Unicode => {
-            let len = rdr.read_u32()?;
-            let value = rdr.read_wtf8(len)?;
+            let len = rdr.read_len("string")?;
+            let value = rdr.read_wtf8(len as u32)?;
             bag.make_str(value)
         }
         Type::AsciiInterned | Type::Interned => {
-            let len = rdr.read_u32()?;
-            let value = rdr.read_wtf8(len)?;
+            let len = rdr.read_len("string")?;
+            let value = rdr.read_wtf8(len as u32)?;
             bag.make_interned_str(value)
         }
         Type::ShortAscii => {
@@ -1056,7 +1076,7 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
             let len = rdr.read_u8()? as usize;
             let d = depth - 1;
             if let Some(index) = slot
-                && let Some(tuple) = bag.make_tuple_placeholder(len)
+                && let Some(tuple) = bag.make_tuple_placeholder(len)?
             {
                 refs[index] = Some(tuple.clone());
                 for item_index in 0..len {
@@ -1070,17 +1090,17 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
             }
         }
         Type::Null => {
-            return Err(MarshalError::BadType);
+            return Err(MarshalError::NullObject);
         }
         Type::Ref => {
             // Handled in deserialize_value_depth before calling this function
             return Err(MarshalError::BadType);
         }
         Type::Tuple => {
-            let len = rdr.read_u32()? as usize;
+            let len = rdr.read_len("tuple")?;
             let d = depth - 1;
             if let Some(index) = slot
-                && let Some(tuple) = bag.make_tuple_placeholder(len)
+                && let Some(tuple) = bag.make_tuple_placeholder(len)?
             {
                 refs[index] = Some(tuple.clone());
                 for item_index in 0..len {
@@ -1094,10 +1114,10 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
             }
         }
         Type::List => {
-            let len = rdr.read_u32()? as usize;
+            let len = rdr.read_len("list")?;
             let d = depth - 1;
             if let Some(index) = slot
-                && let Some(list) = bag.make_list_placeholder(len)
+                && let Some(list) = bag.make_list_placeholder(len)?
             {
                 refs[index] = Some(list.clone());
                 for item_index in 0..len {
@@ -1111,7 +1131,7 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
             }
         }
         Type::Set => {
-            let len = rdr.read_u32()? as usize;
+            let len = rdr.read_len("set")?;
             let d = depth - 1;
             if let Some(index) = slot
                 && let Some(set) = bag.make_set_placeholder()
@@ -1128,7 +1148,7 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
             }
         }
         Type::FrozenSet => {
-            let len = rdr.read_u32()?;
+            let len = rdr.read_len("set")?;
             let d = depth - 1;
             let it = (0..len).map(|_| deserialize_value_depth(rdr, bag, d, refs));
             itertools::process_results(it, |it| bag.make_frozenset(it))??
@@ -1165,8 +1185,8 @@ fn deserialize_value_typed<R: Read, Bag: MarshalBag>(
         }
         Type::Bytes => {
             // After marshaling, byte arrays are converted into bytes.
-            let len = rdr.read_u32()?;
-            let value = rdr.read_slice(len)?;
+            let len = rdr.read_len("bytes object")?;
+            let value = rdr.read_slice(len as u32)?;
             bag.make_bytes(value)
         }
         Type::Code => return Err(MarshalError::BadType),

crates/host_env/src/io.rs

@@ -199,6 +199,29 @@ pub fn is_seekable(fd: crt_fd::Borrowed<'_>) -> bool {
     os::seek_fd(fd, 0, libc::SEEK_CUR).is_ok()
 }
 
+/// Whether a read from `fd` answers from data the file already holds, rather
+/// than waiting for whoever writes the other end.
+///
+/// Seeking answers this everywhere but Windows, where a pipe seeks too --
+/// `lseek` on one succeeds and reports a position, so a reader that took
+/// seekability for an answer would wait on a peer while holding whatever it
+/// holds for the length of the call.
+#[cfg(not(windows))]
+pub fn reads_without_waiting(fd: crt_fd::Borrowed<'_>) -> bool {
+    is_seekable(fd)
+}
+
+#[cfg(windows)]
+pub fn reads_without_waiting(fd: crt_fd::Borrowed<'_>) -> bool {
+    use std::os::windows::io::AsRawHandle;
+    use windows_sys::Win32::Storage::FileSystem::{FILE_TYPE_DISK, GetFileType};
+
+    let Ok(handle) = crt_fd::as_handle(fd) else {
+        return false;
+    };
+    unsafe { GetFileType(handle.as_raw_handle() as _) == FILE_TYPE_DISK }
+}
+
 pub fn validate_whence(whence: i32) -> bool {
     let standard = (0..=2).contains(&whence);
     #[cfg(any(target_os = "dragonfly", target_os = "freebsd", target_os = "linux"))]

crates/host_env/src/io_unsupported.rs

@@ -176,6 +176,10 @@ pub fn is_seekable(_fd: crt_fd::Borrowed<'_>) -> bool {
     false
 }
 
+pub fn reads_without_waiting(_fd: crt_fd::Borrowed<'_>) -> bool {
+    false
+}
+
 pub fn validate_whence(whence: i32) -> bool {
     (0..=2).contains(&whence)
 }

crates/stdlib/src/_asyncio.rs

@@ -724,47 +724,56 @@ pub(crate) mod _asyncio {
 
         /// Add waiter to fut_awaited_by with single-object optimization
         fn awaited_by_add(&self, waiter: PyObjectRef, vm: &VirtualMachine) -> PyResult<()> {
-            let mut awaited_by = self.fut_awaited_by.write();
-            if awaited_by.is_none() {
-                // First waiter - store directly
-                *awaited_by = Some(waiter);
-                return Ok(());
-            }
+            // Storing a waiter in the set runs its __hash__ and __eq__, which can
+            // come back to this future, so the field is locked only while it is
+            // read or written.
+            let existing = {
+                let mut awaited_by = self.fut_awaited_by.write();
+                match awaited_by.as_ref() {
+                    // First waiter - store directly
+                    None => {
+                        *awaited_by = Some(waiter);
+                        return Ok(());
+                    }
+                    Some(existing) => existing.clone(),
+                }
+            };
 
             if self.fut_awaited_by_is_set.load(Ordering::Relaxed) {
                 // Already a Set - add to it
-                let set = awaited_by.as_ref().unwrap();
-                vm.call_method(set, "add", (waiter,))?;
-            } else {
-                // Single object - convert to Set
-                let existing = awaited_by.take().unwrap();
-                let new_set = PySet::default().into_ref(&vm.ctx);
-                new_set.add(existing, vm)?;
-                new_set.add(waiter, vm)?;
-                *awaited_by = Some(new_set.into());
-                self.fut_awaited_by_is_set.store(true, Ordering::Relaxed);
+                return vm.call_method(&existing, "add", (waiter,)).map(drop);
             }
+
+            // Single object - convert to Set
+            let new_set = PySet::default().into_ref(&vm.ctx);
+            new_set.add(existing, vm)?;
+            new_set.add(waiter, vm)?;
+            *self.fut_awaited_by.write() = Some(new_set.into());
+            self.fut_awaited_by_is_set.store(true, Ordering::Relaxed);
             Ok(())
         }
 
         /// Discard waiter from fut_awaited_by with single-object optimization
         fn awaited_by_discard(&self, waiter: &PyObject, vm: &VirtualMachine) -> PyResult<()> {
-            let mut awaited_by = self.fut_awaited_by.write();
-            if awaited_by.is_none() {
-                return Ok(());
-            }
-
-            let obj = awaited_by.as_ref().unwrap();
-            if !self.fut_awaited_by_is_set.load(Ordering::Relaxed) {
-                // Single object - check if it matches
-                if obj.is(waiter) {
-                    *awaited_by = None;
+            // As in awaited_by_add, discarding from the set runs Python.
+            let set = {
+                let mut awaited_by = self.fut_awaited_by.write();
+                let Some(obj) = awaited_by.as_ref() else {
+                    return Ok(());
+                };
+                if !self.fut_awaited_by_is_set.load(Ordering::Relaxed) {
+                    // Single object - check if it matches
+                    if obj.is(waiter) {
+                        *awaited_by = None;
+                    }
+                    return Ok(());
                 }
-            } else {
-                // It's a Set - use discard
-                vm.call_method(obj, "discard", (waiter.to_owned(),))?;
-            }
-            Ok(())
+                obj.clone()
+            };
+
+            // It's a Set - use discard
+            vm.call_method(&set, "discard", (waiter.to_owned(),))
+                .map(drop)
         }
 
         #[pymethod]

crates/stdlib/src/array.rs

@@ -55,6 +55,11 @@ pub mod array {
                 $($n(Vec<$t>),)*
             }
 
+            /// One item, already converted to the array's element type.
+            enum ArrayItem {
+                $($n($t),)*
+            }
+
             impl ArrayContentType {
                 fn from_char(c: char) -> Result<Self, String> {
                     match c {
@@ -303,17 +308,31 @@ pub mod array {
                     }
                 }
 
-                fn setitem_by_index(
-                    &mut self,
-                    i: isize,
+                /// Convert an object to the element type of the array with
+                /// this typecode. This runs the object's conversion methods,
+                /// which can reach the array, so it takes the typecode by
+                /// value and holds no lock on it.
+                fn item_from_object(
+                    typecode: char,
                     value: PyObjectRef,
                     vm: &VirtualMachine
+                ) -> PyResult<ArrayItem> {
+                    match typecode {
+                        $($c => Ok(ArrayItem::$n(<$t>::try_into_from_object(vm, value)?)),)*
+                        _ => unreachable!("array has a typecode"),
+                    }
+                }
+
+                fn setitem_by_item(
+                    &mut self,
+                    i: isize,
+                    item: ArrayItem,
+                    vm: &VirtualMachine
                 ) -> PyResult<()> {
-                    match self {
-                        $(ArrayContentType::$n(v) => {
-                            let value = <$t>::try_into_from_object(vm, value)?;
-                            v.setitem_by_index(vm, i, value)
-                        })*
+                    match (self, item) {
+                        $((ArrayContentType::$n(v), ArrayItem::$n(value)) =>
+                            v.setitem_by_index(vm, i, value),)*
+                        _ => unreachable!("item was converted for this array"),
                     }
                 }
 
@@ -906,6 +925,11 @@ pub mod array {
 
         #[pymethod]
         fn frombytes(&self, b: ArgBytesLike, vm: &VirtualMachine) -> PyResult<()> {
+            // The source is read as bytes, so items of any other width would
+            // be reinterpreted rather than appended.
+            if b.itemsize() != 1 {
+                return Err(vm.new_type_error("a bytes-like object is required"));
+            }
             let b = b.borrow_buf();
             let itemsize = self.read().itemsize();
             self._from_bytes(&b, itemsize, vm)
@@ -1047,7 +1071,11 @@ pub mod array {
             vm: &VirtualMachine,
         ) -> PyResult<()> {
             match SequenceIndex::try_from_borrowed_object(vm, needle, "array")? {
-                SequenceIndex::Int(i) => zelf.write().setitem_by_index(i, value, vm),
+                SequenceIndex::Int(i) => {
+                    let typecode = zelf.read().typecode();
+                    let item = ArrayContentType::item_from_object(typecode, value, vm)?;
+                    zelf.write().setitem_by_item(i, item, vm)
+                }
                 SequenceIndex::Slice(slice) => {
                     let cloned;
                     let guard;
@@ -1408,7 +1436,9 @@ pub mod array {
                 ass_item: atomic_func!(|seq, i, value, vm| {
                     let zelf = PyArray::sequence_downcast(seq);
                     if let Some(value) = value {
-                        zelf.write().setitem_by_index(i, value, vm)
+                        let typecode = zelf.read().typecode();
+                        let item = ArrayContentType::item_from_object(typecode, value, vm)?;
+                        zelf.write().setitem_by_item(i, item, vm)
                     } else {
                         zelf.write().delitem_by_index(i, vm)
                     }
@@ -1443,8 +1473,15 @@ pub mod array {
         type Resizable<'a> = PyRwLockWriteGuard<'a, ArrayContentType>;
 
         fn try_resizable_opt(&self) -> Option<Self::Resizable<'_>> {
-            let w = self.write();
-            (self.exports.load(atomic::Ordering::SeqCst) == 0).then_some(w)
+            // An export is a borrow someone else still holds, so it is
+            // answered before the lock rather than by waiting on it.
+            (self.exports.load(atomic::Ordering::SeqCst) == 0).then(|| self.write())
+        }
+
+        fn try_resizable(&self, vm: &VirtualMachine) -> PyResult<Self::Resizable<'_>> {
+            self.try_resizable_opt().ok_or_else(|| {
+                vm.new_buffer_error("cannot resize an array that is exporting buffers")
+            })
         }
     }
 

crates/stdlib/src/hashlib.rs

@@ -847,15 +847,15 @@ pub(crate) mod _hashlib {
                 if len < 1 {
                     return Err(vm.new_value_error("key length must be greater than 0."));
                 }
-                usize::try_from(len)
-                    .map_err(|_| vm.new_overflow_error("key length is too great."))?
+                i32::try_from(len).map_err(|_| vm.new_overflow_error("key length is too great."))?
+                    as usize
             }
             None => hash_digest_size(&name).ok_or_else(|| unsupported_hash(&name, vm))?,
         };
 
         let password_buf = args.password.borrow_buf();
         let salt_buf = args.salt.borrow_buf();
-        let mut dk = vec![0u8; dklen];
+        let mut dk = vm.new_zeroed_bytes(dklen)?;
 
         macro_rules! do_pbkdf2 {
             ($hash_ty:ty) => {{

crates/stdlib/src/mmap.rs

@@ -1152,24 +1152,35 @@ mod mmap {
         }
 
         #[pymethod]
-        fn write(&self, bytes: ArgBytesLike, vm: &VirtualMachine) -> PyResult<PyIntRef> {
-            let pos = self.pos();
-            let size = self.__len__();
+        fn write(zelf: &Py<Self>, bytes: ArgBytesLike, vm: &VirtualMachine) -> PyResult<PyIntRef> {
+            let self_ = &**zelf;
+            let pos = self_.pos();
+            let size = self_.__len__();
 
-            let data = bytes.borrow_buf();
+            // Writing locks the map, and reading a source that views this same
+            // map locks it too, so such a source is copied out first.
+            let copied;
+            let borrowed;
+            let data: &[u8] = if bytes.source_object().is(zelf.as_object()) {
+                copied = bytes.borrow_buf().to_vec();
+                &copied
+            } else {
+                borrowed = bytes.borrow_buf();
+                &borrowed
+            };
 
             if pos > size || size - pos < data.len() {
                 return Err(vm.new_value_error("data out of range"));
             }
 
-            let len = self.try_writable(vm, |mmap| {
+            let len = self_.try_writable(vm, |mmap| {
                 (&mut mmap[pos..(pos + data.len())])
-                    .write(&data)
+                    .write(data)
                     .map_err(|err| err.to_pyexception(vm))?;
                 Ok(data.len())
             })??;
 
-            self.advance_pos(len);
+            self_.advance_pos(len);
 
             Ok(PyInt::from(len).into_ref(&vm.ctx))
         }

crates/stdlib/src/pystruct.rs

@@ -10,13 +10,14 @@ pub(crate) use _struct::module_def;
 #[pymodule]
 pub(crate) mod _struct {
     use crate::vm::{
-        AsObject, Py, PyObjectRef, PyPayload, PyResult, TryFromObject, VirtualMachine,
+        AsObject, Py, PyObjectRef, PyPayload, PyRef, PyResult, TryFromObject, VirtualMachine,
         buffer::{FormatSpec, new_struct_error, struct_error_type},
         builtins::{PyBytes, PyStr, PyStrRef, PyTupleRef, PyType, PyTypeRef},
-        function::{ArgBytesLike, ArgMemoryBuffer, PosArgs},
+        common::lock::{PyMappedRwLockReadGuard, PyRwLock, PyRwLockReadGuard},
+        function::{ArgBytesLike, ArgMemoryBuffer, FuncArgs, PosArgs},
         match_class,
         protocol::PyIterReturn,
-        types::{Constructor, IterNext, Iterable, Representable, SelfIter},
+        types::{Constructor, Initializer, IterNext, Iterable, Representable, SelfIter},
     };
     use crossbeam_utils::atomic::AtomicCell;
     use rustpython_common::wtf8::{Wtf8Buf, wtf8_concat};
@@ -251,41 +252,76 @@ pub(crate) mod _struct {
         Ok(fmt.format_spec(vm)?.size)
     }
 
+    /// What a `Struct` is once a format has been read into it. Held apart
+    /// from the object because `__new__` hands out a `Struct` that `__init__`
+    /// has not filled in yet, and `__init__` may be called again on one that
+    /// already holds a format.
+    #[derive(Debug)]
+    struct StructSpec {
+        spec: FormatSpec,
+        format: PyStrRef,
+    }
+
     #[pyattr]
     #[pyclass(name = "Struct", traverse)]
     #[derive(Debug, PyPayload)]
     struct PyStruct {
         #[pytraverse(skip)]
-        spec: FormatSpec,
-        format: PyStrRef,
+        inner: PyRwLock<Option<StructSpec>>,
     }
 
     impl Constructor for PyStruct {
-        type Args = IntoStructFormatBytes;
+        type Args = FuncArgs;
 
-        fn py_new(_cls: &Py<PyType>, fmt: Self::Args, vm: &VirtualMachine) -> PyResult<Self> {
-            let spec = fmt.format_spec(vm)?;
-            let format = fmt.0;
-            Ok(Self { spec, format })
+        fn py_new(_cls: &Py<PyType>, _args: Self::Args, _vm: &VirtualMachine) -> PyResult<Self> {
+            Ok(Self {
+                inner: PyRwLock::new(None),
+            })
         }
     }
 
-    #[pyclass(with(Constructor, Representable))]
+    impl Initializer for PyStruct {
+        type Args = IntoStructFormatBytes;
+
+        fn init(zelf: PyRef<Self>, fmt: Self::Args, vm: &VirtualMachine) -> PyResult<()> {
+            // The format is read before anything is replaced, so a format that
+            // cannot be read leaves the object as it was.
+            let spec = fmt.format_spec(vm)?;
+            *zelf.inner.write() = Some(StructSpec {
+                spec,
+                format: fmt.0,
+            });
+            Ok(())
+        }
+    }
+
+    #[pyclass(with(Constructor, Initializer, Representable), flags(BASETYPE))]
     impl PyStruct {
-        #[pygetset]
-        fn format(&self) -> PyStrRef {
-            self.format.clone()
+        /// The format this was initialized with, or an error if `__init__`
+        /// never ran.
+        fn ready(&self, vm: &VirtualMachine) -> PyResult<PyMappedRwLockReadGuard<'_, StructSpec>> {
+            PyRwLockReadGuard::try_map(self.inner.read(), Option::as_ref)
+                .map_err(|_| vm.new_runtime_error("Struct object is not initialized"))
         }
 
         #[pygetset]
-        #[inline]
-        const fn size(&self) -> usize {
-            self.spec.size
+        fn format(&self, vm: &VirtualMachine) -> PyResult<PyStrRef> {
+            Ok(self.ready(vm)?.format.clone())
+        }
+
+        /// The size an uninitialized `Struct` reports, which no format has
+        /// yet given a value.
+        #[pygetset]
+        fn size(&self) -> isize {
+            self.inner
+                .read()
+                .as_ref()
+                .map_or(-1, |inner| inner.spec.size as isize)
         }
 
         #[pymethod]
         fn pack(&self, args: PosArgs, vm: &VirtualMachine) -> PyResult<Vec<u8>> {
-            self.spec.pack(args.into_vec(), vm)
+            self.ready(vm)?.spec.pack(args.into_vec(), vm)
         }
 
         #[pymethod]
@@ -296,23 +332,28 @@ pub(crate) mod _struct {
             args: PosArgs,
             vm: &VirtualMachine,
         ) -> PyResult<()> {
-            let offset = get_buffer_offset(buffer.len(), offset, self.size(), true, vm)?;
+            let inner = self.ready(vm)?;
+            let offset = get_buffer_offset(buffer.len(), offset, inner.spec.size, true, vm)?;
             buffer.with_ref(|data| {
-                self.spec
+                inner
+                    .spec
                     .pack_into(&mut data[offset..], args.into_vec(), vm)
             })
         }
 
         #[pymethod]
         fn unpack(&self, data: ArgBytesLike, vm: &VirtualMachine) -> PyResult<PyTupleRef> {
-            data.with_ref(|buf| self.spec.unpack(buf, vm))
+            let inner = self.ready(vm)?;
+            data.with_ref(|buf| inner.spec.unpack(buf, vm))
         }
 
         #[pymethod]
         fn unpack_from(&self, args: UpdateFromArgs, vm: &VirtualMachine) -> PyResult<PyTupleRef> {
-            let offset = get_buffer_offset(args.buffer.len(), args.offset, self.size(), false, vm)?;
+            let inner = self.ready(vm)?;
+            let size = inner.spec.size;
+            let offset = get_buffer_offset(args.buffer.len(), args.offset, size, false, vm)?;
             args.buffer
-                .with_ref(|buf| self.spec.unpack(&buf[offset..][..self.size()], vm))
+                .with_ref(|buf| inner.spec.unpack(&buf[offset..][..size], vm))
         }
 
         #[pymethod]
@@ -321,14 +362,19 @@ pub(crate) mod _struct {
             buffer: ArgBytesLike,
             vm: &VirtualMachine,
         ) -> PyResult<UnpackIterator> {
-            UnpackIterator::with_buffer(vm, self.spec.clone(), buffer)
+            let spec = self.ready(vm)?.spec.clone();
+            UnpackIterator::with_buffer(vm, spec, buffer)
         }
     }
 
     impl Representable for PyStruct {
         #[inline]
-        fn repr_wtf8(zelf: &Py<Self>, _vm: &VirtualMachine) -> PyResult<Wtf8Buf> {
-            Ok(wtf8_concat!("Struct('", zelf.format.as_wtf8(), "')"))
+        fn repr_wtf8(zelf: &Py<Self>, vm: &VirtualMachine) -> PyResult<Wtf8Buf> {
+            Ok(wtf8_concat!(
+                "Struct('",
+                zelf.ready(vm)?.format.as_wtf8(),
+                "')"
+            ))
         }
     }
 

crates/stdlib/src/select.rs

@@ -79,16 +79,26 @@ mod decl {
         }
         let deadline = timeout.map(|s| time::time(vm).unwrap() + s);
 
-        let seq2set = |list: &PyObject| -> PyResult<(Vec<Selectable>, FdSet)> {
-            let v: Vec<Selectable> = list.try_to_value(vm)?;
+        let max_fds: usize = cfg_select! {
+            windows => FD_SETSIZE as usize,
+            _ => FD_SETSIZE,
+        };
 
-            let too_many_fds = cfg_select! {
-                windows => v.len() > FD_SETSIZE as usize,
-                _ => v.len() > FD_SETSIZE,
-            };
-            if too_many_fds {
-                return Err(vm.new_value_error("too many file descriptors in select()"));
-            }
+        let seq2set = |list: &PyObject| -> PyResult<(Vec<Selectable>, FdSet)> {
+            // The limit is answered while the sequence is walked rather than
+            // from the length of the result. fileno() runs Python and can
+            // append to the very list being walked, and a walk that re-reads
+            // the list each step -- which is what `seq2set` does -- then never
+            // reaches a length to check.
+            let seen = core::cell::Cell::new(0usize);
+            let v: Vec<Selectable> = vm.extract_elements_with(list, |obj| {
+                let selectable = Selectable::try_from_object(vm, obj)?;
+                seen.set(seen.get() + 1);
+                if seen.get() > max_fds {
+                    return Err(vm.new_value_error("too many file descriptors in select()"));
+                }
+                Ok(selectable)
+            })?;
 
             let mut fds = FdSet::new();
             for fd in &v {
@@ -304,7 +314,10 @@ mod decl {
                 timeout: OptionalArg<TimeoutArg<true>>,
                 vm: &VirtualMachine,
             ) -> PyResult<Vec<PyObjectRef>> {
-                let mut fds = self.fds.lock();
+                // Poll a copy: the wait releases the GIL-equivalent and runs
+                // signal handlers, which can register or unregister on the same
+                // object, and a held lock would deadlock them.
+                let mut fds = self.fds.lock().clone();
                 let TimeoutArg(timeout) = timeout.unwrap_or_default();
                 let timeout_ms = match timeout {
                     Some(d) => i32::try_from(d.as_millis())

crates/stdlib/src/socket.rs

@@ -40,7 +40,6 @@ mod _socket {
     }
 
     use core::{
-        mem::MaybeUninit,
         net::{Ipv4Addr, Ipv6Addr, SocketAddr},
         time::Duration,
     };
@@ -1589,7 +1588,10 @@ mod _socket {
             vm: &VirtualMachine,
         ) -> Result<Vec<u8>, IoOrPyException> {
             let flags = flags.unwrap_or(0);
-            let mut buffer = Vec::with_capacity(bufsize);
+            let mut buffer = Vec::new();
+            buffer
+                .try_reserve_exact(bufsize)
+                .map_err(|_| vm.new_memory_error(""))?;
             let sock = self.sock()?;
             let n = self.sock_op(vm, SockWaitKind::Read, || {
                 sock.recv_with_flags(buffer.spare_capacity_mut(), flags)
@@ -1608,8 +1610,6 @@ mod _socket {
         ) -> Result<usize, IoOrPyException> {
             let flags = flags.unwrap_or(0);
             let sock = self.sock()?;
-            let mut buf = buf.borrow_buf_mut();
-            let buf = &mut *buf;
 
             // Handle nbytes parameter
             let read_len = if let OptionalArg::Present(nbytes) = nbytes {
@@ -1621,10 +1621,13 @@ mod _socket {
                 buf.len()
             };
 
-            let buf = &mut buf[..read_len];
-            self.sock_op(vm, SockWaitKind::Read, || {
-                sock.recv_with_flags(unsafe { slice_as_uninit(buf) }, flags)
-            })
+            let mut scratch = alloc_recv_scratch(read_len, vm)?;
+            let n = self.sock_op(vm, SockWaitKind::Read, || {
+                sock.recv_with_flags(&mut scratch.spare_capacity_mut()[..read_len], flags)
+            })?;
+            unsafe { scratch.set_len(n) };
+            buf.borrow_buf_mut()[..n].copy_from_slice(&scratch);
+            Ok(n)
         }
 
         #[pymethod]
@@ -1638,7 +1641,10 @@ mod _socket {
             let bufsize = bufsize
                 .to_usize()
                 .ok_or_else(|| vm.new_value_error("negative buffersize in recvfrom"))?;
-            let mut buffer = Vec::with_capacity(bufsize);
+            let mut buffer = Vec::new();
+            buffer
+                .try_reserve_exact(bufsize)
+                .map_err(|_| vm.new_memory_error(""))?;
             let (n, addr) = self.sock_op(vm, SockWaitKind::Read, || {
                 self.sock()?
                     .recv_from_with_flags(buffer.spare_capacity_mut(), flags)
@@ -1655,24 +1661,28 @@ mod _socket {
             flags: OptionalArg<i32>,
             vm: &VirtualMachine,
         ) -> Result<(usize, PyObjectRef), IoOrPyException> {
-            let mut buf = buf.borrow_buf_mut();
-            let buf = &mut *buf;
-            let buf = match nbytes {
+            let read_len = match nbytes {
                 OptionalArg::Present(i) => {
                     let i = i.to_usize().ok_or_else(|| {
                         vm.new_value_error("negative buffersize in recvfrom_into")
                     })?;
-                    buf.get_mut(..i).ok_or_else(|| {
-                        vm.new_value_error("nbytes is greater than the length of the buffer")
-                    })?
+                    if i > buf.len() {
+                        return Err(vm
+                            .new_value_error("nbytes is greater than the length of the buffer")
+                            .into());
+                    }
+                    i
                 }
-                OptionalArg::Missing => buf,
+                OptionalArg::Missing => buf.len(),
             };
             let flags = flags.unwrap_or(0);
             let sock = self.sock()?;
+            let mut scratch = alloc_recv_scratch(read_len, vm)?;
             let (n, addr) = self.sock_op(vm, SockWaitKind::Read, || {
-                sock.recv_from_with_flags(unsafe { slice_as_uninit(buf) }, flags)
+                sock.recv_from_with_flags(&mut scratch.spare_capacity_mut()[..read_len], flags)
             })?;
+            unsafe { scratch.set_len(n) };
+            buf.borrow_buf_mut()[..n].copy_from_slice(&scratch);
             Ok((n, get_addr_tuple(&addr, vm)))
         }
 
@@ -1684,7 +1694,7 @@ mod _socket {
             vm: &VirtualMachine,
         ) -> Result<usize, IoOrPyException> {
             let flags = flags.unwrap_or(0);
-            let buf = bytes.borrow_buf();
+            let buf = bytes.borrow_buf_unlocked(vm)?;
             let buf = &*buf;
             self.sock_op(vm, SockWaitKind::Write, || {
                 self.sock()?.send_with_flags(buf, flags)
@@ -1704,7 +1714,7 @@ mod _socket {
 
             let deadline = timeout.map(Deadline::new);
 
-            let buf = bytes.borrow_buf();
+            let buf = bytes.borrow_buf_unlocked(vm)?;
             let buf = &*buf;
             let mut buf_offset = 0;
             // now we have like 3 layers of interrupt loop :)
@@ -1741,7 +1751,7 @@ mod _socket {
                 OptionalArg::Missing => (0, arg2),
             };
             let addr = self.extract_address(address, "sendto", vm)?;
-            let buf = bytes.borrow_buf();
+            let buf = bytes.borrow_buf_unlocked(vm)?;
             let buf = &*buf;
             self.sock_op(vm, SockWaitKind::Write, || {
                 self.sock()?.send_to_with_flags(buf, &addr, flags)
@@ -1771,8 +1781,8 @@ mod _socket {
 
             let buffers = buffers
                 .iter()
-                .map(|buf| buf.borrow_buf())
-                .collect::<Vec<_>>();
+                .map(|buf| buf.borrow_buf_unlocked(vm))
+                .collect::<PyResult<Vec<_>>>()?;
             let buffers = buffers
                 .iter()
                 .map(|buf| io::IoSlice::new(buf))
@@ -2380,8 +2390,21 @@ mod _socket {
         Ok(s.to_string_lossy().into_owned())
     }
 
-    unsafe fn slice_as_uninit<T>(v: &mut [T]) -> &mut [MaybeUninit<T>] {
-        unsafe { &mut *(v as *mut [T] as *mut [MaybeUninit<T>]) }
+    /// Room to receive into that belongs to no Python object.
+    ///
+    /// A peer may never send, so the wait for it is unbounded. The export of
+    /// the caller's buffer is held for the whole call, which is what keeps it
+    /// from being resized, but the borrow that reaches its bytes is a lock
+    /// every other thread touching that object waits on, and a thread waiting
+    /// on a lock never reaches a safepoint — holding it across the wait stops
+    /// the world from being stopped at all. The bytes are copied over once
+    /// they have arrived.
+    fn alloc_recv_scratch(len: usize, vm: &VirtualMachine) -> PyResult<Vec<u8>> {
+        let mut scratch = Vec::new();
+        scratch
+            .try_reserve_exact(len)
+            .map_err(|_| vm.new_memory_error(""))?;
+        Ok(scratch)
     }
 
     enum IoOrPyException {

crates/vm/src/anystr.rs

@@ -27,7 +27,7 @@ pub struct SplitLinesArgs {
 #[derive(FromArgs)]
 pub struct ExpandTabsArgs {
     #[pyarg(any, default = 8)]
-    tabsize: isize,
+    tabsize: i32,
 }
 
 impl ExpandTabsArgs {
@@ -132,6 +132,11 @@ where
 {
     fn new() -> Self;
     fn with_capacity(capacity: usize) -> Self;
+    /// `with_capacity`, reporting a capacity that cannot be allocated instead
+    /// of aborting the process on it.
+    fn try_with_capacity(capacity: usize) -> Option<Self>
+    where
+        Self: Sized;
     fn push_str(&mut self, s: &S);
 }
 
@@ -285,27 +290,29 @@ pub(crate) trait AnyStr {
         }
     }
 
-    fn py_pad(&self, left: usize, right: usize, fillchar: Self::Char) -> Self::Container {
-        let mut u = Self::Container::with_capacity(
-            (left + right) * fillchar.bytes_len() + self.bytes_len(),
-        );
+    fn py_pad(&self, left: usize, right: usize, fillchar: Self::Char) -> Option<Self::Container> {
+        let capacity = left
+            .checked_add(right)?
+            .checked_mul(fillchar.bytes_len())?
+            .checked_add(self.bytes_len())?;
+        let mut u = Self::Container::try_with_capacity(capacity)?;
         u.extend(core::iter::repeat_n(fillchar, left));
         u.push_str(self);
         u.extend(core::iter::repeat_n(fillchar, right));
-        u
+        Some(u)
     }
 
-    fn py_center(&self, width: usize, fillchar: Self::Char, len: usize) -> Self::Container {
+    fn py_center(&self, width: usize, fillchar: Self::Char, len: usize) -> Option<Self::Container> {
         let marg = width - len;
         let left = marg / 2 + (marg & width & 1);
         self.py_pad(left, marg - left, fillchar)
     }
 
-    fn py_ljust(&self, width: usize, fillchar: Self::Char, len: usize) -> Self::Container {
+    fn py_ljust(&self, width: usize, fillchar: Self::Char, len: usize) -> Option<Self::Container> {
         self.py_pad(0, width - len, fillchar)
     }
 
-    fn py_rjust(&self, width: usize, fillchar: Self::Char, len: usize) -> Self::Container {
+    fn py_rjust(&self, width: usize, fillchar: Self::Char, len: usize) -> Option<Self::Container> {
         self.py_pad(width - len, 0, fillchar)
     }
 
@@ -402,7 +409,7 @@ pub(crate) trait AnyStr {
         elements
     }
 
-    fn py_zfill(&self, width: isize) -> Vec<u8> {
+    fn py_zfill(&self, width: isize) -> Option<Vec<u8>> {
         let width = width.to_usize().unwrap_or(0);
         let char_len = self.elements().count();
         let width = self

crates/vm/src/buffer.rs

@@ -16,7 +16,7 @@ use malachite_bigint::BigInt;
 use num_traits::{PrimInt, ToPrimitive};
 use std::os::raw;
 
-type PackFunc = fn(&VirtualMachine, PyObjectRef, &mut [u8]) -> PyResult<()>;
+type PackFunc = fn(&VirtualMachine, FormatType, PyObjectRef, &mut [u8]) -> PyResult<()>;
 type UnpackFunc = fn(&VirtualMachine, &[u8]) -> PyObjectRef;
 
 static OVERFLOW_MSG: &str = "total struct size too long"; // not a const to reduce code size
@@ -490,7 +490,7 @@ impl FormatSpec {
                     let pack = code.info.pack.unwrap();
                     for arg in args.by_ref().take(code.repeat) {
                         let (item_buf, rest) = buffer.split_at_mut(code.info.size);
-                        pack(vm, arg, item_buf)?;
+                        pack(vm, code.code, arg, item_buf)?;
                         buffer = rest;
                     }
                 }
@@ -549,7 +549,12 @@ impl FormatSpec {
 }
 
 trait Packable {
-    fn pack<E: ByteOrder>(vm: &VirtualMachine, arg: PyObjectRef, data: &mut [u8]) -> PyResult<()>;
+    fn pack<E: ByteOrder>(
+        vm: &VirtualMachine,
+        code: FormatType,
+        arg: PyObjectRef,
+        data: &mut [u8],
+    ) -> PyResult<()>;
     fn unpack<E: ByteOrder>(vm: &VirtualMachine, data: &[u8]) -> PyObjectRef;
 }
 
@@ -576,10 +581,11 @@ macro_rules! make_pack_prim_int {
         impl Packable for $T {
             fn pack<E: ByteOrder>(
                 vm: &VirtualMachine,
+                code: FormatType,
                 arg: PyObjectRef,
                 data: &mut [u8],
             ) -> PyResult<()> {
-                let i: $T = get_int_or_index(vm, arg)?;
+                let i: $T = get_int_or_index(vm, code, arg)?;
                 i.pack_int::<E>(data);
                 Ok(())
             }
@@ -592,16 +598,28 @@ macro_rules! make_pack_prim_int {
     };
 }
 
-fn get_int_or_index<T>(vm: &VirtualMachine, arg: PyObjectRef) -> PyResult<T>
+fn get_int_or_index<T>(vm: &VirtualMachine, code: FormatType, arg: PyObjectRef) -> PyResult<T>
 where
-    T: PrimInt + for<'a> TryFrom<&'a BigInt>,
+    T: PrimInt + fmt::Display + for<'a> TryFrom<&'a BigInt>,
 {
     let index = arg
         .try_index_opt(vm)
         .unwrap_or_else(|| Err(new_struct_error(vm, "required argument is not an integer")))?;
-    index
-        .try_to_primitive(vm)
-        .map_err(|_| new_struct_error(vm, "argument out of range"))
+    index.try_to_primitive(vm).map_err(|_| {
+        // A pointer is converted rather than checked against the range of a
+        // named format, so what it reports is the conversion failing.
+        let msg = if code == FormatType::VoidP {
+            "int too large to convert".to_owned()
+        } else {
+            format!(
+                "'{}' format requires {} <= number <= {}",
+                code as u8 as char,
+                T::min_value(),
+                T::max_value()
+            )
+        };
+        new_struct_error(vm, msg)
+    })
 }
 
 make_pack_prim_int!(i8);
@@ -620,6 +638,7 @@ macro_rules! make_pack_float {
         impl Packable for $T {
             fn pack<E: ByteOrder>(
                 vm: &VirtualMachine,
+                _code: FormatType,
                 arg: PyObjectRef,
                 data: &mut [u8],
             ) -> PyResult<()> {
@@ -648,7 +667,12 @@ make_pack_float!(f32, "f");
 make_pack_float!(f64, "d");
 
 impl Packable for f16 {
-    fn pack<E: ByteOrder>(vm: &VirtualMachine, arg: PyObjectRef, data: &mut [u8]) -> PyResult<()> {
+    fn pack<E: ByteOrder>(
+        vm: &VirtualMachine,
+        _code: FormatType,
+        arg: PyObjectRef,
+        data: &mut [u8],
+    ) -> PyResult<()> {
         let f_64 = ArgIntoFloat::try_from_object(vm, arg)?.into_float();
         // "from_f64 should be preferred in any non-`const` context" except it gives the wrong result :/
         let f_16 = Self::from_f64_const(f_64);
@@ -666,8 +690,13 @@ impl Packable for f16 {
 }
 
 impl Packable for *mut raw::c_void {
-    fn pack<E: ByteOrder>(vm: &VirtualMachine, arg: PyObjectRef, data: &mut [u8]) -> PyResult<()> {
-        usize::pack::<E>(vm, arg, data)
+    fn pack<E: ByteOrder>(
+        vm: &VirtualMachine,
+        code: FormatType,
+        arg: PyObjectRef,
+        data: &mut [u8],
+    ) -> PyResult<()> {
+        usize::pack::<E>(vm, code, arg, data)
     }
 
     fn unpack<E: ByteOrder>(vm: &VirtualMachine, rdr: &[u8]) -> PyObjectRef {
@@ -676,7 +705,12 @@ impl Packable for *mut raw::c_void {
 }
 
 impl Packable for bool {
-    fn pack<E: ByteOrder>(vm: &VirtualMachine, arg: PyObjectRef, data: &mut [u8]) -> PyResult<()> {
+    fn pack<E: ByteOrder>(
+        vm: &VirtualMachine,
+        _code: FormatType,
+        arg: PyObjectRef,
+        data: &mut [u8],
+    ) -> PyResult<()> {
         let v = ArgIntoBool::try_from_object(vm, arg)?.into_bool() as u8;
         v.pack_int::<E>(data);
         Ok(())
@@ -688,7 +722,12 @@ impl Packable for bool {
     }
 }
 
-fn pack_char(vm: &VirtualMachine, arg: PyObjectRef, data: &mut [u8]) -> PyResult<()> {
+fn pack_char(
+    vm: &VirtualMachine,
+    _code: FormatType,
+    arg: PyObjectRef,
+    data: &mut [u8],
+) -> PyResult<()> {
     let v = PyBytesRef::try_from_object(vm, arg)?;
     let ch = *v
         .as_bytes()

crates/vm/src/builtins/bytearray.rs

@@ -354,7 +354,10 @@ impl PyByteArray {
 
     #[pymethod]
     fn join(&self, iter: ArgIterable<PyBytesInner>, vm: &VirtualMachine) -> PyResult<Self> {
-        Ok(self.inner().join(iter, vm)?.into())
+        // Driving the iterable runs Python, which can reach this bytearray,
+        // so the separator is taken by value rather than left borrowed.
+        let separator = self.inner().clone();
+        Ok(separator.join(iter, vm)?.into())
     }
 
     #[pymethod]
@@ -497,8 +500,8 @@ impl PyByteArray {
     }
 
     #[pymethod]
-    fn zfill(&self, width: isize) -> Self {
-        self.inner().zfill(width).into()
+    fn zfill(&self, width: isize, vm: &VirtualMachine) -> PyResult<Self> {
+        Ok(self.inner().zfill(width, vm)?.into())
     }
 
     #[pymethod]
@@ -532,7 +535,10 @@ impl PyByteArray {
     }
 
     fn __mod__(&self, values: PyObjectRef, vm: &VirtualMachine) -> PyResult<Self> {
-        let formatted = self.inner().cformat(values, vm)?;
+        // Formatting calls the values' conversion methods, which can reach
+        // this bytearray, so the format is taken by value.
+        let format = self.inner().clone();
+        let formatted = format.cformat(values, vm)?;
         Ok(formatted.into())
     }
 
@@ -778,8 +784,9 @@ impl BufferResizeGuard for PyByteArray {
     type Resizable<'a> = PyRwLockWriteGuard<'a, PyBytesInner>;
 
     fn try_resizable_opt(&self) -> Option<Self::Resizable<'_>> {
-        let w = self.inner.write();
-        (self.exports.load(Ordering::SeqCst) == 0).then_some(w)
+        // An export is a borrow someone else still holds, so it is answered
+        // before the lock rather than by waiting on it.
+        (self.exports.load(Ordering::SeqCst) == 0).then(|| self.inner.write())
     }
 }
 

crates/vm/src/builtins/bytes.rs

@@ -505,8 +505,8 @@ impl PyBytes {
     }
 
     #[pymethod]
-    fn zfill(&self, width: isize) -> Self {
-        self.inner.zfill(width).into()
+    fn zfill(&self, width: isize, vm: &VirtualMachine) -> PyResult<Self> {
+        Ok(self.inner.zfill(width, vm)?.into())
     }
 
     #[pymethod]

crates/vm/src/builtins/memory.rs

@@ -1,6 +1,6 @@
 use super::{
     PositionIterInternal, PyBytes, PyBytesRef, PyGenericAlias, PyInt, PyListRef, PySlice, PyStr,
-    PyTuple, PyTupleRef, PyType, PyTypeRef, PyUtf8StrRef, iter::builtins_iter,
+    PyStrRef, PyTuple, PyTupleRef, PyType, PyTypeRef, PyUtf8StrRef, iter::builtins_iter,
 };
 use crate::common::lock::LazyLock;
 use crate::{
@@ -192,6 +192,11 @@ impl PyMemoryView {
         }
     }
 
+    /// The object this view looks at, whose storage it borrows.
+    pub fn viewed_object(&self) -> &PyObject {
+        &self.buffer.obj
+    }
+
     fn try_not_released(&self, vm: &VirtualMachine) -> PyResult<()> {
         if self.released.load() {
             Err(vm.new_value_error("operation forbidden on released memoryview object"))
@@ -826,10 +831,32 @@ impl PyMemoryView {
     }
 
     #[pymethod]
-    fn tobytes(&self, vm: &VirtualMachine) -> PyResult<PyBytesRef> {
+    fn tobytes(&self, args: ToBytesArgs, vm: &VirtualMachine) -> PyResult<PyBytesRef> {
         self.try_not_released(vm)?;
+        let order = match &args.order {
+            None => Order::C,
+            Some(order) => match order.to_str() {
+                Some("C") => Order::C,
+                Some("F") => Order::Fortran,
+                Some("A") => Order::Any,
+                _ => return Err(vm.new_value_error("order must be 'C', 'F' or 'A'")),
+            },
+        };
+
         let mut v = vec![];
-        self.append_to(&mut v);
+        // 'A' asks for the memory as it is laid out, which is what appending a
+        // contiguous view does. Only a Fortran walk of a view that is not
+        // already Fortran-contiguous reorders anything, and a view of fewer
+        // than two dimensions has one layout under either name.
+        if order == Order::Fortran && self.desc.ndim() > 1 {
+            v.reserve(self.desc.len);
+            let bytes = &*self.buffer.obj_bytes();
+            self.desc.for_each_segment_fortran(|range| {
+                v.extend_from_slice(&bytes[range.start as usize..range.end as usize]);
+            });
+        } else {
+            self.append_to(&mut v);
+        }
         Ok(PyBytes::from(v).into_ref(&vm.ctx))
     }
 
@@ -925,10 +952,17 @@ impl PyMemoryView {
 
     fn cast_to_1d(&self, format: PyUtf8StrRef, vm: &VirtualMachine) -> PyResult<Self> {
         let format_str = format.as_str();
-        if Self::native_fmtchar(format_str).is_none() {
+        let Some(dest_char) = Self::native_fmtchar(format_str) else {
             return Err(vm.new_value_error(
                 "memoryview: destination format must be a native single character format prefixed with an optional '@'",
             ));
+        };
+        // One side has to be bytes. Casting between two item types would
+        // reinterpret the items rather than re-divide the memory, and the
+        // source items were written by something that chose their type.
+        let source_is_bytes = Self::native_fmtchar(&self.desc.format).is_some_and(is_byte_fmtchar);
+        if !source_is_bytes && !is_byte_fmtchar(dest_char) {
+            return Err(vm.new_type_error("memoryview: cannot cast between two non-byte formats"));
         }
         let format_spec = Self::parse_format(format_str, vm)?;
         let itemsize = format_spec.size();
@@ -994,7 +1028,7 @@ impl PyMemoryView {
             let mut other = self.cast_to_1d(format, vm)?;
             let itemsize = other.desc.itemsize;
 
-            // 0 ndim is single item
+            // 0 ndim is single item, so the buffer has to be that one item
             if shape_ndim == 0 {
                 if itemsize != other.desc.len {
                     return Err(
@@ -1002,7 +1036,6 @@ impl PyMemoryView {
                     );
                 }
                 other.desc.dim_desc = vec![];
-                other.desc.len = itemsize;
                 return Ok(other.into_ref(&vm.ctx));
             }
 
@@ -1010,7 +1043,19 @@ impl PyMemoryView {
             let mut dim_descriptor = Vec::with_capacity(shape_ndim);
 
             for x in shape {
-                let x = usize::try_from_borrowed_object(vm, x)?;
+                let x = x
+                    .downcast_ref::<PyInt>()
+                    .ok_or_else(|| {
+                        vm.new_type_error("memoryview.cast(): elements of shape must be integers")
+                    })?
+                    .try_to_primitive::<usize>(vm)
+                    .ok()
+                    .filter(|x| *x > 0)
+                    .ok_or_else(|| {
+                        vm.new_value_error(
+                            "memoryview.cast(): elements of shape must be integers > 0",
+                        )
+                    })?;
 
                 if x > isize::MAX as usize / product_shape {
                     return Err(vm.new_value_error("memoryview.cast(): product(shape) > SSIZE_MAX"));
@@ -1084,6 +1129,20 @@ impl Py<PyMemoryView> {
     }
 }
 
+#[derive(FromArgs)]
+struct ToBytesArgs {
+    #[pyarg(any, default)]
+    order: Option<PyStrRef>,
+}
+
+/// The layout a copy of a view is written in.
+#[derive(PartialEq, Eq)]
+enum Order {
+    C,
+    Fortran,
+    Any,
+}
+
 #[derive(FromArgs)]
 struct CastArgs {
     #[pyarg(any)]
@@ -1242,7 +1301,9 @@ impl Hashable for PyMemoryView {
         if !zelf.desc.readonly {
             return Err(vm.new_value_error("cannot hash writable memoryview object"));
         }
-        if !matches!(&*zelf.desc.format, "B" | "b" | "c") {
+        // The hash is over the bytes, so it agrees with the hash of the same
+        // bytes only where an item is a byte.
+        if !Self::native_fmtchar(&zelf.desc.format).is_some_and(is_byte_fmtchar) {
             return Err(
                 vm.new_value_error("memoryview: hashing is restricted to formats 'B', 'b' or 'c'")
             );
@@ -1496,6 +1557,10 @@ fn format_unpack(
     })
 }
 
+/// Whether `ch` names a format whose items are single bytes.
+const fn is_byte_fmtchar(ch: u8) -> bool {
+    matches!(ch, b'c' | b'b' | b'B')
+}
 fn is_equiv_shape(a: &BufferDescriptor, b: &BufferDescriptor) -> bool {
     if a.ndim() != b.ndim() {
         return false;

crates/vm/src/builtins/str.rs

@@ -1320,11 +1320,13 @@ impl PyStr {
     }
 
     #[pymethod]
-    fn zfill(&self, width: isize) -> Wtf8Buf {
-        unsafe {
-            // SAFETY: this is safe-guaranteed because the original self.as_wtf8() is valid wtf8
-            Wtf8Buf::from_bytes_unchecked(self.as_wtf8().py_zfill(width))
-        }
+    fn zfill(&self, width: isize, vm: &VirtualMachine) -> PyResult<Wtf8Buf> {
+        let filled = self
+            .as_wtf8()
+            .py_zfill(width)
+            .ok_or_else(|| vm.new_memory_error(""))?;
+        // SAFETY: this is safe-guaranteed because the original self.as_wtf8() is valid wtf8
+        Ok(unsafe { Wtf8Buf::from_bytes_unchecked(filled) })
     }
 
     #[inline]
@@ -1332,7 +1334,7 @@ impl PyStr {
         &self,
         width: isize,
         fillchar: OptionalArg<PyStrRef>,
-        pad: fn(&Wtf8, usize, CodePoint, usize) -> Wtf8Buf,
+        pad: fn(&Wtf8, usize, CodePoint, usize) -> Option<Wtf8Buf>,
         vm: &VirtualMachine,
     ) -> PyResult<Wtf8Buf> {
         let fillchar = fillchar.map_or(Ok(' '.into()), |ref s| {
@@ -1340,11 +1342,11 @@ impl PyStr {
                 vm.new_type_error("The fill character must be exactly one character long")
             })
         })?;
-        Ok(if self.len() as isize >= width {
-            self.as_wtf8().to_owned()
-        } else {
-            pad(self.as_wtf8(), width as usize, fillchar, self.len())
-        })
+        if self.len() as isize >= width {
+            return Ok(self.as_wtf8().to_owned());
+        }
+        pad(self.as_wtf8(), width as usize, fillchar, self.len())
+            .ok_or_else(|| vm.new_memory_error(""))
     }
 
     #[pymethod]
@@ -2214,6 +2216,12 @@ impl AnyStrContainer<str> for String {
         Self::with_capacity(capacity)
     }
 
+    fn try_with_capacity(capacity: usize) -> Option<Self> {
+        let mut s = Self::new();
+        s.try_reserve_exact(capacity).ok()?;
+        Some(s)
+    }
+
     fn push_str(&mut self, other: &str) {
         Self::push_str(self, other)
     }
@@ -2327,6 +2335,12 @@ impl AnyStrContainer<Wtf8> for Wtf8Buf {
         Self::with_capacity(capacity)
     }
 
+    fn try_with_capacity(capacity: usize) -> Option<Self> {
+        let mut s = Self::new();
+        s.try_reserve_exact(capacity).ok()?;
+        Some(s)
+    }
+
     fn push_str(&mut self, other: &Wtf8) {
         self.push_wtf8(other)
     }
@@ -2447,6 +2461,12 @@ impl AnyStrContainer<AsciiStr> for AsciiString {
         Self::with_capacity(capacity)
     }
 
+    fn try_with_capacity(capacity: usize) -> Option<Self> {
+        let mut v = Vec::new();
+        v.try_reserve_exact(capacity).ok()?;
+        Some(Self::from(v))
+    }
+
     fn push_str(&mut self, other: &AsciiStr) {
         Self::push_str(self, other)
     }

crates/vm/src/builtins/tuple.rs

@@ -342,10 +342,6 @@ impl PyTuple<PyObjectRef> {
         }
     }
 
-    pub(crate) fn new_marshal_placeholder(len: usize, ctx: &Context) -> PyRef<Self> {
-        Self::new_ref(vec![ctx.none(); len], ctx)
-    }
-
     /// # Safety
     /// This tuple must be a marshal placeholder which has not escaped the
     /// decoder, and `index` must not have been replaced previously.

crates/vm/src/bytes_inner.rs

@@ -78,7 +78,7 @@ impl ByteInnerNewOptions {
         } else {
             size as usize
         };
-        Ok(vec![0; size].into())
+        Ok(vm.new_zeroed_bytes(size)?.into())
     }
 
     fn handle_object_fallback(obj: PyObjectRef, vm: &VirtualMachine) -> PyResult<PyBytesInner> {
@@ -576,16 +576,15 @@ impl PyBytesInner {
     fn _pad(
         &self,
         options: ByteInnerPaddingOptions,
-        pad: fn(&[u8], usize, u8, usize) -> Vec<u8>,
+        pad: PadFn,
         vm: &VirtualMachine,
     ) -> PyResult<Vec<u8>> {
         let (width, fillchar) = options.get_value("center", vm)?;
         let len = self.len();
-        Ok(if len as isize >= width {
-            Vec::from(&self.elements[..])
-        } else {
-            pad(&self.elements, width as usize, fillchar, len)
-        })
+        if len as isize >= width {
+            return Ok(Vec::from(&self.elements[..]));
+        }
+        pad(&self.elements, width as usize, fillchar, len).ok_or_else(|| vm.new_memory_error(""))
     }
 
     pub fn center(
@@ -821,8 +820,10 @@ impl PyBytesInner {
         self.elements.py_bytes_splitlines(options, into_wrapper)
     }
 
-    pub fn zfill(&self, width: isize) -> Vec<u8> {
-        self.elements.py_zfill(width)
+    pub fn zfill(&self, width: isize, vm: &VirtualMachine) -> PyResult<Vec<u8>> {
+        self.elements
+            .py_zfill(width)
+            .ok_or_else(|| vm.new_memory_error(""))
     }
 
     // len(self)>=1, from="", len(to)>=1, max_count>=1
@@ -1077,11 +1078,21 @@ impl AnyStrContainer<[u8]> for Vec<u8> {
         Self::with_capacity(capacity)
     }
 
+    fn try_with_capacity(capacity: usize) -> Option<Self> {
+        let mut v = Self::new();
+        v.try_reserve_exact(capacity).ok()?;
+        Some(v)
+    }
+
     fn push_str(&mut self, other: &[u8]) {
         self.extend(other)
     }
 }
 
+/// A padding function from `AnyStr`, returning `None` for a width whose result
+/// cannot be allocated.
+type PadFn = fn(&[u8], usize, u8, usize) -> Option<Vec<u8>>;
+
 const ASCII_WHITESPACES: [u8; 6] = [0x20, 0x09, 0x0a, 0x0c, 0x0d, 0x0b];
 
 impl anystr::AnyChar for u8 {

crates/vm/src/frame.rs

@@ -9362,9 +9362,14 @@ impl ExecutingFrame<'_> {
 
             if has_data_descr {
                 // Check for member descriptor (slot access)
+                // The slot offset only means anything on the layout the
+                // descriptor was defined for; the specialized instruction
+                // guards on the type version alone, so what descr_get()
+                // checks on every access has to be checked here instead.
                 if let Some(ref descr) = cls_attr
                     && let Some(member_descr) = descr.downcast_ref::<PyMemberDescriptor>()
                     && let MemberGetter::Offset(offset) = member_descr.member.getter
+                    && cls.fast_issubclass(&member_descr.common.typ)
                 {
                     unsafe {
                         self.code
@@ -11099,9 +11104,12 @@ impl ExecutingFrame<'_> {
 
         if has_data_descr {
             // Check for member descriptor (slot access)
+            // As in the load specialization, the offset is only valid for
+            // instances of the type the descriptor belongs to.
             if let Some(ref descr) = cls_attr
                 && let Some(member_descr) = descr.downcast_ref::<PyMemberDescriptor>()
                 && let MemberGetter::Offset(offset) = member_descr.member.getter
+                && cls.fast_issubclass(&member_descr.common.typ)
             {
                 unsafe {
                     self.code

crates/vm/src/function/buffer.rs

@@ -49,11 +49,41 @@ impl ArgBytesLike {
         f(&self.borrow_buf())
     }
 
+    /// The bytes to hand to an operation that may wait, and whatever keeps
+    /// them readable while it does.
+    ///
+    /// `borrow_buf` may answer with a lock that every other thread writing to
+    /// the same object waits on, and a thread waiting on a lock never reaches
+    /// a safepoint, so keeping one across a wait for a peer, a pipe or a
+    /// signal stops the world from being stopped at all. Bytes reached that
+    /// way are copied out first. Bytes that lock nothing -- an immutable
+    /// object's -- are borrowed where they lie, which is all CPython holds in
+    /// either case.
+    pub fn borrow_buf_unlocked(&self, vm: &VirtualMachine) -> PyResult<UnlockedBuf<'_>> {
+        let borrowed = self.borrow_buf();
+        if !borrowed.is_locked() {
+            return Ok(UnlockedBuf::Borrowed(borrowed));
+        }
+        let mut copy = Vec::new();
+        copy.try_reserve_exact(borrowed.len())
+            .map_err(|_| vm.new_memory_error(""))?;
+        copy.extend_from_slice(&borrowed);
+        Ok(UnlockedBuf::Copied(copy))
+    }
+
     #[must_use]
     pub const fn len(&self) -> usize {
         self.0.desc.len
     }
 
+    /// The width of one item. Callers that read the buffer as bytes rather
+    /// than as whatever it holds have to ask, since a contiguous buffer of
+    /// wider items is contiguous all the same.
+    #[must_use]
+    pub const fn itemsize(&self) -> usize {
+        self.0.desc.itemsize
+    }
+
     #[must_use]
     pub const fn is_empty(&self) -> bool {
         self.len() == 0
@@ -63,6 +93,16 @@ impl ArgBytesLike {
     pub fn as_object(&self) -> &PyObject {
         &self.0.obj
     }
+
+    /// The object whose storage is borrowed while this buffer is read: a view
+    /// borrows the object it looks at, not itself.
+    #[must_use]
+    pub fn source_object(&self) -> &PyObject {
+        self.0
+            .obj
+            .downcast_ref::<crate::builtins::PyMemoryView>()
+            .map_or(&self.0.obj, |view| view.viewed_object())
+    }
 }
 
 impl From<ArgBytesLike> for PyBuffer {
@@ -113,6 +153,24 @@ impl<'a> TryFromBorrowedObject<'a> for ArgContiguousBytesLike {
     }
 }
 
+/// Bytes that stay readable across a wait, from [`ArgBytesLike::borrow_buf_unlocked`].
+#[derive(Debug)]
+pub enum UnlockedBuf<'a> {
+    Borrowed(BorrowedValue<'a, [u8]>),
+    Copied(Vec<u8>),
+}
+
+impl core::ops::Deref for UnlockedBuf<'_> {
+    type Target = [u8];
+
+    fn deref(&self) -> &[u8] {
+        match self {
+            Self::Borrowed(b) => b,
+            Self::Copied(v) => v,
+        }
+    }
+}
+
 /// A memory buffer, read-write access. Like the `w*` format code for `PyArg_Parse` in CPython.
 #[derive(Debug, Traverse)]
 pub struct ArgMemoryBuffer(PyBuffer);
@@ -139,6 +197,16 @@ impl ArgMemoryBuffer {
     pub const fn is_empty(&self) -> bool {
         self.len() == 0
     }
+
+    /// The object whose storage is borrowed while this buffer is written: a
+    /// view borrows the object it looks at, not itself.
+    #[must_use]
+    pub fn source_object(&self) -> &PyObject {
+        self.0
+            .obj
+            .downcast_ref::<crate::builtins::PyMemoryView>()
+            .map_or(&self.0.obj, |view| view.viewed_object())
+    }
 }
 
 impl From<ArgMemoryBuffer> for PyBuffer {

crates/vm/src/gc_state.rs

@@ -8,7 +8,6 @@ use crate::object::{GC_NO_OWNER, GC_PERMANENT, GC_UNTRACKED, GcLink, GcOwner};
 use crate::{AsObject, PyObject, PyObjectRef};
 use core::ptr::NonNull;
 use core::sync::atomic::{AtomicBool, AtomicU16, AtomicU32, AtomicUsize, Ordering};
-use std::collections::HashSet;
 
 fn elapsed_secs(
     #[cfg(target_arch = "wasm32")] _start: (),
@@ -155,6 +154,45 @@ fn is_owned_by(obj: &PyObject, owner: GcOwner) -> bool {
 #[derive(Clone, Copy, PartialEq, Eq, Hash)]
 struct GcPtr(NonNull<PyObject>);
 
+/// Hashing for the tables a collection keys by an object's address.
+///
+/// The default hasher is SipHash, which buys resistance against a caller
+/// choosing keys that collide. Nothing chooses these keys: they are addresses
+/// this process handed out, and the tables live and die inside one collection.
+/// What a collection needs from them is speed -- it hashes every tracked
+/// object and every edge between them -- so this runs the address through a
+/// handful of multiplies and shifts instead. The shifts are what earns the
+/// speed: a table picks its bucket from the low bits, and an address arrives
+/// with its low bits zeroed by alignment, so entropy has to be carried
+/// downward or every object lands in the same few buckets.
+#[derive(Default)]
+struct GcPtrHasher(u64);
+
+impl core::hash::Hasher for GcPtrHasher {
+    fn finish(&self) -> u64 {
+        self.0
+    }
+
+    fn write_usize(&mut self, value: usize) {
+        let mut z = (value as u64).wrapping_add(0x9E37_79B9_7F4A_7C15);
+        z = (z ^ (z >> 30)).wrapping_mul(0xBF58_476D_1CE4_E5B9);
+        z = (z ^ (z >> 27)).wrapping_mul(0x94D0_49BB_1331_11EB);
+        self.0 = z ^ (z >> 31);
+    }
+
+    fn write(&mut self, bytes: &[u8]) {
+        // Addresses reach this hasher through `write_usize`; a key hashed any
+        // other way still has to land somewhere sensible.
+        for &byte in bytes {
+            self.0 = (self.0 ^ u64::from(byte)).wrapping_mul(0x0100_0000_01B3);
+        }
+    }
+}
+
+type GcBuildHasher = core::hash::BuildHasherDefault<GcPtrHasher>;
+type GcSet<T> = std::collections::HashSet<T, GcBuildHasher>;
+type GcMap<K, V> = std::collections::HashMap<K, V, GcBuildHasher>;
+
 /// RAII barrier that parks every other thread for the pointer-reading phases
 /// of a collection and lets them run again before finalizers execute.
 ///
@@ -556,14 +594,25 @@ impl GcState {
             retired.sort_unstable();
             retired
         };
-        let mut collecting: HashSet<GcPtr> = HashSet::new();
+        // The candidates and their reference counts go in one table, not a set
+        // beside a map: every edge in the heap is looked up here, and the two
+        // held the same keys, so a second table only bought a second hash of
+        // the same address. `candidate_ptrs` keeps them in a walkable order,
+        // since the counts are written while the candidates are read.
+        let mut gc_refs: GcMap<GcPtr, usize> = GcMap::default();
+        let mut candidate_ptrs: Vec<GcPtr> = Vec::new();
         for gen_list in &gen_locks {
             for obj in gen_list.iter() {
                 if retired.binary_search(&obj.gc_owner()).is_ok() {
                     obj.set_gc_owner(GC_NO_OWNER);
                 }
-                if obj.strong_count() > 0 && is_owned_by(obj, owner) {
-                    collecting.insert(GcPtr(NonNull::from(obj)));
+                let strong_count = obj.strong_count();
+                let ptr = GcPtr(NonNull::from(obj));
+                if strong_count > 0
+                    && is_owned_by(obj, owner)
+                    && gc_refs.insert(ptr, strong_count).is_none()
+                {
+                    candidate_ptrs.push(ptr);
                 }
             }
         }
@@ -583,7 +632,7 @@ impl GcState {
                 .retain(|tag| retired.binary_search(tag).is_err());
         }
 
-        if collecting.is_empty() {
+        if candidate_ptrs.is_empty() {
             // Reset counts for generations whose objects were promoted away.
             // For gen2 (oldest), survivors stay in-place so don't reset gen2 count.
             let reset_end = if generation >= 2 { 2 } else { generation + 1 };
@@ -602,26 +651,10 @@ impl GcState {
             };
         }
 
-        let candidates = collecting.len();
+        let candidates = candidate_ptrs.len();
 
         if debug.contains(GcDebugFlags::STATS) {
-            eprintln!(
-                "gc: collecting {} objects from generations 0..={}",
-                collecting.len(),
-                generation
-            );
-        }
-
-        // Step 2: Build gc_refs map (copy reference counts)
-        let mut gc_refs: std::collections::HashMap<GcPtr, usize> = std::collections::HashMap::new();
-
-        #[expect(
-            clippy::iter_over_hash_type,
-            reason = "Iteration order doesn't matter here"
-        )]
-        for &ptr in &collecting {
-            let obj = unsafe { ptr.0.as_ref() };
-            gc_refs.insert(ptr, obj.strong_count());
+            eprintln!("gc: collecting {candidates} objects from generations 0..={generation}");
         }
 
         // Step 3: Subtract internal references
@@ -630,32 +663,31 @@ impl GcState {
         // of each object's children. Without this, a dict whose write lock is
         // held during one traversal but not the other can yield inconsistent
         // results, causing live objects to be incorrectly collected.
-        let mut referents_map: std::collections::HashMap<GcPtr, Vec<NonNull<PyObject>>> =
-            std::collections::HashMap::new();
+        //
+        // Every object's referents go in one buffer, with each object holding
+        // the range that is its own: a vector each would be an allocation per
+        // tracked object, and the collection wants them all at once anyway.
+        let mut referent_ptrs: Vec<NonNull<PyObject>> = Vec::new();
+        let mut referent_ranges: GcMap<GcPtr, (usize, usize)> = GcMap::default();
 
-        #[expect(
-            clippy::iter_over_hash_type,
-            reason = "Iteration order doesn't matter here"
-        )]
-        for &ptr in &collecting {
+        for &ptr in &candidate_ptrs {
             let obj = unsafe { ptr.0.as_ref() };
             if obj.strong_count() == 0 {
                 continue;
             }
-            let referent_ptrs = unsafe { obj.gc_get_referent_ptrs() };
-            referents_map.insert(ptr, referent_ptrs.clone());
-            for child_ptr in referent_ptrs {
-                let gc_ptr = GcPtr(child_ptr);
-                if collecting.contains(&gc_ptr)
-                    && let Some(refs) = gc_refs.get_mut(&gc_ptr)
-                {
+            let start = referent_ptrs.len();
+            unsafe { obj.gc_extend_referent_ptrs(&mut referent_ptrs) };
+            let end = referent_ptrs.len();
+            for &child_ptr in &referent_ptrs[start..end] {
+                if let Some(refs) = gc_refs.get_mut(&GcPtr(child_ptr)) {
                     *refs = refs.saturating_sub(1);
                 }
             }
+            referent_ranges.insert(ptr, (start, end));
         }
 
         // Step 4: Find reachable objects (gc_refs > 0) and traverse from them
-        let mut reachable: HashSet<GcPtr> = HashSet::new();
+        let mut reachable: GcSet<GcPtr> = GcSet::default();
         let mut worklist: Vec<GcPtr> = Vec::new();
 
         #[expect(
@@ -672,16 +704,21 @@ impl GcState {
         while let Some(ptr) = worklist.pop() {
             let obj = unsafe { ptr.0.as_ref() };
             if obj.is_gc_tracked() {
-                // Reuse the pre-computed referent pointers from step 3.
-                // For objects that were skipped in step 3 (strong_count was 0),
-                // compute them now as a fallback.
-                let referent_ptrs = referents_map
-                    .get(&ptr)
-                    .cloned()
-                    .unwrap_or_else(|| unsafe { obj.gc_get_referent_ptrs() });
-                for child_ptr in referent_ptrs {
+                // Reuse the pre-computed referent pointers from step 3, in
+                // place: copying them out again costs a second pass over every
+                // edge in the heap. Objects skipped in step 3 (strong_count was
+                // 0) have none stored and are traversed here instead.
+                let computed;
+                let children: &[NonNull<PyObject>] = match referent_ranges.get(&ptr) {
+                    Some(&(start, end)) => &referent_ptrs[start..end],
+                    None => {
+                        computed = unsafe { obj.gc_get_referent_ptrs() };
+                        &computed
+                    }
+                };
+                for &child_ptr in children {
                     let gc_ptr = GcPtr(child_ptr);
-                    if collecting.contains(&gc_ptr) && reachable.insert(gc_ptr) {
+                    if gc_refs.contains_key(&gc_ptr) && reachable.insert(gc_ptr) {
                         worklist.push(gc_ptr);
                     }
                 }
@@ -689,7 +726,11 @@ impl GcState {
         }
 
         // Step 5: Find unreachable objects
-        let unreachable: Vec<GcPtr> = collecting.difference(&reachable).copied().collect();
+        let unreachable: Vec<GcPtr> = candidate_ptrs
+            .iter()
+            .filter(|ptr| !reachable.contains(ptr))
+            .copied()
+            .collect();
 
         // With the world stopped, every frame on any thread's call stack is a
         // live root that is externally referenced and must have been
@@ -702,7 +743,7 @@ impl GcState {
         // set_current_frame_nosave), not top_frame.
         #[cfg(all(unix, feature = "threading", debug_assertions))]
         if stw.is_stopped() {
-            let unreachable_set: HashSet<GcPtr> = unreachable.iter().copied().collect();
+            let unreachable_set: GcSet<GcPtr> = unreachable.iter().copied().collect();
             let mut cur = crate::vm::thread::get_current_frame();
             while !cur.is_null() {
                 let iframe = unsafe { &*cur };
@@ -802,7 +843,7 @@ impl GcState {
         }
 
         // 6b: Record initial strong counts (for resurrection detection)
-        let initial_counts: std::collections::HashMap<GcPtr, usize> = unreachable_refs
+        let initial_counts: GcMap<GcPtr, usize> = unreachable_refs
             .iter()
             .map(|obj| {
                 let ptr = GcPtr(core::ptr::NonNull::from(obj.as_ref()));
@@ -833,8 +874,8 @@ impl GcState {
         }
 
         // Detect resurrection
-        let mut resurrected_set: HashSet<GcPtr> = HashSet::new();
-        let unreachable_set: HashSet<GcPtr> = unreachable.iter().copied().collect();
+        let mut resurrected_set: GcSet<GcPtr> = GcSet::default();
+        let unreachable_set: GcSet<GcPtr> = unreachable.iter().copied().collect();
 
         for obj in &unreachable_refs {
             let ptr = GcPtr(core::ptr::NonNull::from(obj.as_ref()));
@@ -874,7 +915,7 @@ impl GcState {
 
         // Compute collected count (exclude instance dicts in truly_dead)
         let collected = {
-            let dead_ptrs: HashSet<usize> = truly_dead
+            let dead_ptrs: GcSet<usize> = truly_dead
                 .iter()
                 .map(|obj| obj.as_ref() as *const PyObject as usize)
                 .collect();
@@ -932,10 +973,9 @@ impl GcState {
             // never be observable through the generation lists, or another
             // thread could obtain a strong reference via gc.get_objects()
             // and access the cleared payload.
-            let mut late_resurrected: HashSet<GcPtr> = HashSet::new();
+            let mut late_resurrected: GcSet<GcPtr> = GcSet::default();
             if !save_all {
-                let mut expected_counts: std::collections::HashMap<GcPtr, usize> =
-                    std::collections::HashMap::new();
+                let mut expected_counts: GcMap<GcPtr, usize> = GcMap::default();
                 for obj_ref in &truly_dead {
                     let obj = obj_ref.as_ref();
                     if obj.is_gc_tracked() {
@@ -949,8 +989,7 @@ impl GcState {
                 // the dead set; any surplus in strong_count means another thread
                 // grabbed a reference before untracking (late resurrection) and
                 // the object must not be cleared.
-                let mut referents: std::collections::HashMap<GcPtr, Vec<NonNull<PyObject>>> =
-                    std::collections::HashMap::new();
+                let mut referents: GcMap<GcPtr, Vec<NonNull<PyObject>>> = GcMap::default();
                 for obj_ref in &truly_dead {
                     let referent_ptrs = unsafe { obj_ref.gc_get_referent_ptrs() };
                     for child_ptr in &referent_ptrs {

crates/vm/src/object/core.rs

@@ -1957,11 +1957,20 @@ impl PyObject {
     /// and its contents haven't been modified.
     pub unsafe fn gc_get_referent_ptrs(&self) -> Vec<NonNull<Self>> {
         let mut result = Vec::new();
+        unsafe { self.gc_extend_referent_ptrs(&mut result) };
+        result
+    }
+
+    /// Append this object's referents to `out`, for a caller that holds many
+    /// objects' referents in one buffer rather than one buffer each.
+    ///
+    /// # Safety
+    /// Same as [`Self::gc_get_referent_ptrs`].
+    pub unsafe fn gc_extend_referent_ptrs(&self, out: &mut Vec<NonNull<Self>>) {
         // Traverse the entire object including dict and slots
         self.0.traverse(&mut |child: &Self| {
-            result.push(NonNull::from(child));
+            out.push(NonNull::from(child));
         });
-        result
     }
 
     /// Pop edges from this object for cycle breaking.

crates/vm/src/protocol/buffer.rs

@@ -641,6 +641,47 @@ impl BufferDescriptor {
         }
     }
 
+    /// Visit each item's byte range with the *first* dimension varying
+    /// fastest, which is the order a Fortran-ordered copy is written in.
+    /// `for_each_segment` visits in the opposite order and can hand over whole
+    /// rows at once; here every item is its own range, since consecutive items
+    /// in this order are a row apart.
+    pub fn for_each_segment_fortran<F>(&self, mut f: F)
+    where
+        F: FnMut(Range<isize>),
+    {
+        if self.len == 0 {
+            return;
+        }
+        if self.ndim() == 0 {
+            f(self.offset..self.offset + self.itemsize as isize);
+            return;
+        }
+        let mut indices = vec![0usize; self.ndim()];
+        loop {
+            let pos = self.offset
+                + indices
+                    .iter()
+                    .zip_eq(self.dim_desc.iter())
+                    .map(|(&i, &(_, stride, suboffset))| i as isize * stride + suboffset)
+                    .sum::<isize>();
+            f(pos..pos + self.itemsize as isize);
+
+            let mut dim = 0;
+            loop {
+                indices[dim] += 1;
+                if indices[dim] < self.dim_desc[dim].0 {
+                    break;
+                }
+                indices[dim] = 0;
+                dim += 1;
+                if dim == self.ndim() {
+                    return;
+                }
+            }
+        }
+    }
+
     fn _for_each_segment<F, const CONTIGUOUS: bool>(&self, mut index: isize, dim: usize, f: &mut F)
     where
         F: FnMut(Range<isize>),

crates/vm/src/stdlib/_io.rs

@@ -1258,7 +1258,7 @@ mod _io {
 
             let current_size = self.readahead() as usize;
 
-            let mut out = vec![0u8; n];
+            let mut out = vm.new_zeroed_bytes(n)?;
             let mut remaining = n;
             let mut written = 0;
             if current_size > 0 {
@@ -1673,7 +1673,7 @@ mod _io {
                 check_writable(&raw, vm)?;
             }
 
-            data.buffer = vec![0; buffer_size];
+            data.buffer = vm.new_zeroed_bytes(buffer_size)?;
 
             if Self::READABLE {
                 data.reset_read();
@@ -1938,7 +1938,7 @@ mod _io {
             if data.writable() {
                 data.flush_rewind(vm)?;
             }
-            let mut v = vec![0; n];
+            let mut v = vm.new_zeroed_bytes(n)?;
             data.reset_read();
             let r = data
                 .raw_read(Either::A(Some(&mut v)), 0..n, vm)?
@@ -3364,14 +3364,17 @@ mod _io {
                 *snapshot = Some((cookie.dec_flags, input_chunk.clone()));
                 let decoded = vm.call_method(decoder, "decode", (input_chunk, cookie.need_eof))?;
                 let decoded = check_decoded(decoded, vm)?;
-                let pos_is_valid = decoded
-                    .as_wtf8()
-                    .is_code_point_boundary(cookie.bytes_to_skip as usize);
+                // The position is stored both as a count of characters and as
+                // an offset in bytes, so both have to land inside what was
+                // just decoded: everything read back from here indexes it.
+                let num_to_skip = cookie.num_to_skip();
+                let pos_is_valid = num_to_skip.chars <= decoded.char_len()
+                    && decoded.as_wtf8().is_code_point_boundary(num_to_skip.bytes);
                 textio.set_decoded_chars(Some(decoded));
                 if !pos_is_valid {
                     return Err(vm.new_os_error("can't restore logical file position"));
                 }
-                textio.decoded_chars_used = cookie.num_to_skip();
+                textio.decoded_chars_used = num_to_skip;
             } else {
                 textio.snapshot = Some((cookie.dec_flags, PyBytes::from(vec![]).into_ref(&vm.ctx)))
             }
@@ -4813,8 +4816,20 @@ mod _io {
         }
 
         #[pymethod]
-        fn readinto(&self, obj: ArgMemoryBuffer, vm: &VirtualMachine) -> PyResult<usize> {
-            let mut buf = self.buffer(vm)?;
+        fn readinto(zelf: &Py<Self>, obj: ArgMemoryBuffer, vm: &VirtualMachine) -> PyResult<usize> {
+            // Reading locks this object, and a destination that views it locks
+            // it too, so such a destination is filled after the read is done.
+            if obj.source_object().is(zelf.as_object()) {
+                let mut data = vm.new_zeroed_bytes(obj.len())?;
+                let ret = zelf
+                    .buffer(vm)?
+                    .cursor
+                    .read(&mut data)
+                    .map_err(|_| vm.new_value_error("Error readinto from Take"))?;
+                obj.borrow_buf_mut()[..ret].copy_from_slice(&data[..ret]);
+                return Ok(ret);
+            }
+            let mut buf = zelf.buffer(vm)?;
             let ret = buf
                 .cursor
                 .read(&mut obj.borrow_buf_mut())
@@ -5767,7 +5782,7 @@ mod fileio {
             }
             let handle = zelf.get_fd(vm)?;
             let bytes = if let Some(read_byte) = read_byte.to_usize() {
-                let mut bytes = vec![0; read_byte];
+                let mut bytes = vm.new_zeroed_bytes(read_byte)?;
                 // Loop on EINTR (PEP 475)
                 let n = loop {
                     match vm.allow_threads(|| host_io::read_once(handle, &mut bytes)) {
@@ -5811,6 +5826,26 @@ mod fileio {
             Ok(Some(bytes))
         }
 
+        /// One `read()` into `buf`, retried on EINTR (PEP 475). `None` on EAGAIN.
+        fn read_once_into(
+            zelf: &Py<Self>,
+            handle: crt_fd::Borrowed<'_>,
+            buf: &mut [u8],
+            vm: &VirtualMachine,
+        ) -> PyResult<Option<usize>> {
+            loop {
+                match vm.allow_threads(|| host_io::read_once(handle, buf)) {
+                    Ok(n) => return Ok(Some(n)),
+                    Err(e) if host_io::is_interrupted_error(&e) => {
+                        vm.check_signals()?;
+                    }
+                    // Non-blocking mode: return None if EAGAIN
+                    Err(e) if host_io::is_would_block_error(&e) => return Ok(None),
+                    Err(e) => return Err(Self::io_error(zelf, e, vm)),
+                }
+            }
+        }
+
         #[pymethod]
         fn readinto(
             zelf: &Py<Self>,
@@ -5826,24 +5861,28 @@ mod fileio {
 
             let handle = zelf.get_fd(vm)?;
 
-            let mut buf = obj.borrow_buf_mut();
-            // Loop on EINTR (PEP 475)
-            let ret = loop {
-                match vm.allow_threads(|| host_io::read_once(handle, &mut buf)) {
-                    Ok(n) => break n,
-                    Err(e) if host_io::is_interrupted_error(&e) => {
-                        vm.check_signals()?;
-                        continue;
-                    }
-                    // Non-blocking mode: return None if EAGAIN
-                    Err(e) if host_io::is_would_block_error(&e) => {
-                        return Ok(None);
-                    }
-                    Err(e) => return Err(Self::io_error(zelf, e, vm)),
-                }
-            };
+            if host_io::reads_without_waiting(handle) {
+                // The read answers from the file itself, so it returns without
+                // waiting on anyone; write where the caller asked directly.
+                // Seekability is not the question -- a pipe on Windows seeks.
+                let mut buf = obj.borrow_buf_mut();
+                return Self::read_once_into(zelf, handle, &mut buf, vm);
+            }
 
-            Ok(Some(ret))
+            // A pipe, socket or terminal answers only when the other end
+            // writes, which may be never. Holding the export for the whole
+            // call is what keeps the target from being resized meanwhile, as a
+            // Py_buffer does; but reaching its bytes takes a lock that every
+            // other thread touching the same object waits on, and a thread
+            // waiting on a lock never reaches a safepoint, so holding that one
+            // across the wait stops the world from being stopped at all. Read
+            // aside and take the lock for the copy.
+            let mut scratch = vm.new_zeroed_bytes(obj.len())?;
+            let ret = Self::read_once_into(zelf, handle, &mut scratch, vm)?;
+            if let Some(n) = ret {
+                obj.borrow_buf_mut()[..n].copy_from_slice(&scratch[..n]);
+            }
+            Ok(ret)
         }
 
         #[pymethod]
@@ -5861,9 +5900,14 @@ mod fileio {
 
             let handle = zelf.get_fd(vm)?;
 
+            // A pipe, socket or terminal takes the bytes only when the other
+            // end makes room, which may be never; see readinto above for what
+            // holding the source's lock across that wait costs.
+            let buf = obj.borrow_buf_unlocked(vm)?;
+
             // Loop on EINTR (PEP 475)
             let len = loop {
-                match obj.with_ref(|b| vm.allow_threads(|| host_io::write_once(handle, b))) {
+                match vm.allow_threads(|| host_io::write_once(handle, &buf)) {
                     Ok(n) => break n,
                     Err(e) if host_io::is_interrupted_error(&e) => {
                         vm.check_signals()?;

crates/vm/src/stdlib/_thread.rs

@@ -1207,9 +1207,9 @@ pub(crate) mod _thread {
                         // fall back to top_iframe (may be a stack-allocated frame).
                         let top = slot.top_frame.load(Ordering::Relaxed);
                         if let Some(p) = core::ptr::NonNull::new(top) {
-                            let py = unsafe {
-                                &*Py::<crate::frame::FrameObject>::from_payload_ptr(p.as_ptr())
-                            };
+                            // SAFETY: world stopped -> the owning thread is parked
+                            // with this frame on its chain, so it is alive.
+                            let py = unsafe { p.as_ref() };
                             Some((*id, py.to_owned()))
                         } else {
                             // Stack-allocated frame: materialize from top_iframe.

crates/vm/src/stdlib/atexit.rs

@@ -3,7 +3,9 @@ pub(crate) use atexit::module_def;
 
 #[pymodule]
 mod atexit {
-    use crate::{AsObject, PyObjectRef, PyResult, VirtualMachine, function::FuncArgs};
+    use crate::{
+        AsObject, PyObjectRef, PyResult, VirtualMachine, common::rc::PyRc, function::FuncArgs,
+    };
 
     #[pyfunction]
     fn register(func: PyObjectRef, args: FuncArgs, vm: &VirtualMachine) -> PyObjectRef {
@@ -11,7 +13,7 @@ mod atexit {
         vm.state
             .atexit_funcs
             .lock()
-            .insert(0, Box::new((func.clone(), args)));
+            .insert(0, PyRc::new((func.clone(), args)));
         func
     }
 
@@ -29,24 +31,26 @@ mod atexit {
             funcs.len() as isize - 1
         };
         while i >= 0 {
-            let (cb, entry_ptr) = {
+            let entry = {
                 let funcs = vm.state.atexit_funcs.lock();
                 if i as usize >= funcs.len() {
                     i = funcs.len() as isize;
                     i -= 1;
                     continue;
                 }
-                let entry = &funcs[i as usize];
-                (entry.0.clone(), &**entry as *const (PyObjectRef, FuncArgs))
+                // Keep the entry alive for as long as it is being compared, so
+                // it cannot be dropped and have its address handed to a
+                // callback registered from within __eq__.
+                funcs[i as usize].clone()
             };
             // Lock released: __eq__ can safely call atexit functions
-            let eq = vm.bool_eq(&func, &cb)?;
+            let eq = vm.bool_eq(&func, &entry.0)?;
             if eq {
                 // The entry may have moved during __eq__. Search backward by identity.
                 let mut funcs = vm.state.atexit_funcs.lock();
                 let mut j = (funcs.len() as isize - 1).min(i);
                 while j >= 0 {
-                    if core::ptr::eq(&**funcs.get(j as usize).unwrap(), entry_ptr) {
+                    if PyRc::ptr_eq(funcs.get(j as usize).unwrap(), &entry) {
                         funcs.remove(j as usize);
                         i = j;
                         break;
@@ -70,7 +74,7 @@ mod atexit {
         let funcs: Vec<_> = core::mem::take(&mut *vm.state.atexit_funcs.lock());
         // Callbacks stored in LIFO order, iterate forward
         for entry in funcs {
-            let (func, args) = *entry;
+            let (func, args) = PyRc::try_unwrap(entry).unwrap_or_else(|e| (*e).clone());
             if let Err(e) = func.call(args, vm) {
                 let exit = e.fast_isinstance(vm.ctx.exceptions.system_exit);
                 let msg = func

crates/vm/src/stdlib/marshal.rs

@@ -116,9 +116,6 @@ mod decl {
             )?;
         }
 
-        if !allow_code {
-            check_no_code(&value, vm)?;
-        }
         check_exact_type(&value, vm)?;
         let mut buf = Vec::new();
         let mut refs = if version >= 3 {
@@ -126,7 +123,7 @@ mod decl {
         } else {
             None
         };
-        write_object(&mut buf, &value, &mut refs, version, vm)?;
+        write_object(&mut buf, &value, &mut refs, version, allow_code, vm)?;
         Ok(PyBytes::from(buf))
     }
 
@@ -185,6 +182,7 @@ mod decl {
         obj: &PyObjectRef,
         refs: &mut Option<WriterRefTable>,
         version: i32,
+        allow_code: bool,
         vm: &VirtualMachine,
     ) -> PyResult<()> {
         write_object_depth(
@@ -192,6 +190,7 @@ mod decl {
             obj,
             refs,
             version,
+            allow_code,
             vm,
             marshal::MAX_MARSHAL_STACK_DEPTH,
         )
@@ -202,6 +201,7 @@ mod decl {
         obj: &PyObjectRef,
         refs: &mut Option<WriterRefTable>,
         version: i32,
+        allow_code: bool,
         vm: &VirtualMachine,
         depth: usize,
     ) -> PyResult<()> {
@@ -322,20 +322,20 @@ mod decl {
             buf.write_u8(b'(');
             buf.write_u32(t.len() as u32);
             for elem in t.as_slice() {
-                write_object_depth(buf, elem, refs, version, vm, depth - 1)?;
+                write_object_depth(buf, elem, refs, version, allow_code, vm, depth - 1)?;
             }
         } else if let Some(l) = obj.downcast_ref::<PyList>() {
             buf.write_u8(b'[');
             let items = l.borrow_vec();
             buf.write_u32(items.len() as u32);
             for elem in items.iter() {
-                write_object_depth(buf, elem, refs, version, vm, depth - 1)?;
+                write_object_depth(buf, elem, refs, version, allow_code, vm, depth - 1)?;
             }
         } else if let Some(d) = obj.downcast_ref::<PyDict>() {
             buf.write_u8(b'{');
             for (k, v) in d {
-                write_object_depth(buf, &k, refs, version, vm, depth - 1)?;
-                write_object_depth(buf, &v, refs, version, vm, depth - 1)?;
+                write_object_depth(buf, &k, refs, version, allow_code, vm, depth - 1)?;
+                write_object_depth(buf, &v, refs, version, allow_code, vm, depth - 1)?;
             }
             buf.write_u8(b'0'); // TYPE_NULL terminator
         } else if let Some(s) = obj.downcast_ref::<PySet>() {
@@ -343,16 +343,19 @@ mod decl {
             let elems = s.elements();
             buf.write_u32(elems.len() as u32);
             for elem in &elems {
-                write_object_depth(buf, elem, refs, version, vm, depth - 1)?;
+                write_object_depth(buf, elem, refs, version, allow_code, vm, depth - 1)?;
             }
         } else if let Some(s) = obj.downcast_ref::<PyFrozenSet>() {
             buf.write_u8(b'>');
             let elems = s.elements();
             buf.write_u32(elems.len() as u32);
             for elem in &elems {
-                write_object_depth(buf, elem, refs, version, vm, depth - 1)?;
+                write_object_depth(buf, elem, refs, version, allow_code, vm, depth - 1)?;
             }
         } else if let Some(co) = obj.downcast_ref::<PyCode>() {
+            if !allow_code {
+                return Err(vm.new_value_error("marshalling code objects is disallowed"));
+            }
             buf.write_u8(b'c');
             // `Literal` holds the exact object a constant was built from, so
             // route `co_consts` back through the object writer: it reaches the
@@ -360,7 +363,7 @@ mod decl {
             // reference table the reader indexes against.
             marshal::serialize_code_with(buf, &co.code, |buf, constant| {
                 let constant = PyObjectRef::from(constant.clone());
-                write_object_depth(buf, &constant, refs, version, vm, depth - 1)
+                write_object_depth(buf, &constant, refs, version, allow_code, vm, depth - 1)
             })?;
         } else if let Some(sl) = obj.downcast_ref::<crate::builtins::PySlice>() {
             if version < 5 {
@@ -373,15 +376,17 @@ mod decl {
                 sl.start.as_ref().unwrap_or(&none),
                 refs,
                 version,
+                allow_code,
                 vm,
                 depth - 1,
             )?;
-            write_object_depth(buf, &sl.stop, refs, version, vm, depth - 1)?;
+            write_object_depth(buf, &sl.stop, refs, version, allow_code, vm, depth - 1)?;
             write_object_depth(
                 buf,
                 sl.step.as_ref().unwrap_or(&none),
                 refs,
                 version,
+                allow_code,
                 vm,
                 depth - 1,
             )?;
@@ -431,14 +436,36 @@ mod decl {
     struct PyMarshalBag<'a> {
         vm: &'a VirtualMachine,
         pending_error: &'a RefCell<Option<PyBaseExceptionRef>>,
+        allow_code: bool,
     }
 
     impl<'a> PyMarshalBag<'a> {
         fn new(
             vm: &'a VirtualMachine,
             pending_error: &'a RefCell<Option<PyBaseExceptionRef>>,
+            allow_code: bool,
         ) -> Self {
-            Self { vm, pending_error }
+            Self {
+                vm,
+                pending_error,
+                allow_code,
+            }
+        }
+
+        /// Room for a container the decoder publishes before it reads what
+        /// goes in it. The length is the input's to choose, so the room is
+        /// asked for rather than assumed: a length no allocator can serve is
+        /// a MemoryError, not an aborted process.
+        fn placeholder_elements(
+            &self,
+            len: usize,
+        ) -> Result<Vec<PyObjectRef>, marshal::MarshalError> {
+            let mut elements = Vec::new();
+            elements
+                .try_reserve_exact(len)
+                .map_err(|_| self.remember_python_error(self.vm.new_memory_error("")))?;
+            elements.resize(len, self.vm.ctx.none());
+            Ok(elements)
         }
 
         fn remember_python_error(&self, error: PyBaseExceptionRef) -> marshal::MarshalError {
@@ -484,8 +511,12 @@ mod decl {
         fn make_tuple(&self, elements: impl Iterator<Item = Self::Value>) -> Self::Value {
             self.vm.ctx.new_tuple(elements.collect()).into()
         }
-        fn make_tuple_placeholder(&self, len: usize) -> Option<Self::Value> {
-            Some(PyTuple::new_marshal_placeholder(len, &self.vm.ctx).into())
+        fn make_tuple_placeholder(
+            &self,
+            len: usize,
+        ) -> Result<Option<Self::Value>, marshal::MarshalError> {
+            let elements = self.placeholder_elements(len)?;
+            Ok(Some(PyTuple::new_ref(elements, &self.vm.ctx).into()))
         }
         fn set_tuple_item(
             &self,
@@ -501,8 +532,14 @@ mod decl {
             unsafe { tuple.set_marshal_item(index, value) };
             Ok(())
         }
-        fn make_code(&self, code: CodeObject) -> Self::Value {
-            crate::builtins::PyCode::new_ref_with_bag(self.vm, code).into()
+        fn make_code(&self, code: CodeObject) -> Result<Self::Value, marshal::MarshalError> {
+            if !self.allow_code {
+                return Err(self.remember_python_error(
+                    self.vm
+                        .new_value_error("unmarshalling code objects is disallowed"),
+                ));
+            }
+            Ok(crate::builtins::PyCode::new_ref_with_bag(self.vm, code).into())
         }
         fn make_stop_iter(&self) -> Result<Self::Value, marshal::MarshalError> {
             Ok(self.vm.ctx.exceptions.stop_iteration.to_owned().into())
@@ -513,8 +550,12 @@ mod decl {
         ) -> Result<Self::Value, marshal::MarshalError> {
             Ok(self.vm.ctx.new_list(it.collect()).into())
         }
-        fn make_list_placeholder(&self, len: usize) -> Option<Self::Value> {
-            Some(self.vm.ctx.new_list(vec![self.vm.ctx.none(); len]).into())
+        fn make_list_placeholder(
+            &self,
+            len: usize,
+        ) -> Result<Option<Self::Value>, marshal::MarshalError> {
+            let elements = self.placeholder_elements(len)?;
+            Ok(Some(self.vm.ctx.new_list(elements).into()))
         }
         fn set_list_item(
             &self,
@@ -635,13 +676,20 @@ mod decl {
 
     fn deserialize_value(
         rdr: &mut impl marshal::Read,
+        allow_code: bool,
         vm: &VirtualMachine,
     ) -> PyResult<PyObjectRef> {
         let pending_error = RefCell::new(None);
-        match marshal::deserialize_value(rdr, PyMarshalBag::new(vm, &pending_error)) {
+        match marshal::deserialize_value(rdr, PyMarshalBag::new(vm, &pending_error, allow_code)) {
             Ok(value) => Ok(value),
             Err(error) => Err(pending_error.into_inner().unwrap_or_else(|| match error {
                 marshal::MarshalError::Eof => vm.new_eof_error("marshal data too short"),
+                error @ marshal::MarshalError::NullObject => vm.new_type_error(error.to_string()),
+                error @ (marshal::MarshalError::BadSize(_)
+                | marshal::MarshalError::UnknownType
+                | marshal::MarshalError::InvalidRef) => {
+                    vm.new_value_error(format!("bad marshal data ({error})"))
+                }
                 _ => vm.new_value_error("bad marshal data"),
             })),
         }
@@ -661,11 +709,7 @@ mod decl {
         let LoadsArgs { data, allow_code } = args;
         let buf = data.borrow_buf();
 
-        let result = deserialize_value(&mut &buf[..], vm)?;
-        if !allow_code {
-            check_no_code(&result, vm)?;
-        }
-        Ok(result)
+        deserialize_value(&mut &buf[..], allow_code, vm)
     }
 
     #[derive(FromArgs)]
@@ -685,54 +729,25 @@ mod decl {
             .try_into_value::<i64>(vm)?;
         let read_res = vm.call_method(&args.f, "read", ())?;
         let bytes = ArgBytesLike::try_from_object(vm, read_res)?;
-        let buf = bytes.borrow_buf();
 
-        let mut rdr: &[u8] = &buf;
-        let len_before = rdr.len();
-        let result = deserialize_value(&mut rdr, vm)?;
-        let consumed = len_before - rdr.len();
+        // The borrow ends here: seek() below is the caller's, and reaching the
+        // same buffer from it would deadlock on a borrow still held.
+        let (result, consumed) = {
+            let buf = bytes.borrow_buf();
+            let mut rdr: &[u8] = &buf;
+            let len_before = rdr.len();
+            let result = deserialize_value(&mut rdr, args.allow_code, vm)?;
+            (result, len_before - rdr.len())
+        };
 
         // Seek file to just after the consumed bytes
         let new_pos = tell_before + consumed as i64;
         vm.call_method(&args.f, "seek", (new_pos,))?;
 
-        if !args.allow_code {
-            check_no_code(&result, vm)?;
-        }
         Ok(result)
     }
 
     /// Reject subclasses of marshallable types (int, float, complex, tuple, etc.).
-    /// Recursively check that no code objects are present.
-    fn check_no_code(obj: &PyObjectRef, vm: &VirtualMachine) -> PyResult<()> {
-        if obj.downcast_ref::<PyCode>().is_some() {
-            return Err(vm.new_value_error("unmarshalling code objects is disallowed"));
-        }
-        if let Some(tup) = obj.downcast_ref::<PyTuple>() {
-            for elem in tup.as_slice() {
-                check_no_code(elem, vm)?;
-            }
-        } else if let Some(list) = obj.downcast_ref::<PyList>() {
-            for elem in list.borrow_vec().iter() {
-                check_no_code(elem, vm)?;
-            }
-        } else if let Some(set) = obj.downcast_ref::<PySet>() {
-            for elem in set.elements() {
-                check_no_code(&elem, vm)?;
-            }
-        } else if let Some(fset) = obj.downcast_ref::<PyFrozenSet>() {
-            for elem in fset.elements() {
-                check_no_code(&elem, vm)?;
-            }
-        } else if let Some(dict) = obj.downcast_ref::<PyDict>() {
-            for (k, v) in dict {
-                check_no_code(&k, vm)?;
-                check_no_code(&v, vm)?;
-            }
-        }
-        Ok(())
-    }
-
     fn check_exact_type(obj: &PyObjectRef, vm: &VirtualMachine) -> PyResult<()> {
         let cls = obj.class();
         // bool is a subclass of int but is marshallable

crates/vm/src/stdlib/typevar.rs

@@ -923,11 +923,12 @@ pub(crate) mod typevar {
     impl Representable for ParamSpecArgs {
         #[inline(always)]
         fn repr_str(zelf: &crate::Py<Self>, vm: &VirtualMachine) -> PyResult<String> {
-            // Check if origin is a ParamSpec
-            if let Ok(name) = zelf.__origin__.get_attr("__name__", vm) {
-                return Ok(format!("{name}.args", name = name.str(vm)?));
+            // A ParamSpec origin is named; anything else is shown by its repr,
+            // which carries the recursion guard a Rust `{:?}` walk does not.
+            if let Some(param_spec) = zelf.__origin__.downcast_ref::<ParamSpec>() {
+                return Ok(format!("{}.args", param_spec.__name__().str_utf8(vm)?));
             }
-            Ok(format!("{:?}.args", zelf.__origin__))
+            Ok(format!("{}.args", zelf.__origin__.repr(vm)?))
         }
     }
 
@@ -986,11 +987,12 @@ pub(crate) mod typevar {
     impl Representable for ParamSpecKwargs {
         #[inline(always)]
         fn repr_str(zelf: &crate::Py<Self>, vm: &VirtualMachine) -> PyResult<String> {
-            // Check if origin is a ParamSpec
-            if let Ok(name) = zelf.__origin__.get_attr("__name__", vm) {
-                return Ok(format!("{name}.kwargs", name = name.str(vm)?));
+            // A ParamSpec origin is named; anything else is shown by its repr,
+            // which carries the recursion guard a Rust `{:?}` walk does not.
+            if let Some(param_spec) = zelf.__origin__.downcast_ref::<ParamSpec>() {
+                return Ok(format!("{}.kwargs", param_spec.__name__().str_utf8(vm)?));
             }
-            Ok(format!("{:?}.kwargs", zelf.__origin__))
+            Ok(format!("{}.kwargs", zelf.__origin__.repr(vm)?))
         }
     }
 

crates/vm/src/types/slot.rs

@@ -527,7 +527,11 @@ pub fn hash_not_implemented(zelf: &PyObject, vm: &VirtualMachine) -> PyResult<Py
 }
 
 fn call_wrapper(zelf: &PyObject, args: FuncArgs, vm: &VirtualMachine) -> PyResult {
-    vm.call_special_method(zelf, identifier!(vm, __call__), args)
+    // `__call__` can name the object being called, and dispatching it pushes no
+    // Python frame, so nothing else counts the nesting.
+    vm.with_recursion("while calling a Python object", || {
+        vm.call_special_method(zelf, identifier!(vm, __call__), args)
+    })
 }
 
 fn getattro_wrapper(zelf: &PyObject, name: &Py<PyStr>, vm: &VirtualMachine) -> PyResult {
@@ -616,7 +620,11 @@ fn descr_get_wrapper(
     cls: Option<PyObjectRef>,
     vm: &VirtualMachine,
 ) -> PyResult {
-    vm.call_special_method(&zelf, identifier!(vm, __get__), (obj, cls))
+    // A descriptor whose `__get__` is the descriptor itself resolves it by
+    // fetching `__get__` again, and none of that pushes a Python frame.
+    vm.with_recursion("while calling a Python object", || {
+        vm.call_special_method(&zelf, identifier!(vm, __get__), (obj, cls))
+    })
 }
 
 fn descr_set_wrapper(

crates/vm/src/vm/mod.rs

@@ -100,6 +100,11 @@ pub struct VirtualMachine {
     pub state: PyRc<PyGlobalState>,
     pub initialized: bool,
     recursion_depth: Cell<usize>,
+    /// Depth of native recursion that pushes no Python frame, counted only
+    /// where the stack pointer cannot be read. Everywhere else the native
+    /// stack itself answers, and nothing needs counting.
+    #[cfg(any(miri, target_env = "musl"))]
+    native_recursion_depth: Cell<usize>,
     /// C stack soft limit for detecting stack overflow (like c_stack_soft_limit)
     #[cfg_attr(any(miri, target_env = "musl"), allow(dead_code))]
     c_stack_soft_limit: Cell<usize>,
@@ -384,7 +389,7 @@ impl StopTheWorldState {
     /// is only ever `try_lock`'d. The active requester therefore force-parks
     /// this thread, finishes its whole stop→start span, releases the exclusion,
     /// and only then does this thread resume and acquire it.
-    fn acquire_exclusion(&self) {
+    fn acquire_exclusion(&self, state: &PyGlobalState) {
         if self
             .exclusion
             .compare_exchange(false, true, Ordering::AcqRel, Ordering::Relaxed)
@@ -393,7 +398,7 @@ impl StopTheWorldState {
             return;
         }
         loop {
-            crate::vm::thread::suspend_if_needed(self);
+            crate::vm::thread::suspend_if_needed(state);
             std::thread::yield_now();
             if self
                 .exclusion
@@ -421,7 +426,7 @@ impl StopTheWorldState {
     /// drives the stop→start span at a time; it is released by
     /// `start_the_world`/`reset_after_fork`.
     pub fn stop_the_world(&self, state: &PyGlobalState) {
-        self.acquire_exclusion();
+        self.acquire_exclusion(state);
         let start = std::time::Instant::now();
         let requester_ident = crate::stdlib::_thread::get_ident();
         self.requester.store(requester_ident, Ordering::Relaxed);
@@ -759,7 +764,10 @@ pub struct PyGlobalState {
     pub stacksize: AtomicCell<usize>,
     pub thread_count: AtomicCell<usize>,
     pub hash_secret: HashSecret,
-    pub atexit_funcs: PyMutex<Vec<Box<(PyObjectRef, FuncArgs)>>>,
+    /// Registered `atexit` callbacks, newest first. Shared ownership so
+    /// `atexit.unregister` can keep the entry it is comparing alive while the
+    /// list is unlocked, and still recognize it afterwards by identity.
+    pub atexit_funcs: PyMutex<Vec<PyRc<(PyObjectRef, FuncArgs)>>>,
     pub codec_registry: CodecsRegistry,
     pub finalizing: AtomicBool,
     pub warnings: WarningsState,
@@ -991,6 +999,8 @@ impl VirtualMachine {
             state,
             initialized: false,
             recursion_depth: Cell::new(0),
+            #[cfg(any(miri, target_env = "musl"))]
+            native_recursion_depth: Cell::new(0),
             c_stack_soft_limit: Cell::new(Self::calculate_c_stack_soft_limit()),
             async_gen_firstiter: RefCell::new(None),
             async_gen_finalizer: RefCell::new(None),
@@ -2004,6 +2014,14 @@ impl VirtualMachine {
     const STACK_MARGIN_BYTES: usize =
         (if cfg!(debug_assertions) { 16384 } else { 4096 }) * core::mem::size_of::<usize>();
 
+    /// How deep native recursion may go where the stack cannot be measured
+    /// (`Py_C_RECURSION_LIMIT`). A native step costs far more stack than a
+    /// Python one and debug builds cost more again, so this sits well under
+    /// what a default stack holds rather than at what it would just fit.
+    #[cfg(any(miri, target_env = "musl"))]
+    const NATIVE_RECURSION_LIMIT_UNMEASURED: usize =
+        if cfg!(debug_assertions) { 500 } else { 1500 };
+
     /// Get the stack boundaries using platform-specific APIs.
     /// Returns (base, top) where base is the lowest address and top is the highest.
     #[cfg(all(not(miri), not(target_env = "musl"), windows))]
@@ -2105,16 +2123,34 @@ impl VirtualMachine {
     /// Used to run the body of a (possibly) recursive function. It will raise a
     /// RecursionError if recursive functions are nested far too many times,
     /// preventing a stack overflow.
+    /// `Py_EnterRecursiveCall`: bounds native recursion that pushes no Python
+    /// frame, against the native stack. That is a separate budget from the
+    /// frame limit `sys.setrecursionlimit()` sets, so nesting counted here does
+    /// not come out of what Python code has left to call with.
     pub fn with_recursion<R, F: FnOnce() -> PyResult<R>>(&self, _where: &str, f: F) -> PyResult<R> {
-        self.check_recursive_call(_where)?;
+        // `check_c_stack_overflow()` answers no unconditionally where the stack
+        // pointer cannot be read, which would leave this guard with nothing to
+        // stop. A count of the nesting stands in for the measurement there.
+        #[cfg(any(miri, target_env = "musl"))]
+        let counted_too_deep =
+            self.native_recursion_depth.get() >= Self::NATIVE_RECURSION_LIMIT_UNMEASURED;
+        #[cfg(not(any(miri, target_env = "musl")))]
+        let counted_too_deep = false;
 
-        // Native stack guard: check C stack like _Py_MakeRecCheck
-        if self.check_c_stack_overflow() {
-            return Err(self.new_recursion_error(_where.to_string()));
+        if counted_too_deep || self.check_c_stack_overflow() {
+            return Err(
+                self.new_recursion_error(format!("maximum recursion depth exceeded {_where}"))
+            );
         }
 
-        self.recursion_depth.update(|d| d + 1);
-        scopeguard::defer! { self.recursion_depth.update(|d| d - 1) }
+        #[cfg(any(miri, target_env = "musl"))]
+        let _native_depth_guard = {
+            self.native_recursion_depth.update(|d| d + 1);
+            scopeguard::guard((), |()| {
+                self.native_recursion_depth.update(|d| d.saturating_sub(1))
+            })
+        };
+
         f()
     }
 
@@ -2603,12 +2639,28 @@ impl VirtualMachine {
         // Objects/listobject.c. Each branch takes an atomic snapshot to avoid
         // race conditions from concurrent mutation (no GIL).
         let cls = value.class();
-        let list_borrow;
         let slice = if cls.is(self.ctx.types.tuple_type) {
             value.downcast_ref::<PyTuple>().unwrap().as_slice()
         } else if cls.is(self.ctx.types.list_type) {
-            list_borrow = value.downcast_ref::<PyList>().unwrap().borrow_vec();
-            &list_borrow
+            // The list is re-read on every step, the way map_iterable_object()
+            // does it: func() runs Python, which can mutate or even clear the
+            // same list, and a borrow held across that call deadlocks it.
+            let list = value.downcast_ref::<PyList>().unwrap();
+            let mut results = Vec::new();
+            let mut i = 0;
+            loop {
+                let elem = {
+                    let elements = list.borrow_vec();
+                    let Some(elem) = elements.get(i) else {
+                        break;
+                    };
+                    elem.clone()
+                    // free the lock
+                };
+                results.push(func(elem)?);
+                i += 1;
+            }
+            return Ok(results);
         } else if cls.is(self.ctx.types.dict_type) {
             let keys = value.downcast_ref::<PyDict>().unwrap().keys_vec();
             return keys.into_iter().map(func).collect();
@@ -2797,7 +2849,7 @@ impl VirtualMachine {
 
         // Suspend this thread if stop-the-world is in progress
         #[cfg(feature = "threading")]
-        thread::suspend_if_needed(&self.state.stop_the_world);
+        thread::suspend_if_needed(&self.state);
 
         // Pass a QSBR checkpoint if requested (deferred memory reclamation).
         #[cfg(feature = "threading")]

crates/vm/src/vm/thread.rs

@@ -5,11 +5,13 @@ use crate::builtins::PyBaseExceptionRef;
 #[cfg(feature = "threading")]
 use alloc::sync::Arc;
 
-#[cfg(all(unix, feature = "threading"))]
-use crate::frame::FrameObject;
 use crate::frame::InterpreterFrame;
+#[cfg(feature = "threading")]
+use crate::vm::PyGlobalState;
 use crate::{AsObject, PyObject, VirtualMachine};
 #[cfg(all(unix, feature = "threading"))]
+use crate::{Py, frame::FrameObject};
+#[cfg(all(unix, feature = "threading"))]
 use core::sync::atomic::AtomicPtr;
 use core::{
     cell::{Cell, RefCell},
@@ -44,7 +46,7 @@ pub struct ThreadSlot {
     /// thread at a safepoint and supplies the happens-before edge, so the
     /// pointer and the frames it reaches are quiescent and alive at read time.
     #[cfg(unix)]
-    pub top_frame: AtomicPtr<FrameObject>,
+    pub top_frame: AtomicPtr<Py<FrameObject>>,
     /// Raw InterpreterFrame pointer, published alongside top_frame so
     /// cross-thread readers (sys._current_frames) can materialize
     /// stack-allocated frames that have no FrameObject.
@@ -114,7 +116,7 @@ thread_local! {
     /// initialized; the `Arc<ThreadSlot>` in `CURRENT_THREAD_SLOT` keeps the
     /// pointee alive until `cleanup_current_thread_frames` clears this.
     #[cfg(all(unix, feature = "threading"))]
-    static CURRENT_TOP_FRAME_SLOT: Cell<*const AtomicPtr<FrameObject>> =
+    static CURRENT_TOP_FRAME_SLOT: Cell<*const AtomicPtr<Py<FrameObject>>> =
         const { Cell::new(core::ptr::null()) };
 
     /// Cached pointer to this thread's `ThreadSlot::top_iframe` for the hot
@@ -536,9 +538,10 @@ fn attach_thread(vm: &VirtualMachine) {
     // a thread doing rapid allow_threads calls from re-attaching and running
     // past the requester forever, which would stall stop-the-world. Done
     // outside the CURRENT_THREAD_SLOT borrow above because suspend re-borrows
-    // it. Safe against a concurrent start_the_world: suspend_if_needed only
-    // parks while the request is still live and self-recovers otherwise.
-    suspend_if_needed(&vm.state.stop_the_world);
+    // it. Safe against a concurrent start_the_world: suspend_if_needed decides
+    // whether to park under the registry lock, so it never parks after the
+    // request has been withdrawn.
+    suspend_if_needed(&vm.state);
 }
 
 /// Transition ATTACHED → DETACHED (like `_PyThreadState_Detach`).
@@ -605,102 +608,111 @@ pub fn allow_threads<R>(_vm: &VirtualMachine, f: impl FnOnce() -> R) -> R {
 /// Transitions ATTACHED → SUSPENDED and waits until released
 /// (like `_PyThreadState_Suspend` + `_PyThreadState_Attach`).
 #[cfg(feature = "threading")]
-pub fn suspend_if_needed(stw: &super::StopTheWorldState) {
+pub fn suspend_if_needed(state: &PyGlobalState) {
     let should_suspend = CURRENT_THREAD_SLOT.with(|slot| {
         slot.borrow()
             .as_ref()
             .is_some_and(|s| s.stop_requested.load(Ordering::Relaxed))
     });
-    if !should_suspend {
-        return;
+    if should_suspend {
+        do_suspend(state);
     }
-
-    if !stw.requested.load(Ordering::Acquire) {
-        CURRENT_THREAD_SLOT.with(|slot| {
-            if let Some(s) = slot.borrow().as_ref() {
-                s.stop_requested.store(false, Ordering::Release);
-            }
-        });
-        return;
-    }
-
-    do_suspend(stw);
 }
 
 #[cfg(feature = "threading")]
 #[cold]
-fn do_suspend(stw: &super::StopTheWorldState) {
+fn do_suspend(state: &PyGlobalState) {
+    let stw = &state.stop_the_world;
     CURRENT_THREAD_SLOT.with(|slot| {
-        if let Some(s) = slot.borrow().as_ref() {
-            // ATTACHED → SUSPENDED
+        let borrowed = slot.borrow();
+        let Some(s) = borrowed.as_ref() else {
+            return;
+        };
+
+        // Decide whether to park while holding the thread registry. Both edges
+        // of `requested` are written under that lock: `init_thread_countdown`
+        // sets it, and `start_the_world` clears it and then releases every
+        // SUSPENDED thread without letting go. Publishing SUSPENDED here is
+        // therefore either seen by that release pass or never reached, which
+        // leaves the requester the only writer that takes a thread out of
+        // SUSPENDED. A completion check that observed this thread parked cannot
+        // then be invalidated by the thread resuming on its own.
+        let park = {
+            let _registry = state.thread_frames.lock();
+            if stw.requested.load(Ordering::Acquire) {
+                Some(s.state.compare_exchange(
+                    THREAD_ATTACHED,
+                    THREAD_SUSPENDED,
+                    Ordering::AcqRel,
+                    Ordering::Acquire,
+                ))
+            } else {
+                // The stop already ended; this thread's request bit is stale.
+                s.stop_requested.store(false, Ordering::Release);
+                None
+            }
+        };
+
+        match park {
+            None => {
+                super::stw_trace(format_args!("suspend skip not-requested"));
+                return;
+            }
+            Some(Ok(_)) => {
+                // Consumed this thread's stop request bit.
+                s.stop_requested.store(false, Ordering::Release);
+            }
+            Some(Err(THREAD_DETACHED)) => {
+                // Leaving VM; caller will re-check on next entry.
+                super::stw_trace(format_args!("suspend skip DETACHED"));
+                return;
+            }
+            Some(Err(THREAD_SUSPENDED)) => {
+                // Already parked by another path.
+                s.stop_requested.store(false, Ordering::Release);
+                super::stw_trace(format_args!("suspend skip already-suspended"));
+                return;
+            }
+            Some(Err(state)) => {
+                debug_assert!(false, "unexpected thread state in suspend: {state}");
+                return;
+            }
+        }
+        super::stw_trace(format_args!("suspend ATTACHED->SUSPENDED"));
+
+        // Notify the stop-the-world requester that we've parked. The registry
+        // is released first: the requester's wait loop takes the notify mutex
+        // and then the registry, so taking them the other way round here would
+        // invert the order.
+        stw.notify_suspended();
+        super::stw_trace(format_args!("suspend notified-requester"));
+
+        // Wait until start_the_world sets us back to DETACHED
+        let wait_yields = wait_while_suspended(s);
+        stw.add_suspend_wait_yields(wait_yields);
+
+        // Re-attach (DETACHED → ATTACHED), tstate_wait_attach CAS loop.
+        loop {
             match s.state.compare_exchange(
+                THREAD_DETACHED,
                 THREAD_ATTACHED,
-                THREAD_SUSPENDED,
                 Ordering::AcqRel,
                 Ordering::Acquire,
             ) {
-                Ok(_) => {
-                    // Consumed this thread's stop request bit.
-                    s.stop_requested.store(false, Ordering::Release);
-                }
-                Err(THREAD_DETACHED) => {
-                    // Leaving VM; caller will re-check on next entry.
-                    super::stw_trace(format_args!("suspend skip DETACHED"));
-                    return;
-                }
+                Ok(_) => break,
                 Err(THREAD_SUSPENDED) => {
-                    // Already parked by another path.
-                    s.stop_requested.store(false, Ordering::Release);
-                    super::stw_trace(format_args!("suspend skip already-suspended"));
-                    return;
+                    let extra_wait = wait_while_suspended(s);
+                    stw.add_suspend_wait_yields(extra_wait);
                 }
+                Err(THREAD_ATTACHED) => break,
                 Err(state) => {
-                    debug_assert!(false, "unexpected thread state in suspend: {state}");
-                    return;
+                    debug_assert!(false, "unexpected post-suspend state: {state}");
+                    break;
                 }
             }
-            super::stw_trace(format_args!("suspend ATTACHED->SUSPENDED"));
-
-            // Re-check: if start_the_world already ran (cleared `requested`),
-            // no one will set us back to DETACHED — we must self-recover.
-            if !stw.requested.load(Ordering::Acquire) {
-                s.state.store(THREAD_ATTACHED, Ordering::Release);
-                s.stop_requested.store(false, Ordering::Release);
-                super::stw_trace(format_args!("suspend abort requested-cleared"));
-                return;
-            }
-
-            // Notify the stop-the-world requester that we've parked
-            stw.notify_suspended();
-            super::stw_trace(format_args!("suspend notified-requester"));
-
-            // Wait until start_the_world sets us back to DETACHED
-            let wait_yields = wait_while_suspended(s);
-            stw.add_suspend_wait_yields(wait_yields);
-
-            // Re-attach (DETACHED → ATTACHED), tstate_wait_attach CAS loop.
-            loop {
-                match s.state.compare_exchange(
-                    THREAD_DETACHED,
-                    THREAD_ATTACHED,
-                    Ordering::AcqRel,
-                    Ordering::Acquire,
-                ) {
-                    Ok(_) => break,
-                    Err(THREAD_SUSPENDED) => {
-                        let extra_wait = wait_while_suspended(s);
-                        stw.add_suspend_wait_yields(extra_wait);
-                    }
-                    Err(THREAD_ATTACHED) => break,
-                    Err(state) => {
-                        debug_assert!(false, "unexpected post-suspend state: {state}");
-                        break;
-                    }
-                }
-            }
-            s.stop_requested.store(false, Ordering::Release);
-            super::stw_trace(format_args!("suspend resume -> ATTACHED"));
         }
+        s.stop_requested.store(false, Ordering::Release);
+        super::stw_trace(format_args!("suspend resume -> ATTACHED"));
     });
 }
 
@@ -818,11 +830,8 @@ pub fn set_current_frame(frame: *const InterpreterFrame) -> *const InterpreterFr
                     core::ptr::null_mut()
                 } else {
                     let frame_obj = unsafe { (*frame).frame_obj() };
-                    // The payload address, which is what the cross-thread
-                    // reader hands to `Py::from_payload_ptr`. The `Py` address
-                    // would be off by the object header.
                     frame_obj.map_or(core::ptr::null_mut(), |py| {
-                        core::ptr::from_ref::<FrameObject>(py).cast_mut()
+                        py as *const Py<FrameObject> as *mut Py<FrameObject>
                     })
                 };
                 unsafe { &*slot }.store(fo_ptr, Ordering::Relaxed);
@@ -971,7 +980,7 @@ pub fn reinit_frame_slot_after_fork(vm: &VirtualMachine) {
             core::ptr::null_mut()
         } else {
             match unsafe { (*top_iframe).frame_obj() } {
-                Some(fo) => core::ptr::from_ref::<FrameObject>(fo).cast_mut(),
+                Some(fo) => fo as *const Py<FrameObject> as *mut Py<FrameObject>,
                 None => core::ptr::null_mut(),
             }
         }
@@ -1163,6 +1172,8 @@ impl VirtualMachine {
             state: self.state.clone(),
             initialized: self.initialized,
             recursion_depth: Cell::new(0),
+            #[cfg(any(miri, target_env = "musl"))]
+            native_recursion_depth: Cell::new(0),
             c_stack_soft_limit: Cell::new(Self::calculate_c_stack_soft_limit()),
             async_gen_firstiter: RefCell::new(None),
             async_gen_finalizer: RefCell::new(None),

crates/vm/src/vm/vm_ops.rs

@@ -168,6 +168,27 @@ impl VirtualMachine {
         }
     }
 
+    /// `vec![0; len]` for a length that came from Python, where a request too
+    /// large to satisfy is a `MemoryError` rather than an aborted process.
+    ///
+    /// The bytes are left for the allocator to zero, so a large request costs
+    /// no more than the pages that are actually written to.
+    pub fn new_zeroed_bytes(&self, len: usize) -> PyResult<Vec<u8>> {
+        if len == 0 {
+            return Ok(Vec::new());
+        }
+        let layout =
+            core::alloc::Layout::array::<u8>(len).map_err(|_| self.new_memory_error(""))?;
+        // SAFETY: `len` is not zero, so neither is the layout's size.
+        let ptr = unsafe { alloc::alloc::alloc_zeroed(layout) };
+        if ptr.is_null() {
+            return Err(self.new_memory_error(""));
+        }
+        // SAFETY: `ptr` was just allocated by the global allocator for exactly
+        // this many bytes, and every one of them is initialized to zero.
+        Ok(unsafe { Vec::from_raw_parts(ptr, len, len) })
+    }
+
     /// Calling scheme used for binary operations:
     ///
     /// Order operations are tried until either a valid result or error:

extra_tests/snippets/builtin_bytes.py

@@ -747,3 +747,22 @@ assert not s.istitle(), f"{s}"
 assert "123A".istitle(), f"{s}"
 assert not "123a".istitle(), f"{s}"
 assert not "123A\ta".istitle(), f"{s}"
+
+
+def test_huge_size():
+    # sizes that cannot be allocated are MemoryError, not an aborted process
+    for factory in (bytes, bytearray):
+        assert_raises(MemoryError, lambda factory=factory: factory(2**62))
+        for meth in ("center", "ljust", "rjust", "zfill"):
+            assert_raises(
+                MemoryError,
+                lambda factory=factory, meth=meth: getattr(factory(b"a"), meth)(
+                    1 << 62
+                ),
+            )
+        assert_raises(
+            OverflowError, lambda factory=factory: factory(b"\ta").expandtabs(2**31)
+        )
+
+
+test_huge_size()

extra_tests/snippets/builtin_hash.py

@@ -35,9 +35,10 @@ with assert_raises(TypeError):
 # slot dispatch is what recurses, so that is where the depth is checked.
 
 if sys.implementation.name == "rustpython":
-    # CPython, which also runs this snippet, survives this depth unguarded.
+    # Deep enough to reach the native stack guard; CPython, which also runs
+    # this snippet, dies on the same value.
     deep_tuple = ()
-    for _ in range(sys.getrecursionlimit() * 2):
+    for _ in range(100_000):
         deep_tuple = (deep_tuple,)
     with assert_raises(RecursionError):
         hash(deep_tuple)

extra_tests/snippets/builtin_memoryview.py

@@ -514,3 +514,164 @@ def test_fortran_contiguity():
 
 
 test_fortran_contiguity()
+
+
+def test_cast_arguments():
+    # cast() takes a native single character format, optionally '@'-prefixed;
+    # a zero-size format used to reach a division by zero.
+    assert memoryview(b"abcd").cast("@i").itemsize == 4
+    for fmt in ("0s", "4s", "<i", "", "ss"):
+        assert_raises(ValueError, lambda fmt=fmt: memoryview(b"abcd").cast(fmt))
+
+    # every element of shape is an int > 0; a 0 used to divide by zero while
+    # checking the product against SSIZE_MAX
+    for shape in ([0], [0, 4], [4, 0], [-1, 4], [0, 0]):
+        assert_raises(
+            ValueError, lambda shape=shape: memoryview(b"abcd").cast("B", shape)
+        )
+
+    class Index:
+        def __index__(self):
+            return 4
+
+    for shape in ([2.0, 2], [Index()], ["4"]):
+        assert_raises(
+            TypeError, lambda shape=shape: memoryview(b"abcd").cast("B", shape)
+        )
+
+    assert memoryview(b"abcd").cast("B", [True, 4]).tolist() == [[97, 98, 99, 100]]
+
+
+test_cast_arguments()
+
+
+def test_negative_stride():
+    # A reversed view starts at its last byte, so walking it from there runs
+    # off the front of the exported slice.
+    assert memoryview(b"dcba") == memoryview(b"abcd")[::-1]
+    assert memoryview(b"abcd")[::-1] == memoryview(b"dcba")
+    assert not memoryview(b"abcd") == memoryview(b"abcd")[::-1]
+
+    b = bytearray(b"____")
+    memoryview(b)[0:4] = memoryview(b"abcd")[::-1]
+    assert b == bytearray(b"dcba"), b
+
+    a = array.array("i", [1, 2, 3])
+    assert memoryview(array.array("i", [3, 2, 1])) == memoryview(a)[::-1]
+    assert memoryview(a)[::-1].tolist() == [3, 2, 1]
+
+
+test_negative_stride()
+
+
+def test_write_through_same_object():
+    # Reading the source and writing the destination lock the same object
+    # when they overlap, and converting a value runs Python that can reach it.
+    b = bytearray(b"abcd")
+    memoryview(b)[0:4] = b
+    assert b == bytearray(b"abcd"), b
+
+    b = bytearray(b"abcd")
+    memoryview(b)[0:4] = memoryview(b)[::-1]
+    assert b == bytearray(b"dcba"), b
+
+    b = bytearray(b"abcd")
+    memoryview(b)[0:2] = memoryview(b)[2:4]
+    assert b == bytearray(b"cdcd"), b
+
+    b = bytearray(b"abcd")
+    view = memoryview(b)
+
+    class Index:
+        def __index__(self):
+            view[1] = 66
+            return 65
+
+    view[0] = Index()
+    assert b == bytearray(b"ABcd"), b
+
+
+test_write_through_same_object()
+
+
+def test_cast_between_non_byte_formats():
+    # A cast re-divides bytes into items; going from one item type straight to
+    # another would reinterpret what is already there.
+    view = memoryview(b"abcd").cast("i")
+    for fmt in ("h", "i", "f"):
+        try:
+            view.cast(fmt)
+        except TypeError as e:
+            assert "cannot cast between two non-byte formats" in str(e), e
+        else:
+            raise AssertionError(f"expected TypeError for cast to {fmt!r}")
+
+    # Either side being bytes is allowed.
+    assert view.cast("B").tolist() == [97, 98, 99, 100]
+    assert view.cast("b").format == "b"
+    assert view.cast("c").tolist() == [b"a", b"b", b"c", b"d"]
+    assert memoryview(b"abcd").cast("c").cast("i").format == "i"
+
+
+def test_cast_to_zero_dim():
+    # A zero-dimensional view holds exactly one item, so the buffer has to be
+    # that one item and no more.
+    assert memoryview(b"abcd").cast("I", shape=()).tobytes() == b"abcd"
+    assert memoryview(b"a").cast("B", shape=()).tobytes() == b"a"
+
+    for source, fmt in ((b"abcd", "B"), (b"abcdefgh", "I"), (b"ab", "b")):
+        try:
+            memoryview(source).cast(fmt, shape=())
+        except TypeError as e:
+            assert "product(shape) * itemsize != buffer size" in str(e), e
+        else:
+            raise AssertionError(f"expected TypeError for {source!r} as {fmt!r}")
+
+
+def test_hash_restricted_to_byte_formats():
+    # The hash is over the bytes, so it agrees with the hash of those bytes
+    # only where an item is a byte.
+    data = b"abcdefgh"
+    assert hash(memoryview(data)) == hash(data)
+    assert hash(memoryview(data).cast("c")) == hash(data)
+    assert hash(memoryview(data).cast("b")) == hash(data)
+
+    for fmt in ("I", "i", "h", "d"):
+        try:
+            hash(memoryview(data).cast(fmt))
+        except ValueError as e:
+            assert "hashing is restricted to formats" in str(e), e
+        else:
+            raise AssertionError(f"expected ValueError for format {fmt!r}")
+
+
+def test_tobytes_order():
+    view = memoryview(b"abcdefgh")
+    for order in (None, "C", "F", "A"):
+        assert view.tobytes(order=order) == b"abcdefgh", order
+
+    # A multidimensional view is laid out C-contiguously, so a Fortran-ordered
+    # copy walks it down the columns instead.
+    grid = memoryview(b"abcdefgh").cast("B", shape=(2, 4))
+    assert grid.tolist() == [[97, 98, 99, 100], [101, 102, 103, 104]]
+    assert grid.tobytes() == b"abcdefgh"
+    assert grid.tobytes(order="C") == b"abcdefgh"
+    assert grid.tobytes(order="A") == b"abcdefgh"
+    assert grid.tobytes(order="F") == b"aebfcgdh"
+
+    cube = memoryview(b"abcdefgh").cast("B", shape=(2, 2, 2))
+    assert cube.tobytes(order="F") == b"aecgbfdh"
+
+    for order in ("Z", "c", "f", ""):
+        try:
+            view.tobytes(order=order)
+        except ValueError as e:
+            assert str(e) == "order must be 'C', 'F' or 'A'", e
+        else:
+            raise AssertionError(f"expected ValueError for order {order!r}")
+
+
+test_cast_between_non_byte_formats()
+test_cast_to_zero_dim()
+test_hash_restricted_to_byte_formats()
+test_tobytes_order()

extra_tests/snippets/builtin_str.py

@@ -900,3 +900,18 @@ assert id(b) != id(b * 0)
 assert id(b) != id(b * 1)
 assert id(b) != id(1 * b)
 assert id(b) != id(b * 2)
+
+
+def test_huge_width():
+    # A width that cannot be allocated is a MemoryError, not an aborted
+    # process, and a tabsize wider than a C int does not fit at all.
+    for meth in ("center", "ljust", "rjust", "zfill"):
+        assert_raises(MemoryError, lambda meth=meth: getattr("a", meth)(1 << 62))
+    assert_raises(OverflowError, lambda: "\ta".expandtabs(1 << 62))
+    assert_raises(OverflowError, lambda: "\ta".expandtabs(2**31))
+    # The widest tabsize that still fits is accepted. With no tab to expand
+    # there is nothing to lay out, so the width is never allocated.
+    assert "a".expandtabs(2**31 - 1) == "a"
+
+
+test_huge_width()

extra_tests/snippets/builtin_type.py

@@ -687,3 +687,33 @@ def foo():
 
 code = compile(stmts, "<test>", "exec")
 assert code.co_names == ("blah", "foo")
+
+
+# A slot descriptor carries the layout it was defined for. Reached from another
+# class, it has to report that rather than read the slot at its own offset,
+# whether the access is fresh or has been seen often enough to be specialized.
+
+
+class WideSlots:
+    __slots__ = ("s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7")
+
+
+class NarrowSlots:
+    __slots__ = ("only",)
+
+
+class NoSlots:
+    __slots__ = ()
+
+
+NarrowSlots.borrowed = WideSlots.__dict__["s7"]
+NoSlots.borrowed = WideSlots.__dict__["s7"]
+
+for owner in (NarrowSlots(), NoSlots()):
+    for _ in range(1000):
+        with assert_raises(TypeError):
+            owner.borrowed
+        with assert_raises(TypeError):
+            owner.borrowed = 1
+        with assert_raises(TypeError):
+            del owner.borrowed

extra_tests/snippets/recursion.py

@@ -11,3 +11,36 @@ foo = Foo()
 # Since the default __str__ implementation calls __repr__ and __repr__ is
 # actually __str__, str(foo) should raise a RecursionError.
 assert_raises(RecursionError, str, foo)
+
+
+# A __call__ that is the object being called dispatches through the call slot
+# again, and none of that pushes a Python frame.
+
+
+class Caller:
+    pass
+
+
+caller = Caller()
+Caller.__call__ = caller
+assert_raises(RecursionError, caller)
+
+
+# The same shape through the descriptor protocol: resolving the attribute
+# fetches __get__, which is the descriptor itself.
+
+
+class Descr:
+    pass
+
+
+descr = Descr()
+Descr.__get__ = descr
+Descr.x = descr
+try:
+    descr.x
+except (RecursionError, TypeError):
+    # RecursionError here, TypeError from the call of a non-callable elsewhere
+    pass
+else:
+    raise AssertionError("descr.x should not resolve")

extra_tests/snippets/stdlib_array.py

@@ -143,3 +143,36 @@ class _ReenteringWriter:
 arr = array("b", range(128))
 arr.tofile(_ReenteringWriter(arr))
 assert len(arr) == 129
+
+
+def test_setitem_reentrant():
+    # Converting the value runs Python, which can reach the array, so the
+    # array is not locked while it happens.
+    a = array("i", [1, 2, 3])
+
+    class Index:
+        def __index__(self):
+            a[1] = 9
+            return 7
+
+    a[0] = Index()
+    assert a == array("i", [7, 9, 3]), a
+
+
+test_setitem_reentrant()
+
+
+def test_frombytes_of_itself():
+    # Resizing is refused while a buffer is exported, before any lock is taken.
+    # The typecode is "b" so the view's items are bytes and the resize is what
+    # the call is refused for.
+    a = array("b", [1, 2, 3])
+    m = memoryview(a)
+    with assert_raises(BufferError):
+        a.frombytes(m)
+    del m
+
+    # A view of wider items is not a source of bytes at all.
+    wide = array("i", [1, 2, 3])
+    with assert_raises(TypeError):
+        wide.frombytes(memoryview(wide))

extra_tests/snippets/stdlib_asyncio.py

@@ -72,4 +72,31 @@ finally:
     asyncio.InvalidStateError = saved_invalid_state_error
     asyncio.exceptions.InvalidStateError = saved_invalid_state_error
 
+# The awaited-by set is built with the waiter's __hash__, which can come back
+# to the same future; the field must not be locked while that runs.
+
+
+class Reentrant:
+    def __hash__(self):
+        _asyncio.future_add_to_awaited_by(awaited, Reentrant())
+        return 1
+
+    def __eq__(self, other):
+        return self is other
+
+
+awaited = _asyncio.Future(loop=object())
+_asyncio.future_add_to_awaited_by(awaited, Reentrant())
+with assert_raises(RecursionError):
+    # converting the single waiter into a set hashes both of them
+    _asyncio.future_add_to_awaited_by(awaited, Reentrant())
+
+plain = _asyncio.Future(loop=object())
+waiter = object()
+_asyncio.future_add_to_awaited_by(plain, waiter)
+_asyncio.future_add_to_awaited_by(plain, object())
+assert waiter in plain._asyncio_awaited_by
+_asyncio.future_discard_from_awaited_by(plain, waiter)
+assert waiter not in plain._asyncio_awaited_by
+
 print("ok")

extra_tests/snippets/stdlib_atexit.py

@@ -0,0 +1,101 @@
+"""atexit.unregister() compares callbacks with arbitrary Python code.
+
+The comparison runs with the callback list unlocked, so __eq__ may clear it
+and register something new. unregister() then has to tell whether the entry
+it compared is still there, and must not mistake a later registration that
+happens to occupy the same storage for that entry.
+"""
+
+import atexit
+
+
+def make(name):
+    def f():
+        ran.append(name)
+
+    f.tag = name
+    return f
+
+
+ran = []
+a, b, c, d = (make(n) for n in "abcd")
+
+
+class Probe:
+    def __init__(self, action=None, result=True):
+        self.action = action
+        self.result = result
+        self.seen = []
+
+    def __eq__(self, other):
+        self.seen.append(getattr(other, "tag", "?"))
+        if self.action is not None:
+            self.action()
+        return self.result
+
+
+def remaining():
+    del ran[:]
+    atexit._run_exitfuncs()
+    return list(ran)
+
+
+# A callback the probe does not match is left alone.
+atexit._clear()
+atexit.register(a)
+atexit.register(b)
+probe = Probe(result=False)
+atexit.unregister(probe)
+assert probe.seen == ["a", "b"], probe.seen
+assert remaining() == ["b", "a"], ran
+
+# Matching callbacks are dropped, oldest compared first.
+atexit._clear()
+atexit.register(a)
+atexit.register(b)
+atexit.register(c)
+probe = Probe(result=True)
+atexit.unregister(probe)
+assert probe.seen == ["a", "b", "c"], probe.seen
+assert atexit._ncallbacks() == 0
+assert remaining() == [], ran
+
+# __eq__ empties the list: there is nothing left to drop.
+atexit._clear()
+atexit.register(a)
+atexit.register(b)
+probe = Probe(action=atexit._clear, result=True)
+atexit.unregister(probe)
+assert probe.seen == ["a"], probe.seen
+assert remaining() == [], ran
+
+# __eq__ empties the list and registers a replacement. The replacement is a
+# different callback, so it survives however its storage was reused.
+atexit._clear()
+atexit.register(a)
+atexit.register(b)
+atexit.register(c)
+
+
+def replace():
+    atexit._clear()
+    atexit.register(d)
+
+
+probe = Probe(action=replace, result=True)
+atexit.unregister(probe)
+assert probe.seen == ["a", "d"], probe.seen
+assert remaining() == ["d"], ran
+
+# __eq__ registers without clearing: every entry the walk had already passed
+# stays, and so does each newly registered one.
+atexit._clear()
+atexit.register(a)
+atexit.register(b)
+probe = Probe(action=lambda: atexit.register(c), result=True)
+atexit.unregister(probe)
+assert probe.seen == ["a", "c"], probe.seen
+assert remaining() == ["c", "c", "b", "a"], ran
+
+atexit._clear()
+print("ok")

extra_tests/snippets/stdlib_hashlib.py

@@ -2,6 +2,8 @@ import _md5
 import _sha1
 import hashlib
 
+from testutils import assert_raises
+
 # print(hashlib.md5)
 h = hashlib.md5()
 h.update(b"a")
@@ -56,3 +58,9 @@ assert (
 
 assert _md5.md5(b"").hexdigest() == "d41d8cd98f00b204e9800998ecf8427e"
 assert _sha1.sha1(b"").hexdigest() == "da39a3ee5e6b4b0d3255bfef95601890afd80709"
+
+# a derived key wider than a C int does not fit, and never gets allocated.
+# Which OverflowError comes out depends on the width of a C long: where it is
+# narrower than the length asked for, converting the argument fails first.
+with assert_raises(OverflowError):
+    hashlib.pbkdf2_hmac("sha256", b"password", b"salt", 1, 2**62)

extra_tests/snippets/stdlib_io.py

@@ -197,3 +197,48 @@ expect_value_error(
     f"cannot fit '{truncated_non_ascii_type_name}' into an index-sized integer",
     lambda: setattr(textio, "_CHUNK_SIZE", NonAsciiNamedChunkSize()),
 )
+
+
+# A buffer size or read size that cannot be allocated is a MemoryError, not an
+# aborted process.
+assert_raises(MemoryError, lambda: BufferedReader(BytesIO(b"a"), buffer_size=2**62))
+assert_raises(MemoryError, lambda: BufferedReader(BytesIO(b"a")).read(2**62))
+assert_raises(MemoryError, lambda: BufferedReader(BytesIO(b"a")).read1(2**62))
+
+
+def _text_cookie(
+    start_pos=0,
+    dec_flags=0,
+    bytes_to_feed=0,
+    chars_to_skip=0,
+    need_eof=0,
+    bytes_to_skip=0,
+):
+    packed = (
+        start_pos.to_bytes(8, "little", signed=True)
+        + dec_flags.to_bytes(4, "little", signed=True)
+        + bytes_to_feed.to_bytes(4, "little", signed=True)
+        + chars_to_skip.to_bytes(4, "little", signed=True)
+        + bytes([need_eof])
+        + bytes_to_skip.to_bytes(4, "little", signed=True)
+    )
+    return int.from_bytes(packed, "little")
+
+
+# A cookie names a position both in characters and in bytes, and everything
+# read back from it indexes what was decoded, so a position past the end is
+# refused rather than stored.
+for _bad in (
+    _text_cookie(bytes_to_feed=10, chars_to_skip=1000, bytes_to_skip=0),
+    _text_cookie(bytes_to_feed=10, chars_to_skip=100000, bytes_to_skip=3),
+    _text_cookie(bytes_to_feed=10, chars_to_skip=1, bytes_to_skip=1000),
+):
+    _textio = TextIOWrapper(BytesIO(b"hello world " * 20), encoding="utf-8")
+    _textio.read(1)
+    try:
+        _textio.seek(_bad)
+    except (OSError, OverflowError):
+        pass
+    else:
+        assert _textio.read(50) is not None
+        _textio.tell()

extra_tests/snippets/stdlib_io_blocking_buffer.py

@@ -0,0 +1,176 @@
+"""Transfers that wait for a peer must not hold the buffer they were given.
+
+A pipe or a socket answers when the other end does, which may be never. The
+buffer is exported for the whole call, so it cannot be resized meanwhile, but
+nothing else about it changes: another thread can still read it, write to it,
+and the interpreter can still stop the world. An implementation that holds the
+buffer's storage for the duration of the wait takes all of that away, and a
+thread parked on that storage never reaches a safepoint, so a collection that
+wants every thread stopped ends up waiting for the peer too.
+"""
+
+import gc
+import os
+import socket
+import threading
+import time
+
+# The peer acts after DELAY; the checks below have to finish well inside it.
+DELAY = 1.0
+SLACK = DELAY / 2
+
+
+def measure(buf, expected_len, writable):
+    """Time each operation on `buf` that does not need the peer, separately, so
+    a failure names the one that waited rather than the group."""
+    elapsed = {}
+
+    def timed(name, operation):
+        start = time.monotonic()
+        value = operation()
+        elapsed[name] = time.monotonic() - start
+        return value
+
+    assert timed("len", lambda: len(buf)) == expected_len, len(buf)
+    assert isinstance(timed("bytes", lambda: bytes(buf)), bytes)
+    if writable:
+        timed("setitem", lambda: buf.__setitem__(0, buf[0]))
+    timed("gc.collect", gc.collect)
+    return elapsed
+
+
+def run(buf, blocking_call, release_peer, writable):
+    started = threading.Event()
+    expected_len = len(buf)
+    result = []
+
+    def transfer():
+        started.set()
+        result.append(blocking_call(buf))
+
+    def peer():
+        time.sleep(DELAY)
+        release_peer()
+
+    threads = [threading.Thread(target=transfer), threading.Thread(target=peer)]
+    for t in threads:
+        t.start()
+    started.wait()
+    time.sleep(0.2)  # the transfer is now waiting on its peer
+
+    elapsed = measure(buf, expected_len, writable)
+    waited = ["%s %.2fs" % item for item in elapsed.items() if item[1] >= SLACK]
+    assert not waited, "waited on the peer: " + ", ".join(waited)
+
+    # The transfer is still in flight, so its export is still held and the
+    # buffer cannot be resized. An operating system that took the whole
+    # transfer without a peer leaves nothing here to observe.
+    assert not result, "the transfer finished without its peer"
+    try:
+        buf.append(0)
+    except BufferError:
+        pass
+    else:
+        raise AssertionError("append during an export should raise BufferError")
+
+    for t in threads:
+        t.join()
+    return result[0]
+
+
+# --- reading: the buffer is written into, so nothing else may touch it at all
+
+
+read_fd, write_fd = os.pipe()
+pipe = open(read_fd, "rb", buffering=0)
+try:
+    target = bytearray(16)
+    n = run(target, pipe.readinto, lambda: os.write(write_fd, b"pipe"), writable=False)
+    assert n == 4, n
+    assert bytes(target[:4]) == b"pipe", bytes(target)
+finally:
+    pipe.close()
+    os.close(write_fd)
+
+if hasattr(socket, "socketpair"):
+    left, right = socket.socketpair()
+    try:
+        target = bytearray(16)
+        n = run(target, left.recv_into, lambda: right.send(b"socket"), writable=False)
+        assert n == 6, n
+        assert bytes(target[:6]) == b"socket", bytes(target)
+    finally:
+        left.close()
+        right.close()
+
+
+# --- writing: the buffer is only read, so it stays writable meanwhile
+
+
+read_fd, write_fd = os.pipe()
+sink = open(write_fd, "wb", buffering=0)
+try:
+    # More than any pipe will hold, so the write cannot finish on its own.
+    source = bytearray(4 * 1024 * 1024)
+    drained = []
+
+    def drain():
+        with open(read_fd, "rb", buffering=0) as f:
+            while True:
+                chunk = f.read(1 << 16)
+                if not chunk:
+                    break
+                drained.append(len(chunk))
+
+    reader = threading.Thread(target=drain, daemon=True)
+    # One unbuffered write() reports what it transferred, which a signal can
+    # cut short, so the reader is measured against that rather than the source.
+    written = run(source, sink.write, reader.start, writable=True)
+    sink.close()
+    reader.join()
+    assert sum(drained) == written, (sum(drained), written)
+finally:
+    if not sink.closed:
+        sink.close()
+
+if hasattr(socket, "socketpair"):
+    left, right = socket.socketpair()
+    try:
+        # How much a connection holds before it makes the sender wait is the
+        # operating system's to decide, and asking for a small send buffer does
+        # not settle it -- a socketpair is already connected, and on Windows it
+        # is a loopback pair whose receiver has a window of its own. So fill it
+        # until it refuses rather than guess a size that outruns it.
+        left.setblocking(False)
+        filled = 0
+        while True:
+            try:
+                filled += left.send(bytes(1 << 16))
+            except (BlockingIOError, InterruptedError):
+                break
+        left.setblocking(True)
+
+        source = bytearray(1 << 16)
+        received = []
+
+        def receive():
+            wanted = filled + len(source)
+            while sum(received) < wanted:
+                chunk = right.recv(1 << 16)
+                if not chunk:
+                    break
+                received.append(len(chunk))
+
+        reader = threading.Thread(target=receive, daemon=True)
+        run(source, left.sendall, reader.start, writable=True)
+        reader.join()
+        assert sum(received) == filled + len(source), (
+            sum(received),
+            filled,
+            len(source),
+        )
+    finally:
+        left.close()
+        right.close()
+
+print("ok")

extra_tests/snippets/stdlib_io_bytesio.py

@@ -106,3 +106,11 @@ if __name__ == "__main__":
     test_05()
     test_06()
     test_07()
+
+
+# Reading into a buffer that views this same object locks it twice unless the
+# read finishes first.
+_bio = BytesIO(b"x" * 60)
+assert _bio.readinto(_bio.getbuffer()) == 60
+_bio = BytesIO(b"x" * 60)
+assert _bio.readinto(memoryview(_bio.getbuffer())) == 60

extra_tests/snippets/stdlib_marshal.py

@@ -96,5 +96,63 @@ class MarshalTests(unittest.TestCase):
         self.assertIs(loaded_code.co_consts[0], loaded_shared)
 
 
+class AllowCodeTests(unittest.TestCase):
+    """allow_code is answered where a code object is written or read, so a
+    graph that walks back on itself is not a second walk of its own."""
+
+    def test_recursive_value(self):
+        recursive = []
+        recursive.append(recursive)
+        loaded = marshal.loads(
+            marshal.dumps(recursive, allow_code=False), allow_code=False
+        )
+        self.assertIs(loaded[0], loaded)
+
+    def test_too_deeply_nested(self):
+        nested = []
+        for _ in range(100_000):
+            nested = [nested]
+        with self.assertRaises(ValueError):
+            marshal.dumps(nested, allow_code=False)
+
+    def test_code_is_rejected(self):
+        code = compile("1", "", "exec")
+        for value in (code, [code], (code,), {0: code}):
+            with self.assertRaises(ValueError):
+                marshal.dumps(value, allow_code=False)
+            data = marshal.dumps(value)
+            with self.assertRaises(ValueError):
+                marshal.loads(data, allow_code=False)
+
+
+class BadDataTests(unittest.TestCase):
+    def test_container_size_out_of_range(self):
+        import struct
+
+        # a length is signed, so the top bit set is out of range rather than
+        # four billion items to reserve room for
+        for marker in b"([<>":
+            data = bytes([marker | 0x80]) + struct.pack("<I", 0xFFFFFF00)
+            with self.assertRaises(ValueError):
+                marshal.loads(data)
+
+    def test_unknown_type_code(self):
+        for data in (b"\x00", b"\x01", b"?", b"\xff"):
+            with self.assertRaises(ValueError):
+                marshal.loads(data)
+
+    def test_null_object(self):
+        # TYPE_NULL stands for no object, which is not a value to hand back
+        with self.assertRaises(TypeError):
+            marshal.loads(b"0")
+
+    def test_invalid_reference(self):
+        import struct
+
+        for index in (0, 0xFFFFFFFF):
+            with self.assertRaises(ValueError):
+                marshal.loads(b"r" + struct.pack("<I", index))
+
+
 if __name__ == "__main__":
     unittest.main()

extra_tests/snippets/stdlib_select.py

@@ -1,4 +1,5 @@
 import select
+import signal
 import socket
 import sys
 
@@ -72,8 +73,78 @@ if sys.platform != "win32":
             max_fd = sock.fileno()
             max_fd_sock = sock
     assert_raises(ValueError, select.select, [max_fd_sock], [], [], 0)
+for sock in sockets:
+    sock.close()
 del sockets
 a, b = socket.socketpair()
 # CPython disallows this on *nix systems too.
 assert_raises(ValueError, select.select, [a] * TOO_MANY_SELECT_FDS, [], [], 0)
-del a, b
+a.close()
+b.close()
+
+
+# fileno() runs while the sequence is being read, and it can mutate the very
+# list it was handed.
+mutable_pair, other_end = socket.socketpair()
+
+
+class MutatesTheList:
+    def __init__(self, elements, fd):
+        self.elements = elements
+        self.fd = fd
+
+    def fileno(self):
+        self.elements.clear()
+        self.elements.append(self)
+        return self.fd
+
+
+elements = []
+elements.extend([MutatesTheList(elements, mutable_pair.fileno())] * 40)
+assert select.select(elements, [], [], 0) == ([], [], [])
+
+
+class GrowsTheList:
+    def __init__(self, elements, fd):
+        self.elements = elements
+        self.fd = fd
+
+    def fileno(self):
+        self.elements.append(self)
+        return self.fd
+
+
+# A list that grows by one on every fileno() never reaches its own end, so the
+# limit has to be answered during the walk rather than from the final length.
+elements = []
+elements.append(GrowsTheList(elements, mutable_pair.fileno()))
+assert_raises(ValueError, select.select, elements, [], [], 0)
+
+mutable_pair.close()
+other_end.close()
+
+# poll() waits with signal handlers able to run, and a handler may register on
+# the same poll object.
+if hasattr(select, "poll") and hasattr(signal, "setitimer"):
+    poller = select.poll()
+    idle, idle_peer = socket.socketpair()
+    poller.register(idle.fileno(), select.POLLIN)
+    handled = []
+
+    def register_from_handler(signum, frame):
+        poller.register(idle_peer.fileno(), select.POLLIN)
+        handled.append(signum)
+
+    previous = signal.signal(signal.SIGALRM, register_from_handler)
+    try:
+        signal.setitimer(signal.ITIMER_REAL, 0.05)
+        try:
+            poller.poll(1000)
+        except InterruptedError:
+            pass
+    finally:
+        signal.setitimer(signal.ITIMER_REAL, 0)
+        signal.signal(signal.SIGALRM, previous)
+    assert handled == [signal.SIGALRM], handled
+    idle.close()
+    idle_peer.close()

extra_tests/snippets/stdlib_socket.py

@@ -171,3 +171,18 @@ for exc, expected_name in [
 
 # assert socket.timeout.__module__ == "builtins"
 # assert socket.timeout.__name__ == "TimeoutError"
+
+
+# recv() sizes its buffer from the argument, so an unreachable size has to be
+# reported rather than reserved.
+sizes = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+with sizes:
+    for bufsize in (2**62, 2**48):
+        try:
+            sizes.recv(bufsize)
+        except (MemoryError, OSError, OverflowError):
+            # A size that does not fit the platform's C int is reported while
+            # converting the argument, before there is anything to reserve.
+            pass
+    with assert_raises(ValueError):
+        sizes.recvfrom(-1)

extra_tests/snippets/stdlib_struct.py

@@ -92,3 +92,67 @@ with assert_raises(UnicodeEncodeError):
 
 with assert_raises(struct.error):
     struct.Struct(b"\xff")
+
+
+# A value the format has no room for names the format and the range it holds.
+for fmt, message in (
+    ("B", "'B' format requires 0 <= number <= 255"),
+    ("b", "'b' format requires -128 <= number <= 127"),
+    (">H", "'H' format requires 0 <= number <= 65535"),
+    (">i", "'i' format requires -2147483648 <= number <= 2147483647"),
+    ("N", "'N' format requires 0 <= number <= 18446744073709551615"),
+    ("P", "int too large to convert"),
+):
+    try:
+        struct.pack(fmt, 10**30)
+    except struct.error as e:
+        assert str(e) == message, (fmt, str(e))
+    else:
+        raise AssertionError(f"expected struct.error for {fmt!r}")
+
+try:
+    struct.pack("B", "x")
+except struct.error as e:
+    assert str(e) == "required argument is not an integer", e
+else:
+    raise AssertionError("expected struct.error")
+
+
+# __init__ reads a new format into a Struct that already holds one.
+s = struct.Struct(">h")
+s.__init__(">hh")
+assert s.format == ">hh"
+assert s.size == 4
+assert s.pack(1, 2) == b"\x00\x01\x00\x02"
+assert s.unpack(b"\x00\x01\x00\x02") == (1, 2)
+
+# A format that cannot be read leaves the Struct as it was.
+for bad in ("\udc00", "$"):
+    with assert_raises((UnicodeEncodeError, struct.error)):
+        s.__init__(bad)
+    assert s.format == ">hh"
+    assert s.pack(1, 2) == b"\x00\x01\x00\x02"
+
+
+# A subclass may do its own __init__ and pass the format up.
+class BigShort(struct.Struct):
+    def __init__(self):
+        super().__init__(">h")
+
+
+assert BigShort().pack(12345) == b"\x30\x39"
+
+# Until __init__ runs there is no format to answer with.
+blank = struct.Struct.__new__(struct.Struct)
+assert blank.size == -1
+for call in (
+    lambda: blank.format,
+    lambda: blank.pack(1),
+    lambda: blank.unpack(b"aa"),
+    lambda: blank.unpack_from(b"aaaa"),
+    lambda: blank.pack_into(bytearray(4), 0, 1),
+    lambda: blank.iter_unpack(b"aa"),
+    lambda: repr(blank),
+):
+    with assert_raises(RuntimeError):
+        call()

extra_tests/snippets/stdlib_threading_current_frames.py

@@ -0,0 +1,100 @@
+"""Take sys._current_frames() while other threads are running Python.
+
+The frame each thread is executing is published for cross-thread readers, and
+_current_frames() takes a reference to it with the world stopped. A reader that
+disagrees with the publisher about what the published pointer addresses reads
+and reference-counts the wrong memory, which corrupts a neighbouring object
+rather than failing at the read: the damage surfaces later, in the thread that
+owns it, as a crash or a wedge.
+
+Workers therefore run ordinary Python calls (which publish a frame) in a tight
+loop while the main thread hammers _current_frames().
+"""
+
+import sys
+import threading
+import time
+
+DURATION = 1.5
+
+
+def leaf():
+    return sum(range(8))
+
+
+def nest(n):
+    if n:
+        return nest(n - 1)
+    return leaf()
+
+
+def worker(stop):
+    while not stop.is_set():
+        nest(16)
+
+
+def frames_are_sane(frames):
+    # Every key is a thread id, every value a frame of this process.
+    for tid, frame in frames.items():
+        assert isinstance(tid, int), tid
+        assert tid > 0, tid
+        assert type(frame).__name__ == "frame", frame
+        assert isinstance(frame.f_lineno, int), frame
+        assert isinstance(frame.f_code.co_name, str), frame
+
+
+# The main thread sees itself where it stands.
+me = sys._current_frames()[threading.get_ident()]
+assert me is sys._getframe(), me
+
+stop = threading.Event()
+threads = [threading.Thread(target=worker, args=(stop,)) for _ in range(4)]
+for t in threads:
+    t.start()
+
+deadline = time.time() + DURATION
+calls = 0
+while time.time() < deadline:
+    frames_are_sane(sys._current_frames())
+    calls += 1
+stop.set()
+for t in threads:
+    t.join()
+
+assert calls > 0, calls
+
+
+# A thread parked in a call the main thread can name is reported inside it,
+# with its callers reachable through f_back.
+entered = threading.Event()
+leave = threading.Event()
+seen = []
+
+
+def g456():
+    seen.append(threading.get_ident())
+    entered.set()
+    leave.wait()
+
+
+def f123():
+    g456()
+
+
+t = threading.Thread(target=f123)
+t.start()
+entered.wait()
+try:
+    chain = []
+    frame = sys._current_frames()[seen[0]]
+    while frame is not None:
+        chain.append(frame.f_code.co_name)
+        frame = frame.f_back
+    assert "g456" in chain, chain
+    assert "f123" in chain, chain
+    assert chain.index("g456") < chain.index("f123"), chain
+finally:
+    leave.set()
+    t.join()
+
+print("ok")

extra_tests/snippets/stdlib_types.py

@@ -47,14 +47,14 @@ if sys.implementation.name == "rustpython":
         list[self_referential]
 
     nested = [0]
-    for _ in range(sys.getrecursionlimit() * 2):
+    for _ in range(100_000):
         nested = [nested]
     with assert_raises(RecursionError):
         list[nested]
 
     # hashing an alias walks the same shape
     deep_alias = int
-    for _ in range(sys.getrecursionlimit() * 2):
+    for _ in range(100_000):
         deep_alias = list[deep_alias]
     with assert_raises(RecursionError):
         hash(deep_alias)

extra_tests/snippets/stdlib_typing.py

@@ -45,3 +45,21 @@ class ClassWithUnionParams:
 assert _typing._idfunc(1) == 1
 with assert_raises(TypeError):
     _typing._idfunc()
+
+
+# ParamSpecArgs shows a non-ParamSpec origin by its repr, which is where the
+# recursion guard lives; nesting them deeply must not walk the native stack.
+
+from typing import ParamSpec, ParamSpecArgs
+
+spec = ParamSpec("spec")
+assert repr(spec.args) == "spec.args"
+assert repr(spec.kwargs) == "spec.kwargs"
+
+nested = object()
+for _ in range(2000):
+    nested = ParamSpecArgs(nested)
+try:
+    repr(nested)
+except RecursionError:
+    pass