v/TESTS.md

Automated tests

TLDR: do run v test-all locally, after making your changes, and before submitting PRs.

Tip: use v -cc tcc when compiling tests, because TCC is much faster, compared to most other C compilers like clang/gcc/msvc. Most test commands will use the V compiler and the V tools many times, potentially hundreds/thousands of times.

v test-all

Test and build everything. Useful to verify locally, that the CI will most likely pass. Slowest, but most comprehensive.

It works, by running these in succession:

• v test-cleancode
• v test-self
• v test-fmt
• v build-tools
• v build-examples
• v check-md -hide-warnings .
• v install nedpals.args

The self-build checks include serial transformation and C generation with -no-parallel. Transformation always runs in the current compiler; the historical -new-transformer switch is no longer needed.

Details:

In the v repo there are many tests. The main types are:

_test.v tests - these are the normal V test files.

All test_ functions in these files, will be ran automatically by V's test framework.

NB 1: You can run test files one by one, with: v file_test.v - this will run the test_ functions in file_test.v, and will exit with a 0 exit code, if they all had 0 failing assertions.

v -stats file_test.v - this will run the test_ functions, and show a report about how much time it took to run each of them too.

Implicit test executables and their macOS debug symbols are removed after the test process exits, including failed assertions, panics, and nonzero exits. This also applies with -stats, -keepc, and v run file_test.v; -keepc still retains generated C. An explicit -o keeps the requested output for later execution or debugging.

NB 2: You can also run many test files at once (in parallel, depending on how many cores you have), with: v test folder - this will run all _test.v files in folder, recursively.

v -stats test folder - same, but will also produce timing reports about how fast each test_ function in each _test.v file ran.

By default, v test uses at most four parallel workers and budgets one worker per 8 GiB of memory. On Linux it uses the lower of physical memory and the active cgroup memory limit. Set VJOBS to a positive value to explicitly choose a different worker count when your test workload and machine capacity are known.

Skipped test paths are resolved before comparison, so selecting a file through a symlink does not bypass its platform or architecture exclusion.

On Linux, -race test builds use _setjmp and _longjmp to continue after failed assertions, avoiding ThreadSanitizer's incompatible longjmp interceptor on glibc 2.44. Later tests and race detection still run. A later race report can retain frames from the failed test because the sanitizer does not observe that jump.

Tests ending in _windows_test.v or _windows_test.c.v run only on Windows. The compound _android_outside_termux_test.v and _android_outside_termux_test.c.v suffixes select Android outside Termux; _termux_test.v and _termux_test.c.v select Termux.

v test vlib/v/tests:

This folder contains _test.v files, testing the different features of the V compiler. Each of them will be compiled, and all the features in them have to work (verified by assertions).

v vlib/v/slow_tests/inout/compiler_test.v

This is a test runner, that checks whether the output of running a V program, matches an expected .out file. You can also check for code that does panic using this test runner - just paste the start of the panic() output in the corresponding .out file.

Note

These tests, expect to find a pair of .vv and .out files, in the folder: vlib/v/slow_tests/inout

The test runner will run each .vv file, and will check that its output, matches the contents of the .out file with the same base name. This is particularly useful for checking that errors and panics are printed.

v test vlib/v/gen/c/

The C backend has focused unit and integration tests beside its implementation. Many tests compile a small V source to C and assert on the generated declarations, expressions, ABI, linker inputs, or runtime behavior.

v test cmd/v/

The compiler process regressions workflow runs every launcher test in cmd/v/ on Linux, macOS, and Windows with the default compiler. This covers argument routing, fallback messages, executable discovery, and tool-cache behavior, including platform-specific code.

Line coverage

Collect coverage with v -coverage coverage_dir path/to/file_test.v, then inspect it with v cover coverage_dir. Add -no-skip-unused when compiling to include uncalled functions in the report as well as executed code.

On Windows, test statistics and coverage use a 64-bit monotonic clock resolved at runtime, so they also work with bundled TCC versions whose import libraries omit GetTickCount64.

REPL tests

The test runner for these is vlib/v/slow_tests/repl/repl_test.v.

The test cases for the V REPL, are stored in .repl files, in the folder vlib/v/slow_tests/repl/. Each .repl file in this folder, contains several lines of input to the V repl, followed by a single line of ===output===, and then the output lines, that the repl would normally show for the input lines.

If you change the compiler or the REPL source, and you have breaks in those .repl files, you can replace the current output in them, by running several times VAUTOFIX=1 ./vlib/v/slow_tests/repl/repl_test.v, until all the files are fixed and the test pass.

v vlib/v/slow_tests/run_project_folders_test.v

This test runner, checks whether whole project folders, can be compiled, and run.

Note

Each project in these folders, should finish with an exit code of 0, and it should output OK as its last stdout line.

v vlib/v/tests/known_errors/known_errors_test.v

This test runner, checks whether a known program, that was expected to compile, but did NOT, due to a buggy checker, parser or cgen, continues to fail. The negative programs are collected in the vlib/v/tests/known_errors/testdata/ folder. Each of them should FAIL to compile, due to a known/confirmed compiler bug/limitation.

The intended use of this, is for providing samples, that currently do NOT compile, but that a future compiler improvement WILL be able to compile, and to track, whether they were not fixed incidentally, due to an unrelated change/improvement. For example, code that triggers generating invalid C code can go here, and later when a bug is fixed, can be moved to a proper _test.v or .vv/.out pair, outside of the vlib/v/tests/known_errors/testdata/ folder.

Test building of actual V programs (examples, tools, V itself)

• v build-tools
• v build-examples
• v build-vbinaries

v build-examples discovers files with a main, no_main, or implicit main module, and builds configured projects as folders. Leading comments, attributes, and directives do not affect module discovery. Library modules are checked when the programs that import them compile.

Formatting tests

cmd/tools/vfmt_test.v checks the formatter command, while vlib/v/gen/v/gen_test.v and the other tests in vlib/v/gen/v/ check flat-AST-to-V formatting and round trips.

• v test-cleancode

Check that most .v files, are invariant of v fmt runs.

• v test-fmt

This tests that all .v files in the current folder are already formatted. It is useful for adding to CI jobs, to guarantee, that future contributions will keep the existing source nice and clean.

Markdown/documentation checks:

• v check-md -hide-warnings .

Ensure that all .md files in the project are formatted properly, and that the V code block examples in them can be compiled/formatted too.

Note: if that command finds formatting errors, they can be fixed with: VAUTOFIX=1 ./v check-md file.md or with v check-md -fix file.md.

v test-self

Run vlib module tests, including the compiler tests. Test discovery includes architecture-suffixed files such as _test.amd64.v when the suffix matches the host architecture; files for other architectures are excluded.

To run the same suite across separate machines, set VTEST_SELF_SHARD_COUNT to the number of machines and set VTEST_SELF_SHARD_INDEX to a different zero-based index on each one. For example, VTEST_SELF_SHARD_COUNT=5 VTEST_SELF_SHARD_INDEX=0 ./v test-self vlib runs the first shard. Every test file belongs to exactly one shard. Leave both variables unset for the full suite.

v vlib/v/compiler_errors_test.v

This runs tests for:

• vlib/v/scanner/tests/*.vv
• vlib/v/checker/tests/*.vv
• vlib/v/parser/tests/*.vv

Some fixtures require the V 0.5.2 compatibility compiler: legacy compiler-module diagnostics, legacy JSON code-generation errors, and the with_check_option cases. The current driver selects that compiler for those fixtures; other fixtures start with the current compiler and may use its normal compatibility fallback. The suite remains runnable on master even though the legacy compiler sources were removed from this tree.

Prepare the compatibility compiler from the repository root before running the complete suite:

make v1
./v -silent vlib/v/compiler_errors_test.v

make v1 installs the pinned V 0.5.2 release and its matching library in a cache, or builds that release with oldv when a usable release binary is unavailable. Initial installation needs network access and the installer tools; source builds also need Git and a C compiler. On Windows, run this from MSYS2 with GNU make (make or mingw32-make) available. Building the ordinary compiler with makev.bat alone does not provision this compatibility compiler.

The driver can run make v1 automatically when needed. If it reports that no usable fallback was found and make is unavailable, install GNU make and run the preparation command above. This is a missing test prerequisite, and also affects the complete v test vlib/v/ run, which includes this suite. After installation, a usable cached compatibility compiler can run without make. Successful automatic installation keeps progress output out of exact-output fixture diagnostics. If installation fails, the driver prints the installer output to explain the missing prerequisite.

Use -new-compiler when investigating an individual fixture with the current compiler. It overrides explicit legacy fixture selection, so its diagnostics need not match that fixture's existing .out file. Disabling automatic fallback with V_MACOS_V3_NO_FALLBACK=1 still permits explicit legacy fixture selection and does not remove the suite's compatibility-compiler prerequisite.

Note

There are special folders, that compiler_errors_test.v will try to run/compile with specific options:

vlib/v/checker/tests/globals_run/ - -enable-globals run; results stored in .run.out files, matching the .vv ones.

NB 2: in case you need to modify many .out files, run twice in a row: VAUTOFIX=1 ./v vlib/v/compiler_errors_test.v This will fail the first time, but it will record the new output for each .vv file, and store it into the corresponding .out file. The second run should be now successful, and so you can inspect the difference, and commit the new .out files with minimum manual effort.

NB 3: To run only some of the tests, use: VTEST_ONLY=mismatch ./v vlib/v/compiler_errors_test.v This will check only the .vv files, whose paths match the given filter.

NB 4: To run tests, but without printing status lines for all the successful ones, use: VTEST_HIDE_OK=1 ./v test vlib/math/ This will print only the total stats, and the failing tests, but otherwise it will be silent. It is useful, when you have hundreds or thousands of individual _test.v files, and you want to avoid scrolling.

NB 5: To show only the currently running test, use: ./v -progress test vlib/math/ In this mode, the output lines will be limited, no matter how many _test.v files there are. The output will contain the total stats and the output of the failing tests too.

NB 6: Set VTEST_SKIP_OWNERSHIP=1 to omit ownership and autofree tests from v test, v test-self, and vlib/v/test_all.vsh. GitHub Actions enables this behavior automatically while ownership/autofree coverage is disabled there. A _test.v file is an ownership/autofree test when it is in an ownership folder, its name contains ownership or autofree, its // vtest vflags: line has -ownership, -autofree or -d ownership, or it has this line:

// vtest ownership

Add that line to a test that builds the ownership compiler, or compiles with one of those flags (for example with os.exec([vexe, '-ownership', ...])). Flags elsewhere in a test's source are not enough, since many tests pass them only to check how they are parsed or rejected.

.github/workflows/ci.yml

This is a Github Actions configuration file, that runs various CI tests in the main V repository, for example:

• v vet vlib/v - run a style checker.
• v test-self (run self tests) in various compilation modes.

Note

The VDOC test vdoc_file_test.v now also supports VAUTOFIX, which is useful, if you change anything inside cmd/tools/vdoc, or inside the modules that it depends on (like markdown). After such changes, just run this command 2 times, and commit the resulting changes in cmd/tools/vdoc/testdata as well: VAUTOFIX=1 ./v cmd/tools/vdoc/vdoc_file_test.v