mirror of
https://github.com/nodejs/node.git
synced 2026-10-28 18:25:13 +08:00
Signed-off-by: Matteo Collina <hello@matteocollina.com> PR-URL: https://github.com/nodejs/node/pull/63973 Refs: https://github.com/nodejs/node/pull/63570 Refs: https://github.com/nodejs/undici/issues/5002 Reviewed-By: Jithil P Ponnan <jithil@outlook.com> Reviewed-By: James M Snell <jasnell@gmail.com>
106 lines
2.3 KiB
JavaScript
106 lines
2.3 KiB
JavaScript
'use strict';
|
|
require('../common');
|
|
const { Readable } = require('stream');
|
|
const assert = require('assert');
|
|
|
|
{
|
|
// Call .setEncoding() while there are bytes already in the buffer.
|
|
const r = new Readable({ read() {} });
|
|
|
|
r.push(Buffer.from('a'));
|
|
r.push(Buffer.from('b'));
|
|
|
|
r.setEncoding('utf8');
|
|
const chunks = [];
|
|
r.on('data', (chunk) => chunks.push(chunk));
|
|
|
|
process.nextTick(() => {
|
|
assert.deepStrictEqual(chunks, ['ab']);
|
|
});
|
|
}
|
|
|
|
{
|
|
// Call .setEncoding() while the buffer contains a complete,
|
|
// but chunked character.
|
|
const r = new Readable({ read() {} });
|
|
|
|
r.push(Buffer.from([0xf0]));
|
|
r.push(Buffer.from([0x9f]));
|
|
r.push(Buffer.from([0x8e]));
|
|
r.push(Buffer.from([0x89]));
|
|
|
|
r.setEncoding('utf8');
|
|
const chunks = [];
|
|
r.on('data', (chunk) => chunks.push(chunk));
|
|
|
|
process.nextTick(() => {
|
|
assert.deepStrictEqual(chunks, ['🎉']);
|
|
});
|
|
}
|
|
|
|
{
|
|
// Call .setEncoding() while the buffer contains an incomplete character,
|
|
// and finish the character later.
|
|
const r = new Readable({ read() {} });
|
|
|
|
r.push(Buffer.from([0xf0]));
|
|
r.push(Buffer.from([0x9f]));
|
|
|
|
r.setEncoding('utf8');
|
|
|
|
r.push(Buffer.from([0x8e]));
|
|
r.push(Buffer.from([0x89]));
|
|
|
|
const chunks = [];
|
|
r.on('data', (chunk) => chunks.push(chunk));
|
|
|
|
process.nextTick(() => {
|
|
assert.deepStrictEqual(chunks, ['🎉']);
|
|
});
|
|
}
|
|
|
|
{
|
|
// Call .setEncoding() before the stream starts flowing, after _read()
|
|
// has buffered the first chunk of a split character.
|
|
const chunks = [
|
|
Buffer.from([0xf0, 0x9f]),
|
|
Buffer.from([0x8e, 0x89]),
|
|
null,
|
|
];
|
|
const r = new Readable({
|
|
read() {
|
|
this.push(chunks.shift());
|
|
},
|
|
});
|
|
|
|
r.read(0);
|
|
assert.strictEqual(r.readableFlowing, null);
|
|
assert.strictEqual(r.readableLength, 2);
|
|
|
|
r.setEncoding('utf8');
|
|
const received = [];
|
|
r.on('data', (chunk) => received.push(chunk));
|
|
|
|
process.nextTick(() => {
|
|
assert.deepStrictEqual(received, ['🎉']);
|
|
});
|
|
}
|
|
|
|
{
|
|
// Call .setEncoding() after EOF while the buffer contains an
|
|
// incomplete character.
|
|
const r = new Readable({ read() {} });
|
|
|
|
r.push(Buffer.from([0xf0]));
|
|
r.push(Buffer.from([0x9f]));
|
|
r.push(null);
|
|
|
|
r.setEncoding('utf8');
|
|
const chunks = [];
|
|
r.on('data', (chunk) => chunks.push(chunk));
|
|
|
|
process.nextTick(() => {
|
|
assert.deepStrictEqual(chunks, ['�']);
|
|
});
|
|
}
|