Repository navigation
[ php-wasm ] Terminate the non-CLI additional functions list - #4524
Merged
Merged
Conversation
Contributor
|
Caution Review failedThe pull request is closed. Finishing Touches
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Collaborator
Author
|
Rebuilding the whole set of php.wasm binaries won't help here, since they already use |
mho22
marked this pull request as ready for review
October 27, 2026 13:11
mho22
added a commit
that referenced
this pull request
Nov 5, 2026
…ing reads on slow pipes (#3511) ## Motivation for the change, related issues `js_fd_read` in `php_wasm.c` has a hardcoded 5-second polling timeout that causes blocking `fread()` calls to silently return empty data when the pipe takes longer than 5 seconds to produce output. This breaks any use case where PHP reads from a slow process via `proc_open`. The timeout was introduced in #931 as a safety net against hanging reads, then carried over into `php_wasm.c` in #1867. ## Why the timeout is no longer necessary #931 was the first time `fread()` on a blocking pipe polled for data instead of failing with `EWOULDBLOCK`. At that point a child process that died or never started could leave the pipe open, and the read would poll forever with no visible feedback. PHP's own `max_execution_time` does not help here: it is `0` in CLI mode, and the timer does not fire while the read awaits in JavaScript. The 5-second cap made that hang visible, notably when running the PHPUnit test suite. Every situation the cap guarded against now ends the read on its own: - When the child process exits, the `exit` handler closes the child's end of the pipes, and the `pipe.refcnt < 2` check returns. - When the child process emits `error` without `exit`, the `error` handler closes the child's end of the pipes. - When the spawn handler never signals `spawn`, `js_open_process` stops waiting after 5 seconds and closes the child's end of the pipes. - When Node cannot start the process at all, the child has no pid. `proc_open()` returns `false` with the spawn error, so there is no pipe to read. - When PHP closes the stream, the `FS.isClosed` check returns. - When the pipe is non-blocking, `js_fd_read` returns `EWOULDBLOCK` immediately without polling. The one remaining case is a child process that is alive, never writes, and never exits. Native PHP blocks forever on that read too, so this matches the POSIX behaviour. The read awaits in JavaScript, so the event loop keeps running and the host can still kill the child, which closes the pipe and unblocks the read. A read that silently gives up after 5 seconds is worse than a read that blocks, because the caller cannot tell truncated data from a completed read. ## Implementation details - Remove the retry counter and `maxRetries` check from the `js_fd_read` polling loop. - Move the cleanup of the child's pipe ends from the `exit` handler into `closeChildPipes()`. It runs on `exit`, on `error`, and when the spawn wait fails. - Return the negative errno from `js_open_process`, and set `errno` from it in `proc_open.c`. Before, `js_open_process` called `___errno_location(code)`. That export takes no argument, and as a JSPI export it returns a Promise, so `errno` stayed `0` and `proc_open()` warned `proc_open() failed - Success`. It now warns `proc_open() failed - No such file or directory` for a missing `cwd`. - Copy `config.guess` and `config.sub` from the base image's automake in `Dockerfile-5-2` instead of downloading them from savannah, so the PHP 5.2 build does not depend on that server. - Update the PHP 8.5 wasm path in the `php-wasm/web` README from `8_5_10` to `8_5_12`. - Rebuild every PHP.wasm binary on top of trunk: PHP 5.2 and 7.4 to 8.5, for web and Node, in JSPI and Asyncify. The rebuild also ships the `php_wasm.c` fix from #4524, which trunk merged without a rebuild. ## Breaking change A blocking `fread()` on a pipe from a child process that never writes and never exits now waits until the child exits or the host closes the pipe, instead of returning empty data after 5 seconds. ## Testing Instructions (or ideally a Blueprint) CI The `proc_open()` tests in `php-part-1.spec.ts` cover each exit condition: - `blocking fread waits for slow pipe output`: the child writes after 6 seconds, and `fread()` receives the data. - `blocking fread returns when the spawn handler never signals spawn`: `fread()` returns empty after the 5-second spawn wait. - `blocking fread returns when the process errors without exiting`: `fread()` returns empty after the `error` event. - `reports the spawn error when the process never starts`: `proc_open()` returns `false` and warns `No such file or directory`. `Handle process spawn timeout gracefully` is removed. Its spawn handler, built with `createSpawnHandler()`, emits `spawn` as soon as its callback returns a promise, so the test covered a live child that never writes. That case now blocks like native PHP. ### Verification against the WordPress PHPUnit suite Since #931 was built to run PHPUnit, I ran the WordPress core test suite from `wordpress-develop` through the Playground CLI `php` subcommand (PHP 8.3, WordPress nightly) on both trunk and this branch. The results are identical: `Tests_Option_Option` passes 85 tests with 3 skipped, and the `Tests_Functions_*` classes run 1028 tests with the same 42 failures on both builds. Those failures come from environment differences and from PHPUnit process isolation, not from this change. In the same CLI, a `proc_open` child that writes after 7 seconds is read correctly on this branch, while trunk returns an empty string after 5 seconds. --------- Co-authored-by: Brandon Payton <brandon@happycode.net> Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A small one, and honestly a low-value one: this only matters if you build PHP.wasm with
WITH_CLI_SAPI=no, which no committed binary does. But that option is offered bybuild.js, and right now it produces a PHP that can crash at startup. So it is broken, even if nobody uses it.What is wrong
php_wasm.chas two versions ofadditional_functions[]. The CLI version ends with the usual{NULL, NULL, NULL}entry. The non-CLI version, compiled whenWITH_CLI_SAPI=no, does not.zend_register_functions()reads entries until it finds aNULLname, so without that last entry it keeps reading the memory after the array.How I found it
While I tried a wasm64 build of PHP 8.5 to answer the question in #4499, I used a reduced build with
WITH_CLI_SAPI=no. It crashed at startup instrlen()insidezend_register_functions(). With this line added, the same build starts and runsZend/bench.php. I did not check what wasm32 does without the fix: it depends on what sits in memory after the array.What this changes
Nothing for current users. Every committed binary uses
WITH_CLI_SAPI=yes, so no binary is rebuilt here. It only makesWITH_CLI_SAPI=nousable for whoever needs it.