[PHP.wasm] Upgrade emscripten version from 4.0.5 to 4.0.19

[mho22]

Motivation for the change

We need this pull request in order for our PHP.wasm build to be built without RELOCATABLE and avoid size increases when building with MAIN_MODULE. The pull request is in 4.0.19 release.

Implementation details

• Modified emscripten version in base-image Dockerfile.
• Removed setErrNo and malloc from EXPORT_RUNTIME_METHODS
• Replaced setErrNo by __errno_location in phpwasm-emscripten-library
• Replaced PHPLoader["free"] with _free in phpwasm-emscripten-library
• Bound wasmExports['free'] to PHPLoader['free'] in php/Dockerfile
• Bound wasmExports['malloc'] to PHPLoader['malloc'] in php/Dockerfile
• Removed fd_close from JSPI_IMPORTS and JSPI_EXPORTS due to SuspendError: trying to suspend JS frames
• Ignored default.profraw file in .gitignore
• Added __funcs_on_exit function in JSPI_EXPORTS to solve issues from PHP7.4 and PHP7.3
• Added __funcs_on_exit function in EXPORTED_FUNCTIONS to solve issues from PHP7.4 and PHP7.3
• Modified patches related to signature mismatch in zend_list_free_wrapper in PHP7.4 and PHP7.3
• Added HEAPU32 and HEAPU8 in EXPORT_RUNTIME_METHODS
• Refactored php-crash.spec.ts file to run test on each supported php version
• Added _malloc function in EXPORTED_FUNCTIONS
• Rearranged first test in php-crash.spec.ts

✅ make all

✅ npm run recompile:php:node:jspi:all
✅ nx run php-wasm-compile:xdebug:jspi:all
✅ nx run php-wasm-compile-shared:intl:jspi:all 

✅ npm run recompile:php:node:asyncify:all
✅ nx run php-wasm-compile:xdebug:asyncify:all
✅ nx run php-wasm-compile-shared:intl:asyncify:all

✅ npm run recompile:php:web:jspi:all

✅ npm run recompile:php:web:asyncify:all

Testing Instructions

CI

[adamziel]

The remaining failure seems unrelated to this PR 👍 Let's rebuild all the wasm files to make sure all the tests pass and we're good! I love how much smaller the built binaries are becoming.

[mho22]

@adamziel Me too! I am currently rebuilding PHP.wasm node Asyncify as the 5th step in the description list. This will unlock the Intl dynamic extension for Web PR.

I will create a new pull request to try to correct the failed test as I think the current failure could be related to PHP not running exit() in the afterEach step in the blueprints tests. I couldn't reproduce the failure locally so this is difficult to diagnose but I remember doing this for the different test-unit-asyncify some months ago.

[mho22]

@adamziel The Does not crash due to an unhandled Asyncify error in php-crash.spec.ts doesn't return an unreachable Error anymore. It only hangs indefinitely. In order to make the tests pass with this upgrade I commented out the test and I suggest creating a follow-up pull request if you consider that a test like that is mandatory. What do you think?

[adamziel]

@mho22 following up sounds good, but I'm suspicious about it not failing anymore. Are we adding something to the asyncify list in this PR? Why is it not failing anymore?

[mho22]

@adamziel I will investigate. I didn't add anything to the ASYNCIFY list but I think there could be two reasons :

1. The new EXPORTED_RUNTIME_METHODS [ HEAPU32 and HEAPU8 ] and new EXPORTED_FUNCTIONS [ ___errno_location ]. These functions could have been related to that. But I doubt it.

2. The multiple improvements between emscripten 4.0.5 and 4.0.19.

I suspect the second option.

[mho22]

In order to keep consistency between every PHP version I will need to re-run the compilation of every PHPs... Since a new version of PHP 8.4 has been released and a minor change has been added in Dockerfile for PHP Web ASYNCIFY...

I hope we won’t need to upgrade Emscripten again for a long time after this pull request 😅

[mho22]

@adamziel That's it! It is now ready for review!

[mho22]

I will convert this to draft for now by adding the failing crash test again. It works for versions 8.4 8.3 and 8.2 so I guess it times out even if it returns the unreachable errors.

[mho22]

@adamziel I was wrong earlier, nothing was broken, the code had to be rearranged! Let's wait the check results. Ready for review again!

[adamziel]

🎉 🎉 🎉

[adamziel]

Node.js WASM binaries are significantly smaller, I understand that's because we've extracted the INTL extension:

However, Web Binaries are a little bit larger:

Do we understand why? Let's not block the intl work on that, but it would be great to investigate as a follow-up. Maybe we have to live with this now, but maybe there's a way to remove 6MB or so?

[mho22]

@adamziel Ow! This pull request still had Intl supported as a static extension. The Intl dynamic extension in web pull request should have done the 6Mb decrease. However it didn't... I don't know why. I am on it.

[mho22]

I am mixing things here since I am on the Intl for web pull request :

Before this pull request we had :

1. PHP.wasm Node as a MAIN_MODULE
2. PHP.wasm Web without MAIN_MODULE

Because of MAIN_MODULE, the node version had that 8Mb spike due to the magic_php_database file not sliced during build [ which it did without MAIN_MODULE ].

This PR helped remove that 8MB increase when using MAIN_MODULE. But however added 2-3 more MB. This means why Node lost 6Mb while Web became 3Mb heavier.

I will of course still investigate to find ways to reduce those sizes. [ MAIN_MODULE=2 is one of those probable ways ].  

[adamziel]

Good digging, thank you @mho22!