Installation on Alpine and php 8.2+

[Kabingeran]

Hi! Does not install on Alpine with php 8.2+ via pecl. Has anyone tried to install it?

[andrewdalpino]

Hi @Kabingeran I don't think it compiles in Alpine. We had an Alpine environment set up in our CI pipeline and it failed to compile. Since we've shifted our focus away from Tensor and to contributing to NumPower (https://github.com/RubixML/numpower) I haven't had much of a reason to look into the issue further than that.

Here's the GitHub Actions logs for the last time we tried to run the compile workflow on Alpine with PHP 8.2

https://github.com/RubixML/Tensor/actions/runs/14185339324/job/40287690335

Is there something that stands out to you in those logs? I'd be happy to work with you to fix it.

[Kabingeran]

Please add Alpine. Otherwise, your package will not make it into most builds. It looks like you are passing the wrong argument type?

[andrewdalpino]

/tmp/pear/temp/tensor/ext/kernel/require.c is part of Zephir, a third-party dependency we use to compile to a PHP extension. Perhaps this is an issue for them?

https://github.com/zephir-lang/zephir

The extension compiles fine on Ubuntu with PHP 8.2 and 8.3, not sure why you changed the title of this issue.

https://github.com/RubixML/Tensor/actions/runs/14185879466

[Kabingeran]

Why you use Zephir? This is an unnecessary dependence that brings problems. If this bug is not your responsibility, then who will work on it?

[rgr]

Hi,

I’m trying to install Tensor 3.0.7 via PECL on Alpine Linux with PHP 8.3 (Docker image: php:8.3-fpm-alpine or dunglas/frankenphp:php8.3-alpine).

The build fails with the following error:

error: passing argument 1 of 'zval_ptr_dtor' from incompatible pointer type [-Wincompatible-pointer-types]
zval_ptr_dtor(zend_string_path);
             ^~~~~~~~~~~~~~~~
note: expected 'zval *' but argument is of type 'zend_string *'

This happens in kernel/require.c (and possibly other files) because the extension tries to call zval_ptr_dtor on a zend_string* rather than a zval*.
This change in PHP’s internal API seems to have started in PHP 8.1+.
Despite the PECL page stating that 3.0.7 supports up to PHP 8.3, the current codebase is not compatible.

Environment:
• PHP 8.3.0 (Alpine Linux)
• PECL tensor 3.0.7

Suggested next steps:
• Review and update the codebase for compatibility with PHP 8.3 internal API changes (especially type handling in memory management functions).
• Adjust the zval_ptr_dtor calls, and test across PHP 8.0, 8.1, 8.2, and 8.3.

Thanks!

[andrewdalpino]

Why you use Zephir? This is an unnecessary dependence that brings problems. If this bug is not your responsibility, then who will work on it?

Zephir provides the PHP extension framework. Not only is it incorrect to say that it's not necessary it's actually critical to the core functionality of the extension.

[Kabingeran]

@andrewdalpino Use the C language without any dependencies.

[andrewdalpino]

I used Zephir because using pure C would have taken alot longer to develop and would have created more of a burden to maintain. For most of the life of Tensor, I have been the sole maintainer in addition to the rest of the Rubix ecosystem. In a world where the PHP community would have put more emphasis on the development of their ML ecosystem through funding and contributions, this may have been possible - but unfortunately, that is not the world we live in so Zephir was the best choice.

[Kabingeran]

@andrewdalpino Use free code generation, your packages or AI if this is critical.

[sheunl]

@andrewdalpino I could take this on, if you want. Would be great to contribute to the ecosystem.

[andrewdalpino]

Hi @sheunl thanks for the interest! How many years of PHP extension and/or internals development do you have under your belt?

We've shifted our focus away from Tensor and toward NumPower which solves many of the inherent performance issues with Tensor and adds things like NDArray and GPU support.

https://github.com/RubixML/numpower

Tensor was meant to be a stop-gap or "bridge" to something better when the ecosystem caught up. I didn't anticipate that it would take this long for PHP to catch up and so Tensor, as a result, has taken on more responsibility than it was ever intended.

We could use some support for NumPower - but if you feel like you'd rather contribute to Tensor and you could commit to the necessary time to maintain it as PHP internals evolve - then I would be open to talking about it.

Thanks, take care!

P.S. Feel free to join our Telegram channel https://t.me/RubixML