Official PyPy support?

[rassie]

What's the current official stance on PyPy support? There seems to be some interest there, judging by some closed PRs, defunct TravisCI configuration and some bugfixes, but still PyPy is not part of testing and benchmarking matrix -- would it make sense to add PyPy there?

[stefan6419846]

What's the current official stance on PyPy support?

I do not know if it works or not. There have been no special attempts to look into this.

There seems to be some interest there, judging by some closed PRs, defunct TravisCI configuration and some bugfixes

I am not aware of this or any specific requests/support. Travis is surely not used for testing for multiple years now.

PyPy is not part of testing and benchmarking matrix

Benchmarking is only done on the latest CPython anyway. For the remaining parts: How different is PyPy from CPython? If we start running against PyPy as well, do we need to check all other Python variants as well?

[rassie]

Thanks for the quick reply.

I feel like I should give a bit of context: we are doing low-level content stream extraction and modification (mostly around spot colors) and have some massive files with operation count in the double-digit millions. I've done some simple benchmarking and it seems that whatever we are doing profits massively from running under PyPy, ranging from 3x to 50x speedup, especially for big files, especially in long-running processes. Is seems that pypdf being pure Python helps here a lot.

My question was mainly about whether PyPy is of general interest, I have no problem running my code on PyPy without it being officially supported. I have no idea about other Python variants. But if the gains are real (which is where benchmarking comes in), it might be of general interest as well.

[stefan6419846]

You are of course invited to propose a PR adding PyPy to our regular testing, similar to the Windows and macOS tests running only the latest PyPy release. If all tests pass fine as-is, merging it should not be an issue and you might want to provide a follow-up PR to add a suitable declaration/documentation if this helps. If there are issues with our current tests failing, we have to investigate and evaluate how much work fixing this might be involved.

[rassie]

Fair enough, I'll give it a try. Thanks!