On Sun, 1 Feb 2026 at 19:57, Oscar Benjamin <[email protected]> wrote: > > In case anyone is interested in looking these are the PRs that are > currently inactive for about 2 months: > https://github.com/sympy/sympy/pulls?page=5&q=is%3Apr+is%3Aopen+sort%3Aupdated
I went through these 25 PRs and summarised them. One thing to bear in mind is that these are mostly from around the time that I started this thread about AI generated pull requests and obviously predate having any AI policy. Also there would have been many more pull requests at the time but what we are looking at here is the PRs that did not get closed or merged. Out of all time 93% of PRs got closed or merged so these are something like 7% of the PRs and are obviously a skewed sample. This is my summary of each PR: - A reasonable looking PR that has been overlooked by reviewers. - Merge conflicts - AI description - Code mispasted from AI - AI assisted description - not sure if code is correct but don't trust probable AI - An old GSOC PR from Summer - 4000 lines. I guess mentors don't have time. - Reviewer said that the changes are not a good idea - author probably doesn't understand. - AI description and code - not sure if code is good. Was reviewed once but not again after changes. - AI everything - Messing with test runner but I don't trust author or AI to do that. - Experimental PR marked as draft. - AI everything - Feedback given by reviewer and author has not addressed it. - AI everything - Reviewer already told the author that the PR is pointless. - Experimental PR described as draft (not actually marked as such). - AI everything - Author has already been told that the PR is wrong. - Reasonable PR - Reviewer agreed to merge but there are conflicts and author needs to fix them. - Not correct but it is unclear exactly what would be correct so difficult to give feedback. - Reviewer has already said that the PR is wrong. Further changes have been made but I think the reviewer does not feel the need to explain again. - Old GSOC PR - Reviewers seem unsure about the PR. - Weird (wrong) AI PR - Not even sure what kind of AI is being used. No one has reviewed it. - AI PR - Probably okay but not sure if it is a good idea. - AI everything - possibly good but no one has reviewed. - AI everything - added tests have failed but author doesn't know why. - Duplicate of weird AI PR. I think that only a few of these are actually reasonable PRs where the author is ready, capable and willing to work on it but is waiting for review or where there are some good reviews but a little more is needed from the author. My hope would be that a 1 month reminder PR would handle some fraction of those by reminding the author/reviewers to finish/review the PR. The rest I think should just be closed but it is a question of who takes the decision to close and when. It also matters how the decision to close is explained to the author and for anyone else looking to make a PR for the same issue in future. I think that just closing due to inactivity after 2 months is a reasonable approach here. If someone doesn't like it then they can ask for it to be reopened or ask for feedback or something but there shouldn't be any obligation for anyone to actually respond to that. A lot of this is just AI spam and we should not need to spend effort giving feedback when the effort is not being given on the other side. This particular PR probably is a good illustration of how PRs (especially AI PRs) can easily stall: https://github.com/sympy/sympy/pull/28508 The issue is to add docstrings to some code in the LLVM code printer. The author clearly used an AI to write the docstrings. Then the doctests have failed and the author (via AI) said: """ The current doctest failure seem to be coming from llvmlite not being installed in the CI environment (ModuleNotFoundError: No module named 'llvmlite'), which causes cascading NameError exceptions in examples using ir and ll. I believe these errors are unrelated to my docstring changes. Could you please confirm if these are expected/known failures for optional dependencies? Thanks! """ If they had any idea what the LLVM printing code is for then they would have understood the connection with llvmlite. Even if you just think a little bit about what you are saying then the name "llvmlite" is a big clue (maybe that's like counting the r's in strawberry though). I guess that they also probably don't understand how doctests work. They can't have tested any of the code that they are putting in the docstrings if they didn't install llvmlite themselves. So I commented at the bottom of the PR "The llvmlite issue is obviously not unrelated." and then no one has replied to that. Now I could put more effort into explaining what llvmlite is but why should I? What effort has this person put in? My comment is *deliberately* terse because I don't want to have anything to do with their PR unless they are going to put more effort at least to have some understanding of the code they are "writing". It would be possible to install llvmlite in CI to fix the doctest failure although a bit more is needed. Why bother coaching this particular PR to do that though? There is no value in that PR when it is just a few AI generated docstrings pasted in by someone who has no idea what they are doing. They are not the only person to do this either: there have been 8 PRs trying to add those docstrings! I already gave feedback on that PR and did not close it, leaving open the possibility that the author could put in a bit more effort to get it working or at least demonstrate some understanding. Now after 3 months I think it is reasonable to just close it and I don't think anything we can say is particularly better than just a bot saying "closing due to inactivity". -- Oscar -- You received this message because you are subscribed to the Google Groups "sympy" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/sympy/CAHVvXxQXM7Zs7sfUCJZvzyDn71ibTv3BcVKz1NHn2jCxkWbw0A%40mail.gmail.com.
