When doc builds fail, most of the time it's due to ReST formatting or referencing issues.
Most issues have a form similar to
/home/circleci/project/doc/api/next_api_changes/behavior/22135-JSS3.rst:3: WARNING: py:obj reference target not found: Figure._localaxes
My standard way to find out the error is:
Open CircleCI
Download the build log
Search for "WARNING"
This is cumbersome and not very discoverable for new contributors.
Proposed fix
It would be great to have a bot post these warnings if CicleCI fails. I assume that would cover 95% of doc build failures and point contributors (new and experienced) to the problem so that they can solve it faster.
As it turns out, the sections added in #23868 don't actually run on failed doc builds, rendering them ineffective. (They do run on successful doc builds, not that it says anything other than "no errors or warnings")
See, for example this run for which the doc build failed, but no such separate job for pulling out the errors/warnings.
I believe @QuLogic is planning on implementing a more full solution in the near term, including a bot-generated comment on the PR (correct me if I'm wrong/misremembering a conversation)
Agreed that they do not solve it. Not really clear why it is not executed though...
Just want to point out that it would also be good if a solution that writes back a GH comment also takes any deprecations into account. That part of the code actually seems to work as it should...
Activity
story645 commented on Dec 2, 2022
is this closed by #23868?
ksunden commented on Dec 2, 2022
As it turns out, the sections added in #23868 don't actually run on failed doc builds, rendering them ineffective. (They do run on successful doc builds, not that it says anything other than "no errors or warnings")
See, for example this run for which the doc build failed, but no such separate job for pulling out the errors/warnings.
I believe @QuLogic is planning on implementing a more full solution in the near term, including a bot-generated comment on the PR (correct me if I'm wrong/misremembering a conversation)
QuLogic commented on Dec 3, 2022
Yes, and thinking about it since then, I think I have a simpler (in some ways) solution than we discussed, but I haven't fully tested it out yet.
oscargus commented on Dec 3, 2022
Agreed that they do not solve it. Not really clear why it is not executed though...
Just want to point out that it would also be good if a solution that writes back a GH comment also takes any deprecations into account. That part of the code actually seems to work as it should...
oscargus commented on Dec 3, 2022
I think I figured out the issue...