Sphinx may not generate an api folder

[lod]

Trivial fix for a bug with simple projects where sphinx does not generate a docs/build/html/api/ folder. Without this fix the rename fails because the output_path does not exist.

[adiroiban]

Hi David,

Thanks for the PR.

Can you please add more details about this use case?

Why is Sphinx not generating the API sub-folder ?

If Sphinx is not generating the API sub-folder, it looks like you are not using the pydoctor extension as designed.

If this is a valid use case, we also need documentation and automated test for this to make sure this feature is supported in future versions.

Regards

[tristanlatr]

Hello, this seems related to #728, which I cannot figure out! So if this check fixes the issue I’m very very happy.

Did you find a minimal reproducer for this issue ? We should really have a test for this, would it possible to add it ?

Thanks a lot.

[tristanlatr]

You can check the pr #742 in which I added some sphinx extension test in the hope to find a reproducer but I did not managed.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.56%. Comparing base (dca9625) to head (21ca88c).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #799      +/-   ##
==========================================
- Coverage   92.62%   92.56%   -0.06%     
==========================================
  Files          47       47              
  Lines        8393     8393              
  Branches     1850     1850              
==========================================
- Hits         7774     7769       -5     
- Misses        356      359       +3     
- Partials      263      265       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tristanlatr]

Hi @lod, did you managed to get an actual reproducer for that issue?

[tristanlatr]

Hello @lod, have you been able to pin point why Sphinx might not generate the api folder?

[adiroiban]

I put together a minimal executable reproduction
I reproduced the same path locally 3/3 with the originally reported

@fly1d you reproduced this error or an LLM agent did it... or you are an LLM agent ?

We are happy to review contributions created by humans

[glyph]

@fly1d One of the side-effects that I find concerning about LLM use is a sort of hypomanic state that it can put users into, which seems to be a precursor to the much-vaunted "AI psychosis" that has been featured in the news. The extreme verbosity, inappropriate emotional intensity, defensiveness on behalf of models, and hostility that your comment exhibits suggests to me that you may be suffering these effects.

I'll be banning you from the project because this outburst was completely inappropriate, but I would seriously urge you to seek out some mental health support before this escalates to serious real-life consequences for you.

[adiroiban]

Now... back to the topic. We have this error and nobody else reproduced this error due to 2 main reasons ... I might be wrong :

• The whole building sphinx + epydoc impementation is one big hack. I have implemented it just to get the docs.twisted.org working
• There are not many people using epydoc + sphinx

---

Regarding the other part with disclosure. I am only now aware that the patch was generated by ReproMin. And the tool itself didn't disclose it.
I am not saying that it tried to hide on purpose, but not disclosing this is a big product design failure.

Codecov or codspeed or other tools will disclose by default that the output from those tools was not human generated and don't even try to look like human generated.

Regarding the superiority of the model.
We know for mare that 10 years that Neural Machine Translation can generate output that is better than a person. I was using Google Translate for a long time and hallucination rate was low.

[adiroiban]

@glyph regarding blocking person... I understand that it can take a big emotional toll to interact with LLM agents.

I just wanted to say that in this case , this block feels like censorship.

I know there a big debate regarding freedom of speach and death penalty, and I don't wanted to go here into details or contest your decision.

I just wanted to say that I am not 100% happy with a person being blocked forever.

Maybe we can have something like block for 1 day... or 1 week

[glyph]

It's just moderation. This person is not censored, they are welcome to use whatever other channels are available for their advertisements for OpenAI's models; there has never been a time in history where people had so many opportunities to publicize a message, and never before has there been a message so widely publicized as "use ChatGPT". I actually laughed out loud at the idea that I could "censor" it.

They are just not welcome to use resources under our control to spread such a message.

That said, if you think that a permanent ban is too harsh, that such a person has valuable things to contribute, we could do 90 days as a default for this sort of thing in the future, and consider unbanning them after that window. I am skeptical.

[adiroiban]

If GitHub does not provide tools too automatically unblock, then we should not spend more time on that

[glyph]

If GitHub does not provide tools too automatically unblock, then we should not spend more time on that

Agreed it's not worth much more effort. I'll stick with a 90 day ban in the future instead of unlimited (1 week is far too short; people do not recover from this state of mind within a week). For this person I guess I will set a 90-day reminder for myself but not do anything else.

[glyph]

(To be clear: Github does have an "auto-unban" but only when you select it at the time of the ban and I didn't in this case.)