(Git-1) use tmpdir for feedstocks instead of static directory

[ytausch]

This is first step of #2812.

• Pydantic model updated or no update needed

[codecov]

Codecov Report

Attention: Patch coverage is 91.93548% with 10 lines in your changes missing coverage. Please review.

Project coverage is 77.49%. Comparing base (90a1db8) to head (c9091c9).
Report is 15 commits behind head on master.

Files	Patch %	Lines
conda_forge_tick/auto_tick.py	50.00%	6 Missing ⚠️
conda_forge_tick/git_utils.py	33.33%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2872      +/-   ##
==========================================
+ Coverage   77.07%   77.49%   +0.41%     
==========================================
  Files         112      115       +3     
  Lines       12053    12215     +162     
==========================================
+ Hits         9290     9466     +176     
+ Misses       2763     2749      -14     

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

[ytausch]

@beckermr ready for review

[beckermr]

Cool. I'm away for the next few days so it will likely be a bit.

[beckermr]

pre-commit.ci autofix

[beckermr]

I have some comments / requests / questions. Also, in general I was in the process of removing the context module and its associated dataclasses. They had carried sensitive state in the past and needing them made it a pain to debug the bot. This PR adds more things. If we want to keep them, it might be useful to add factory functions to make a context given some simple data (e.g., the feedstock name).

[ytausch]

Also, in general I was in the process of removing the context module and its associated dataclasses. They had carried sensitive state in the past and needing them made it a pain to debug the bot.

Are you referring to the attrs field? Because the default_branch and feedstock_name should not make so much trouble. Would be helpful to see what exactly is making the trouble. I planned to extend the functionality in further PRs, e.g. to access the automerge feedstock attribute directly as an attribute of the context: See here

This PR adds more things. If we want to keep them, it might be useful to add factory functions to make a context given some simple data (e.g., the feedstock name).

I think I don't really get it. What for? So that setting attrs is not needed anymore?

[beckermr]

Yeah it was setting the attrs field which is a pain, but also I don't like having more than one effective api for things related to the feedstocks. Then, for example, we get code that looks up the automerge attribute either directly from attrs or from this new attribute on the context. It's confusing to have both.

[ytausch]

Understood. But let's get this though nonetheless then? In my imagination, the FeedstockContext serves as kind of a proxy between attrs and the individual attributes accessed in the code - and I have pursued this rather consequently in my changes I think.

Open to refactor this further in separate PRs of course.

[ytausch]

What is wrong with the pre-commit CI integration? It shows a timeout for some reason

[beckermr]

Idk. Maybe try and retrigger with an empty commit.

[beckermr]

I'm seeing this in other repos though.

[ytausch]

It also happened in previous commits here - maybe it is an upstream issue?