fetchTree/fetchGit: add test for .gitattributes

[DavHau]

...with the intention to prevent future regressions in fetchGit

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/2023-11-20-nix-team-meeting-minutes-105/35902/1

[thufschmitt]

Discussed in the Nix maintenance meeting.

Not sure how to work forward with that (and #9240 probably just broke it), but we should at least either

• Reimplement that on top of libgit2, or
• Warn of the behaviour change

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/2023-11-24-nix-team-meeting-minutes-106/35955/1

[roberth]

Rebased

[DavHau]

I noticed there is actually another thing that gets processed by git-archive, which is export-subst.

I added a commit to the current PR adding a test which passes on the current stable nix but fails on master.

The bad news is that export-subst seems to be more complex to implement than export-ignore.

I posted some more info on a related issue on libgit2: libgit2/libgit2#918 (comment)

[roberth]

more complex

I wouldn't expect the log format to be particularly stable in future versions, so this could be painful to support. We'll probably want to support a subset, and error out if there's anything non-reproducible in there.
Even something seemingly innocuous like an abbreviated git hash might be impure by depending on the number of commits or objects.