Add .git-blame-ignore-revs

[akx]

This makes GitHub (and other tools that support this file, or any Git client with git config blame.ignoreRevsFile .git-blame-ignore-revs set) ignore the large reformatting/automatic commits, to make blame inspection easier.

[akx]

@Pierre-Sassoulas Added the file to MANIFEST.in too to make the check Tox step pass.

[Pierre-Sassoulas]

76fb2a6 could be removed from the git blame but not ff3fdd8 which introduce some non formatting changes (remove six, maybe others) and some unsafe fixes from ruff. I'm pretty sure the migration from os.path to pathlib is not automated either (I initially thought it was because of the MR ff3fdd8 was introduced in, until I actually tried and the deception was strong).

[webknjaz]

I'd argue it's not just a formatting change but a functional one. F-strings are more performant so it's more of an enhancement rather than formatting. Usually only patches that are solely formatting are ignored. Bits that are useful for git paleontology aren't supposed to be hidden...

[Pierre-Sassoulas]

I'm not sure what's the point of .git-blame-ignore-revs in Manifest.in : after packaging git is not involved anymore, right ? So if f-strings removed from git blame are not consensual, there would be nothing left of this MR if we also remove that 😅

[akx]

I'm not sure what's the point of .git-blame-ignore-revs in Manifest.in

Without it there, check-manifest would fail.

[Pierre-Sassoulas]

What is check-manifest and why does it fail ? manifest.in is linked to python packaging and does not involve git, right ?

[ionelmc]

check-manifest is used to make sure the sdist contains all the source files and nothing is missing. Maybe this new file needs to be an exclude in there.

But alas, I am not sure who is this file for. I will not use it - I prefer to see everything in the gitblame, and jumping back a commit is pretty easy enough (it's just one button in github anyway).

[webknjaz]

But alas, I am not sure who is this file for. I will not use it - I prefer to see everything in the gitblame, and jumping back a commit is pretty easy enough (it's just one button in github anyway).

@ionelmc this file is picked up by GitHub automatically, and it hides the listed commits from blame, unlike local Git invocations where that needs to be configured.

[webknjaz]

check-manifest is used to make sure the sdist contains all the source files and nothing is missing. Maybe this new file needs to be an exclude in there.

Have you considered using setuptools-scm which auto-includes everything Git-tracked into sdists?

[webknjaz]

@ionelmc would you be open to integrating setuptools-scm?

[ionelmc]

@webknjaz I am not opposed in principle. I use it on projects with C extensions where it's useful to have uniquely versioned pre-release distributions. But on this project I don't see the value to change an already working packaging setup. As a matter of fact I only see myself responding to bug reports caused by such a change.

Anyway, regarding this ignore-revs change - I don't want it. I simply don't trust even myself to not introduce behavior changes in formatting commits.