Git for Windows/ARM64 improperly uses `/clangarm64/etc/gitconfig` instead of `/etc/gitconfig` #5431

dscho · 2025-02-18T20:46:08Z

Actually, I just noticed something: for some reason, the system gitconfig file lives in /etc on x86 installs, but it seems the clangarm64 git expects it in /clangarm64/etc ? In any event, git config --list is empty after that procedure.

Originally posted by @jeremyd2019 in #5292

The text was updated successfully, but these errors were encountered:

rimrul · 2025-02-19T12:16:43Z

Looks like fb5e337 needs a little extension.

In fb5e337 (mingw: move Git for Windows' system config where users expect it, 2021-06-22), I moved the location of Git for Windows' system config and system Git attributes file to the top-level `/etc/` directory (because it is a much more obvious location than, say, `/mingw64/etc/`). The patch relied on a very specific scenario that the newly-supported Windows/ARM64 builds of `git.exe` fails to fall into. So let's broaden the condition a bit, so that Windows/ARM64 builds also use that location (instead of the even more obscure `/clangarm64/etc/` directory). This fixes #5431. Signed-off-by: Johannes Schindelin <[email protected]>

rimrul · 2025-02-26T08:31:31Z

We'll need to remember to add a follow up like git-for-windows/build-extra#262

dscho · 2025-02-26T09:19:49Z

We'll need to remember to add a follow up like git-for-windows/build-extra#262

Good point, thank you!

dscho · 2025-02-26T11:48:52Z

We'll need to remember to add a follow up like git-for-windows/build-extra#262

Good point, thank you!

@rimrul actually, I had a look, and it would appear that I was careful to use whatever git -c core.editor=echo config --system -e would report as the location. Do you see anything that needs to be done in addition?

rimrul · 2025-02-26T11:53:43Z

Wouldn't we try to restore the settings from the new location when an ARM user upgrades?

dscho · 2025-02-26T12:07:43Z

Wouldn't we try to restore the settings from the new location when an ARM user upgrades?

Ah. I do not actually think so, we are talking about the system config, and the system config should really be owned by Git for Windows, not by the user. Custom Git configuration goes into ~/.gitconfig, after all.

rimrul · 2025-02-26T12:11:13Z

Wouldn't we try to restore the settings from the new location when an ARM user upgrades?

Ah. I do not actually think so, we are talking about the system config, and the system config should really be owned by Git for Windows, not by the user. Custom Git configuration goes into ~/.gitconfig, after all.

Yes, but we do try to infer config from the system config in the installer.

dscho · 2025-02-26T12:20:18Z

Wouldn't we try to restore the settings from the new location when an ARM user upgrades?

Ah. I do not actually think so, we are talking about the system config, and the system config should really be owned by Git for Windows, not by the user. Custom Git configuration goes into ~/.gitconfig, after all.

Yes, but we do try to infer config from the system config in the installer.

We do, indeed. And we do that by running git config -l -z --system. That is done before the previous Git for Windows installation is removed, i.e. it would use the old behavior and retrieve the values from /clangarm64/etc/gitconfig. Then, the installer would first run the uninstaller, and subsequently install the new Git for Windows version, reinstating the system config settings using the new git.exe, i.e. writing into /etc/gitconfig.

So I think this already works as desired.

Most notably, this fixes #5431, i.e. that the ARM64 flavor of Git for Windows does not use `/etc/gitconfig` by mistake, but `/clangarm64/etc/gitconfig` instead. My original intention was to fix only that issue, but when I fired up a Git for Windows/ARM64 SDK and tried to build, I immediately ran into trouble, then noticed places where ARM64 was not yet handled, and one thing led to another and now it's a 6-patch PR instead of a single-patch one.