Inside git’s internal plumbing folder, git holds a file with the branch name and all of the references (files and changes) for that branch.
When you make a new branch git will update its internal plumbing checking to see if the new branch already exists, updates its references to the new branch if it doesn’t (all held internally in a case sensitive way). It will then make that new branch file, git has already checked that the case senitive name for the branch doesn’t exist internally, so it should be good to go.
Part of its process is creating that internal branch file… But wait!
Windows doesn’t have case sensitive naming so when it tries to make that new branch file it will overwrite the old one (since it shouldn’t exist by git’s own reference!) All of the files and references for it now get nuked.
Now you’re at best back to wherever that originally named branch came from, at worse your .git folder is properly borked.
I’m probably going to get downvoted to Hell and back, but someone’s gotta say it: that’s a git problem, not Windows.
First of all, I agree that case-insensitive file systems suck. It makes things inconsistent, especially from a development standpoint.
But, everyone has known that Windows (and macOS) use case insensitive file systems. At least for Windows, it always has been that way.
Git was written in Linux, which uses a case sensitive file system. So it’s no surprise that its internals use case insensitive storage. Someone ported it over to Windows, and I’m sure they knew about the file system differences. They could’ve taken that into account for file systems that are case insensitive, but chose not to do anything to safe guard Windows users.
But until the day that somebody fixes Git, everybody who is not using case sensitive file systems needs to care more about how they name things (and make sure their team does too). Because fuck everyone else, right?
The issue isn’t just a simple oversight. Git includes the file name as part of the tree and commit hash. The hash has security implications. There’s really no way to make the hash support case insensitivity without opening up a multitude of holes there. So there will always be a mismatch, and you can’t just fix it without changing how git works from the ground up.
I was talking about branch names, not file names. File duplicates due to case sensitivity aren’t a problem on Windows anyway because those are already enforced by the file system. Unless you have people working on Linux that have multiple files with a similar name but with different casing but those should know better.
You want all applications to explisitly support each individual filesystem? That sounds insane. It is absolutly fair to demand some common ground like posix compliance. And a windows user can like anyone else just mount their git repo area using any other filesystem.
I agree that the meme is silly to point to this problem specifically. However, if we leave aside application support as a reason to prefer an OS, then there is really not a lot of arguments left for choosing Windows…
Also, if you use an NTFS USB drive to move the .git folder, you will be in trouble.
Thankfully, moving those things to pen drives is very slow, making most users tar it first, anyway, hence sidestepping the problem.
I never understood this. I’ve been using macOS for a long, long time but in Terminal with either bash or zsh it’s always been as picky as Linux/other UNIX systems with casing.
Even in Finder if you navigate to a directory by path you have to use the proper case.
Am I missing something? I haven’t manually chosen a case-sensitive filesystem, but I sure would if it didn’t already seem case-sensitive.
Huh. I had forgotten that git does actually create a file with the branch name. But it doesn’t actually screw up the .git folder or lose your data when you try to do a rename like this; it just rejects the rename unless you also use the “force” option. This has been the case since at least January of 2020. But apparently it actually doesn’t always use a local file for branch names, so sometimes there’s a problem and sometimes there isn’t, which I guess is arguably worse than just having consistently-surprising behavior.
Inside git’s internal plumbing folder, git holds a file with the branch name and all of the references (files and changes) for that branch.
When you make a new branch git will update its internal plumbing checking to see if the new branch already exists, updates its references to the new branch if it doesn’t (all held internally in a case sensitive way). It will then make that new branch file, git has already checked that the case senitive name for the branch doesn’t exist internally, so it should be good to go.
Part of its process is creating that internal branch file… But wait!
Windows doesn’t have case sensitive naming so when it tries to make that new branch file it will overwrite the old one (since it shouldn’t exist by git’s own reference!) All of the files and references for it now get nuked.
Now you’re at best back to wherever that originally named branch came from, at worse your .git folder is properly borked.
I’m probably going to get downvoted to Hell and back, but someone’s gotta say it: that’s a git problem, not Windows.
First of all, I agree that case-insensitive file systems suck. It makes things inconsistent, especially from a development standpoint.
But, everyone has known that Windows (and macOS) use case insensitive file systems. At least for Windows, it always has been that way.
Git was written in Linux, which uses a case sensitive file system. So it’s no surprise that its internals use case insensitive storage. Someone ported it over to Windows, and I’m sure they knew about the file system differences. They could’ve taken that into account for file systems that are case insensitive, but chose not to do anything to safe guard Windows users.
But until the day that somebody fixes Git, everybody who is not using case sensitive file systems needs to care more about how they name things (and make sure their team does too). Because fuck everyone else, right?
The issue isn’t just a simple oversight. Git includes the file name as part of the tree and commit hash. The hash has security implications. There’s really no way to make the hash support case insensitivity without opening up a multitude of holes there. So there will always be a mismatch, and you can’t just fix it without changing how git works from the ground up.
Of course you can, make it lowercase internally and store the case formatted string for output.
That’d break git repos where files with the same name, but different case exist.
I was talking about branch names, not file names. File duplicates due to case sensitivity aren’t a problem on Windows anyway because those are already enforced by the file system. Unless you have people working on Linux that have multiple files with a similar name but with different casing but those should know better.
Yes you can
Well … everyone using case insensitive FSs need to worry how they name stuff anyway.
Absolutly. And they can just mount the git store using another filesystem. Would be the by far easiest solution.
Or use OS’s where case sensitive FS’s are the default. The ones I have in mind are much better for programming anyway…
Well obviously. But I assume they would not use windows unless they are forced to.
You want all applications to explisitly support each individual filesystem? That sounds insane. It is absolutly fair to demand some common ground like posix compliance. And a windows user can like anyone else just mount their git repo area using any other filesystem.
I agree that the meme is silly to point to this problem specifically. However, if we leave aside application support as a reason to prefer an OS, then there is really not a lot of arguments left for choosing Windows…
@[email protected]
Beware neckbeards with pitchforks.
It’s not even actually that bad, at least not since January of 2020: https://stackoverflow.com/a/59687740/1858225
That’s called a workaround. No end user should have to rely on a workaround as a solution to a bug; and make no mistake, it’s a bug.
But whose bug is it? Really, Git origins have it tied to Linux development.
Case sensitivity, or the lack thereof, on a filesystem is opinionated. That’s the real issue and is not a bug.
Sounds like a feature to me
My point is that the claim in the comic and in other comments that this corrupts your repo or loses work simply isn’t true.
Also, if you use an NTFS USB drive to move the .git folder, you will be in trouble.
Thankfully, moving those things to pen drives is very slow, making most users tar it first, anyway, hence sidestepping the problem.
Does Mac suffer the same bugs as windows in this case?
Yes. Thankfully in my experience I’ve only dealt with this once or twice. But it’s a pita every time.
I’ve tried switching macOS to a case sensitive file system, but not all programs can handle it (at the time it was Photoshop).
I never understood this. I’ve been using macOS for a long, long time but in Terminal with either bash or zsh it’s always been as picky as Linux/other UNIX systems with casing.
Even in Finder if you navigate to a directory by path you have to use the proper case.
Am I missing something? I haven’t manually chosen a case-sensitive filesystem, but I sure would if it didn’t already seem case-sensitive.
Huh. I had forgotten that git does actually create a file with the branch name. But it doesn’t actually screw up the
.gitfolder or lose your data when you try to do a rename like this; it just rejects the rename unless you also use the “force” option. This has been the case since at least January of 2020. But apparently it actually doesn’t always use a local file for branch names, so sometimes there’s a problem and sometimes there isn’t, which I guess is arguably worse than just having consistently-surprising behavior.