Re: fast-import [mis?]-honors core.ignorecase

From: Joshua Jensen <jjensen@workspacewhiz.com>
Date: Jun 26, 2012, 03:45 UTC
Message-ID: <4FE93070.1050507@workspacewhiz.com>
In-Reply-To: <CAG+J_DygPXtjD-0gv8XXpV0JErw_jpwRLOZ00H9bem5hN8g7ZA@mail.gmail.com>

----- Original Message -----
From: Jay Soffian
Date: 6/25/2012 2:48 PM

Show 16 quoted lines

> I was using a "fast-export | <filter program> | fast-import" as a much
> faster filter-branch --env-filter (I needed to rewrite
> author/committer email-addresses) and was surprised that after the
> pipeline was done, the new ref's trees didn't match the old ref's
> trees. i.e.:
>
>    $ cmp <(git log old-ref --pretty='%T')  <(git log --pretty='%T' new-ref)
>
> didn't return 0. Comparing one of the differing tree-pairs indicated
> it was due to a case-difference in a file which had been renamed.
> After some experimenting I finally tracked this down to having
> core.ignorecase set to "true" on the repo in which I was running the
> pipeline. By setting it to "false" the filtering pipeline completes
> with new-ref (and all its ancestors) being tree-identical to old-ref.
>
> Is there any reason for fast-import to honor ignorecase?

I provided this change in ac1c80f7645b6fa90534890e1f83005d40d98281.

In my mind, core.ignorecase=true means that the file system cannot handle case-sensitive filenames. Files held in the repository should conform to a single case.

When I was getting a 'fast export' of our Perforce repository, I found file case was all over the map. Perforce internally handles its files for a case preserving case insensitive file system in a case insensitive manner. File case can change from changelist to changelist! The fast-exported file was a mess. In the original fast-imported version, 'git status' thought that one case of file was modified when another case of the same file was the same. Git could not follow the history over all of the filename case changes.

The case folding patches were created to handle this. Specifically, the fast-import patch serves as a catch-all to handle repositories that do not conform to a single case as they were created and in operation. It folds the incoming file case to match that of what is in the existing repository. The end result is a clean repository with one file case for a given filename. 'git status' works. File history can be followed. 'git add' knows which name to match against.

(I am a firm believer that Git should handle *all* case _internally_ as insensitive when core.ignorecase=true, and I have work-in-progress patches that illustrate this behavior. Others here (particularly on the msysgit mailing list) are adamantly against those patches. Nonetheless, there would be no need to fold case if the Git internals handled comparisons in a case insensitive fashion.)

I do understand your point above, and I have no good answer. Your file case was changed during a rename. The fast-import core.ignorecase=true process did not preserve that case change despite it being right and appropriate for your repository. core.ignorecase=false did preserve the case. I guess the solution here depends on what core.ignorecase=true should mean.

Nevertheless, I am certain Git fast-import needs to retain the case folding behavior during fast-import, even if it means enabling it via a command-line flag instead of core.ignorecase=true.

-Josh