Show 62 quoted lines
> On 04 Mar 2016, at 13:16, Torsten Bögershausen <tboegi@web.de> wrote:
>
> On 03/04/2016 10:07 AM, Alexander Rinass wrote:
>> Hallo,
>>
>> It appears that the git diff command does not precompose file path arguments, even if the option core.precomposeunicode is set to true (which is the default on OS X).
>>
>> Passing the decomposed form of a file path to the git diff command will yield no diff for a modified file.
>>
>> In my case, the decomposed form of the file path is sent by the OS X Cocoa framework's NSTask, wich I am using in an application. It can be simulated on OS X by using $(iconv -f utf-8 -t utf-8-mac <<< FILE_PATH) as file path argument on the shell.
>>
>> Git commands like add, log, ls-tree, ls-files, mv, ... accept both file path forms, git diff does not.
>>
>> It can be tested with the following setup on OS X (as iconv's utf-8-mac encoding is only available on OS X):
>>
>> git init .
>> git config core.quotepath true
>> git config core.precomposeunicode true # (default on OS X)
>> touch .gitignore && git add .gitignore && git commit -m "Initial commit"
>> echo "." >> Ä
>> git add Ä
>> git commit -m "Create commit with unicode file path"
>> echo "." >> Ä
>> This gives the following status, showing the precomposed form of "Ä":
>>
>> git status --short
>> M "\303\204"
>> Running git add with both forms does work as expected:
>>
>> git add Ä
>> git status --short
>> M "\303\204"
>> git reset HEAD -- Ä
>> git add $(iconv -f utf-8 -t utf-8-mac <<< Ä)
>> git status --short
>> M "\303\204"
>> git reset HEAD -- Ä
>> However, running git diff only works with the precomposed form:
>>
>> git status --short
>> M "\303\204"
>> git --no-pager diff -- Ä
>> [...shows diff...]
>> git --no-pager diff -- $(iconv -f utf-8 -t utf-8-mac <<< Ä)
>> [...shows NO diff...]
>>
>> I took a look at the Git source code, and the builtin/diff*.c do not contain the parse_options call (which does the precompose_argv call) that the other builtins use.
>>
>> But I am not really familiar with either C or the Git project structure, so this may not mean anything.
>>
>> Best regards,
>> Alexander Rinass
>>
> Good analyzes, and thanks for the report.
> It should be possible to stick in a
>
> precompose_arrgv(argc, argv)
>
> into builtin/diff.c
>
> Do you you can test that ?
>
Sticking a precompose_argv(argc, argv) into diff.c’s cmd_diff function fixes the issue.
But I had to disable the check (precomposed_unicode != 1) in precompose_argv to make it work. That’s probably because precompose_argv is usually called from parse_options and is missing some other call before it?
I think it is clear that diff.c and friends are definitely missing the precomposing step. I am not sure about the right way to fix though (should parse_options be used in the end?) and my C skills are basic at best, otherwise I would create a patch.
Alexander Rinass <alex@fournova.com>