Re: [PATCH v1 1/1] git diff --quiet exits with 1 on clean tree with CRLF conversions

[Junio C Hamano <gitster@pobox.com>]

Torsten Bögershausen <tboegi@web.de> writes:

>>>         enum safe_crlf crlf_warn = (safe_crlf == SAFE_CRLF_FAIL
>>>                                     ? SAFE_CRLF_WARN
>>>                                     : safe_crlf);
>>> +       if (size_only)
>>> +               crlf_warn = SAFE_CRLF_FALSE;
>> 
>> If you were to go this route, it may be sufficient to change its
>> initialization from WARN to FALSE _unconditionally_, because this
>> function uses the convert_to_git() only to _show_ the differences by
>> computing canonical form out of working tree contents, and the
>> conversion is not done to _write_ into object database to create a
>> new object.
>
> Hm, since when (is it not used) ?

Since forever, but my statement above said "this function", which may
have confused you, where it could have said diff_populate_filespec().

Surely it is possible for somebody to diff_populate_filespec(s, 0)
and then call hash_sha1_file(s->data, s->size, "blob", ...) to write
the data into the object database to create a new object.  But that
sounds really crazy, no?

> The SAFE_CRLF_FAIL was converted into WARN here:
> commit 5430bb283b478991a979437a79e10dcbb6f20e28
> Author: Junio C Hamano <gitster@pobox.com>
> Date:   Mon Jun 24 14:35:04 2013 -0700
>
>     diff: demote core.safecrlf=true to core.safecrlf=warn

Yes.

> Does this all means that, looking back,  5430bb283b478991 could have been more
> aggressive and could have used SAFE_CRLF_FALSE ?

That is pretty much the statement, to which you said "since when",
suspects.

> And we can do this change now?

I am not sure.  The conversion the safe-crlf code does is unsafe and
it is a disservice to users not to warn whenever we notice they are
risking information loss.  Maybe time they run "git diff" is not a
good time to warn, as they may not be actually adding the file
as-is, but if warning against information loss at "git diff" time is
important enough, the I think that should not be squelched by the
"--quiet" option, which is about "do not show the patch text
output".  It should not be taken as "do not diagnose any errors".