Re: [PATCH] .gitattributes: CR at the end of the line is an error

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 21, 2009, 09:31 UTC
Message-ID: <7vvdmqrl06.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20090619194253.6117@nanako3.lavabit.com>

Nanako Shiraishi <nanako3@lavabit.com> writes:

Show 23 quoted lines

> When a CR is accidentally added at the end of a C source file in the git
> project tree, "git diff --check" doesn't detect it as an error.
>
>     $ echo abQ | tr Q '\015' >>fast-import.c
>     $ git diff --check
>
> I think this is because the "whitespace" attribute is set to *.[ch] files
> without specifying what kind of errors are caught. It makes git "notice
> all types of errors" (as described in the documentation), but I think it
> is incorrectly setting cr-at-eol, too, and hides this error.
>
> Signed-off-by: Nanako Shiraishi <nanako3@lavabit.com>
> ---
>
> diff --git a/.gitattributes b/.gitattributes
> index 6b9c715..bb03350 100644
> --- a/.gitattributes
> +++ b/.gitattributes
> @@ -1,2 +1,2 @@
>  * whitespace=!indent,trail,space
> -*.[ch] whitespace
> +*.[ch] whitespace=indent,trail,space
>

I like the result of applying this patch to my tree.

A "whitespace" attribute that is Set, which is what the original has, is defined to "notice all types of errors known to git", it is a poor way to define the project policy, which was what 14f9e12 (Define the project whitespace policy, 2008-02-10) tried to do. It means the policy will silently change when newer git learns to detect more types of whitespace errors.

And it never meant to allow trailing carriage-returns. I think the implementation of whitespace attribute handling is broken.