Re: [PATCH] git-send-email.perl: check for lines longer than 998 characters

From: Jeff King <peff@peff.net>
Date: Jan 18, 2008, 21:30 UTC
Message-ID: <20080118213051.GA21321@coredump.intra.peff.net>
In-Reply-To: <7v8x2mdf7e.fsf@gitster.siamese.dyndns.org>

On Fri, Jan 18, 2008 at 12:57:41PM -0800, Junio C Hamano wrote:

Show 7 quoted lines

> >   2/3: send-email: validate patches before sending anything
> >
> >        The actual up front long-lines check.
> 
> I wonder what the performance implication of this approach would
> be, though.  I am tempted to say that it would be negligible --
> scanning text in Perl is fast enough.

We now open and do one conditional per line for each file (in addition to already going through each file a separate time and doing more complex processing). Doing that over the entirety of "git log --pretty=email -p" on git.git takes about 1 second on my machine for 11402 patches. Obviously there's slightly more syscall overhead as you have to open() each patch, but I think think it is clear that the parsing overhead is negligible.

Show 6 quoted lines

> I suspect that taking this "Safe against SMTP line length limit"
> topic all the way ("all the way" is post 1.5.4, I am inclined to
> agree that this may be a good fix to an existing bug) would
> require that git-format-patch --attach to learn to apply QP on
> patch text to avoid producing very long lines to root-cause the
> issue [*1*].

Perhaps. If such things are sufficiently rare, one could simply attach the patch in their MUA. I think the most important thing is for git to at least stop and warn the user that it might not be sending something valid. But implementing N different fixes that haven't even been requested by users seems like a waste of time.

-Peff