[BUG] git-am: all colons in the beginning of a subject are lost

[BUG] git-am: all colons in the beginning of a subject are lost

[Max Filenko]

Hi!

If there are colons in the beginning of a patch subject line `git-am'
will drop them.

Consider the following patch:

    $ cat 0001-four-colons-prepended.patch
    From e8213a2d10a61c9dc75521d88d656b8d5330e6bb Mon Sep 17 00:00:00 2001
    From: Max Filenko <contact@filenko.ms>
    Date: Tue, 12 Feb 2019 12:21:21 +0100
    Subject: [PATCH] :::: four colons prepended
    
    ---
     file.txt | 2 +-
     1 file changed, 1 insertion(+), 1 deletion(-)
    
    diff --git a/file.txt b/file.txt
    index 4dd1ef7..b5da95d 100644
    --- a/file.txt
    +++ b/file.txt
    @@ -1 +1 @@
    -This is a file.
    +This is a plain text file.
    -- 
    2.17.1

There will be no colons in the beginning of a commit message if I apply
this patch:

    $ git am 0001-four-colons-prepended.patch
    Applying: four colons prepended

The four colons already gone in the log message above. There are neither
no colons in the commit subject line:

    $ git show
    commit 6341a6a2872f850ecb376c268b1b3bae54a6a74f (HEAD -> master)
    Author: Max Filenko <contact@filenko.ms>
    Date:   Tue Feb 12 12:21:21 2019 +0100
    
        four colons prepended
    
    diff --git a/file.txt b/file.txt
    index 4dd1ef7..b5da95d 100644
    --- a/file.txt
    +++ b/file.txt
    @@ -1 +1 @@
    -This is a file.
    +This is a plain text file.

I was able to reproduce this with git 2.17.1 on Ubuntu 18.04.2 LTS as
well as with git 2.17.2 (Apple Git-113) on macOS 10.14.3.

I was able to trace this down to <builtin/am.c>. It seems like there are
no colons already in the `state->msg' which to my understanding is being
filled by `read_commit_msg()' function. I would really appreciate a hand
on debugging it further.

I'm re-submitting this bug report because the original one [1] wasn't
really noticed. Hopefully, it's just because I've missed the proper
prefix in my email's subject line :)

[1]:
http://public-inbox.org/git/m2lg2lxmmm.fsf@bouncer.i-did-not-set--mail-host-address--so-tickle-me/

--
Best,
Max

Re: [BUG] git-am: all colons in the beginning of a subject are lost

[Jeff King]

On Thu, Feb 28, 2019 at 11:02:11AM +0100, Max Filenko wrote:

>     Subject: [PATCH] :::: four colons prepended
> [...]
> There will be no colons in the beginning of a commit message if I apply
> this patch:
> 
>     $ git am 0001-four-colons-prepended.patch
>     Applying: four colons prepended

I suspect this has to do with the sanitization that happens as part of
removing "[PATCH]". Note that if you use "-k" (to preserve the subject)
it doesn't happen, though of course you also get "[PATCH]" then.

If you want to pass the subject lines through verbatim, use "-k" with
both format-patch and git-am.

> I was able to trace this down to <builtin/am.c>. It seems like there are
> no colons already in the `state->msg' which to my understanding is being
> filled by `read_commit_msg()' function. I would really appreciate a hand
> on debugging it further.

It's probably easier to debug with git-mailinfo, which has the same
behavior:

  $ git mailinfo msg patch <0001-four-colons-prepended.patch
  Author: Jeff King
  Email: peff@peff.net
  Subject: four colons prepended
  Date: Thu, 28 Feb 2019 06:12:50 -0500

and is based on the same routines.

The contents are preserved until we end up in mailinfo.c's
cleanup_subject(). And there leading colons are explicitly removed:

       case ' ': case '\t': case ':':
               strbuf_remove(subject, at, 1);
               continue;

That behavior goes all the way back to 2744b2344d (Start of early patch
applicator tools for git., 2005-04-11), when Git was only 4 days old.

Since it also handles cruft like "Re:", I suspect the goal there was I
suspect the goal there was to remove cruft like "Re::::" or "Re: :"
which sometimes happens. I don't know if anybody would complain if we
were more careful about leaving lone colons that weren't part of a "Re"
chain.

-Peff