Re: [PATCH] Add -k option to cvsexportcommit to squash CVS keywords

[Alex Bennee <kernel-hacker@bennee.com>]

2009/5/31 Junio C Hamano <gitster@pobox.com>:
> Alex Bennee <kernel-hacker@bennee.com> writes:
>
>> <snip>
>>
>>
>> From a1a9477d6e332617526aaab488602552b77832d9 Mon Sep 17 00:00:00 2001
>
> Thanks.
>
> You would want to put all the commentary up to here below the "---" line,
> if you want to give an introductory text that is not part of your proposed
> commit log message.

I'm not sure I quite follow. The email is an email with description
above followed by the output of git-format-patch -1. Or should the
email commentary be part of the commit commentary? Should I have just
fired the commit at the mailing list direct from the command line?

>
>> From: Alex Bennee <alex@bennee.com>
>> Date: Thu, 28 May 2009 14:31:51 +0100
>> Subject: [PATCH] Add -k option to cvsexportcommit to squash CVS keywords
>
> You generally do not want these three lines, unless the "From: " your
> e-mail shows is different from the name and e-mail of the real author of
> the patch.
>
> I presume you would want alex-at-bennee-dot-com in the commit, so in this
> case it might be Ok, but then it may be more straightforward to use that
> name on the "From: " line of your e-mail to begin with, but on the other
> hand, you have a commit as kernel-dash-hacker-at-the-same-domain in our
> history already.  If you prefer the latter, you do not need any of the
> above three lines (the subject is the same as your e-mail "Subject: "
> anyway).

Hmmm yes. My actual email is as referenced, kernel-hacker is the alias
I use for my mailing list activity. If people have questions they wish
to email me directly about I would assume the master email makes the
best option for the commit message?

>
>> Depending on how your CVS->GIT conversion went you will have some
>> unexpanded CVS keywords in your
>> GIT repo. If any of your git commits touch these lines then the patch
>> application will fail. This
>> patch addresses that by filtering files before applying the patch
>> ---
>
> The message is wrapped in a funny way, and it lacks sign-off.

Arrgh. It looks like the Google mail client has futzed with it. Trying
to get inline patches not munged by mail clients seems to be an
exercise in frustration. Usually I would create an attachment which
would then stay unmolested, however inline is the preferred posting
style for the list.

I'll resend with a sign-off.

>
> Isn't it "expanded" (not "unexpanded") keyword the problem you are trying
> to address?  "By filtering files" in what way?  I can guess "filtering them
> back to unexpanded form", but please do not make me guess.

The files in the CVS tree will have expanded keywords, in the GIT tree
they will be in the unexpanded state. This patch will return the files
in the working CVS tree to the unexpanded state so the patch from the
GIT tree applies cleanly.

I'll try and re-word to make it clearer.

>
>> diff --git a/Documentation/git-cvsexportcommit.txt
>> b/Documentation/git-cvsexportcommit.txt
>> index 2da8588..b328dd4 100644
>> --- a/Documentation/git-cvsexportcommit.txt
>> +++ b/Documentation/git-cvsexportcommit.txt
>> @@ -63,6 +63,9 @@ OPTIONS
>>  -u::
>>       Update affected files from CVS repository before attempting export.
>>
>> +-k::
>> +     Filter CVS keywords (like $Revision$) before applying patch.
>> +
>
> "Filter in what way" applies here as well.
>
> Is it really sufficient to unmunge "$Revision$? and "$Id"?  What about
> "$Date$", for example?

You're right. Lazily I had only hit the previous cases in the actual
problem files I had. I'll have a dig in the manual and add the rest.

>
>> @@ -266,9 +266,32 @@ foreach my $f (@files) {
>>       $dirty = 1;
>>       warn "File $f not up to date but has status '$cvsstat{$f}' in your
>> CVS checkout!\n";
>>      }
>> +
>> +    # Depending on how your GIT tree got imported some of the CVS
>> +    # expansion keywords would have been squashed. This will break
>> +    # application of the patch if you touched any lines that had them.
>
> I am not quite getting this comment.  "Squashed" sounds like "$Revision$"
> without expansion instead of "$Revision: 1.4 $"; I thought the issue you
> are addressing is that the automated change that comes from the CVS side
> to the expanded keyword gets in the way, i.e. if these always were
> "squashed", then you would not have to fight with spurious conflicts.
>
>> +    if ($opt_k)
>> +    {
>
> This open brace come on the same line as "if", like "if (...) {" to match
> the style of the surrounding code.
>
>> +     my $orig_file ="$f.orig";
>> +     rename $f, $orig_file;
>> +     open(FILTER_IN, "<$orig_file") or die "Cannot open $orig_file\n";
>> +     open(FILTER_OUT, ">$f") or die "Cannot open $f\n";
>> +     while (<FILTER_IN>)
>> +     {
>> +         my $line = $_;
>> +         $line =~ s#\$Revision:[ \.\d]+ \$#\$Revision\$#;
>> +         $line =~ s#\$Id: [^\$]+\$#\$Id\$#;
>
> When there is no '/' in substitution or pattern, it is _far_ easier to
> read if you used the standard s/foo/bar/, not custom s#foo#bar#.

I'd used # as there where \'s to escape some of the meta-characters
and I find large numbers of \/'s hard to follow.

>
> Can "$Revision:" immediately be followed by a digit while "$Id:" must
> always be followed by a whitespace?  I doubt it.
>
> Why isn't this something like:
>
>        $line =~ s/\$(Revision|Id|Date|....):[^\$]+\$/\$\1\$/g;
>
> or even (not bothering to enumerate the possible set of keywords):
>
>        $line =~ s/\$([A-Z][a-z]+):[^\$]+\$/\$\1\$/g;

I bow to your superior regex formulation ;-)

I'll resend in a bit.

-- 
Alex, homepage: http://www.bennee.com/~alex/
CV: http://www.bennee.com/~alex/cv.php