Wrapper scripts to help git-cvsimport cope with cvs 1.12 breakage

From: Yann Dirson <ydirson@altern.org>
Date: Jul 3, 2007, 21:43 UTC
Message-ID: <20070703214336.GC6361@nan92-1-81-57-214-146.fbx.proxad.net>

Some of you may have noticed that cvs 1.12.x has a bad habit, when a file gets added on a branch, of creating a dummy 1.x.2.1 revision of the empty file on the branch. What's annoying, is that the dummy revision gets an erroneous timestamp, which makes the cvsps output wrong, following the well-known GIGO principle.

Until someone (maybe me, but I can give no guaranty whatsoever) finds the time to deal with this bug in cvsps, as well as avoiding at the cvsps level the extra "file initially added on branch" dummy revisions, here is a little perl filter that sanitizes the csvps output, and a customizable git-cvsimport wrapper using it.

Best regards,

-- 
Yann


#!/usr/bin/perl
use strict;
use warnings;

$/='---------------------';

our @branches = qw/HEAD/;
our %parents = ();		# parents not yet recorded

# parsing state
our $patchset_is_valid;
our $patchset;

while(<>) {
  $patchset = $_;

  # skip first separator line
  if ($patchset eq $/) {
    print $patchset;
    next;
  }

  $patchset_is_valid = 1;
  my ($branch, $ancestor);
  if (m/^Branch: (.*)/m) {
    $branch = $1;
    if (!grep { $_ eq $branch } @branches) {
      push @branches, $branch;
    }
  } else {
    die "no branch in $patchset";
  }

  if (m/^Ancestor branch: (.*)/m) {
    $ancestor = $1;
    if (!grep { $_ eq $ancestor } @branches or
	m/^file .* was added on branch /m) {
      print STDERR "preparing $branch fixup\n";
      $patchset_is_valid = 0;
      $parents{$branch} = $ancestor;
    }
  }

  if (m/^file .* was added on branch /m) {
    $patchset_is_valid = 0;
  }

  if (defined $parents{$branch} and $patchset_is_valid) {
    print STDERR "finalizing $branch fixup\n";
    my $line = 'Ancestor branch: ' . $parents{$branch} . "\n";
    $patchset =~ s/^(?=Tag:)/$line/m;
    delete $parents{$branch};
  }