patchimport-tars: Allow per-tar author and commit message.

Instead of having each imported tar ball's commit message be "Imported from filename.tar", optionally take a commit message from a file called "filename.tar.msg".

Instead of having each commit have the same author and committer information, optionally read the committer information from a file called "filename.tar.committer" and author from a file called "filename.tar.author".

Signed-off-by: Peter Krefting <peter@softwolves.pp.se>
---
I used this (albeit based on a slightly earlier verison of the script)
to generate a better-looking history when importing
http://git.debian.org/?p=crashmail/jamnntpd.git and
http://git.debian.org/?p=crashmail/crashmail.git into Git, using
excerpts from the embedded change history as commit messages.

 contrib/fast-import/import-tars.perl |   42 +++++++++++++++++++++++++++++++--
 1 files changed, 39 insertions(+), 3 deletions(-)

Quoting Peter Krefting <peter@softwolves.pp.se>

Instead of requiring the user to have millions of separate files, how about reading a single metainfo file that may look like this?

    [file "git-1.0.0.tar"]
	message = "Commit log message for the first revision"
	author = "A U Thor <au.thor@example.xz>"

    [file "git-1.0.1.tar"]
	message = "Commit log message for the first maintenance revision"
	author = "F I Xer <fi.xer@example.xz>"

and give the name of that metainfo file from the command line? The message may be awkward to put in the metainfo file itself, and it might be easier to refer to the name of a file that is outside the metainfo file itself, but the point is that doing so will both reduce the clutter in your folders, and will give people an option to avoid accidentally reading from random files in the same folder as the tar files that unfortunately have names that ends with the suffixes you randomly chose in this patch.

Peter Krefting wrote:

It's not necessary to duplicate code like that.

Also I wonder if there isn't a nicer interface for users. Why not allow the file to specify From:, Committer: etc as header lines

	# Optionally read a commit message from <filename.tar>.msg
	my $commit_msg = "Imported from $tar_file.";
	my ($committer, $commit_date, $author, $author_date);
	if (open MSG, '<', "${tar_file}.desc")
	{
		my $state = "header";
		$commit_msg = '';
		my $last_val;
		while (<MSG>)
		{
			if ($state eq "header") {
				if (m{^([A-Z][\w\-]+): (.*)}) {
					my ($header, $value) = $1;
					$last_val = ($header eq "Date" ? \$author_date :
						$header eq "From" ? \$author :
						$header eq "Subject" ? \$commit_msg :
						$header eq "Commit-Date" ? \$commit_date :
						$header eq "Committer" ? \$committer :
						undef );
					if ($last_val) {
						$$last_val = $value;
					}
					else {
						warn "ignoring header '$header' in ${tar_file}.desc line $.\n";
					}
				}
				elsif (m{^\s+(.+)}) {
					if ($last_val) {
						$$last_val .= " $1";
					}
				}
				elsif (m{^\s*$}) {
					$commit_msg .= ($commit_msg ? "\n" : "") . "\n";
					$state = "body";
					next;
				}
			}
			if ($state eq "body") {
				$commit_msg .= $_;
			}
		}
		close MSG;
	}

	($this_committer_name, $this_committer_email) =
		choose_email($committer, $committer_name, $committer_email);
	($this_author_name, $this_author_email) =
		choose_email($author, $author_name, $author_email);

	sub choose_email {
		my ($spec_combined, $def_name, $def_email) = @_;
		return ($def_name, $def_email) unless $spec_combined;
		if ($spec_combined =~ m{^([^<]+) <([^@]+@[^@]+)>$})) {
			($1, $2);
		}
		elsif ($spec_combined =~ m{^([^@]+@[^@]+) \((.*)\)$}) {
			($2, $1);
		}
		else {
			warn "Couldn't parse e-mail address '$spec_combined'";
			($def_name, $def_email);
		}
	}

Something like that, anyway...

Sam

Nanako Shiraishi:

The advantage of having one (or more) files per tarball is that I can use the shell's tab completion to make sure I spell the name of the archive correctly. And that it makes it a lot easier to parse.

Sam Vilain:

Good idea. My initial patch was simple, just using the .msg file to add a message. Then I needed to have a different author for one commit, so I added the .author file. Then came the changes that are in the current import-tars that separates author and committer, hence the .committer file. Somewhere along there it became a bit too complex.

Parsing the .msg file for (optional) committer and author info is probably better. I'll update the patch.

Peter Krefting <peter@softwolves.pp.se> writes:

That sounds like a sensible thing to do.

I however am a bit uneasy to see that the patch didn't seem to (and the rerolled one does not seem to, either) allow any way to forbid reading of the .msg files, other than an obvious workaround of running "find | xargs rm" beforehand. Unlike your "import-directories" that is a brand new program without any existing users, you are touching code that other people have already used, and you do not want to change the behaviour for them only because they happen to have unrelated files in the same directory.

Junio C Hamano:

Indeed. Not that it is likely that one have stray filetar.gz.msg files just laying around, but I'll add a command line switch to enable the new functinality. That sounds like the most reasonable way to go, leaving the old usage completely unaffected by the change.

Peter Krefting <peter@softwolves.pp.se> writes:

And the switch could be "--metainfo=<ext>", so that people can choose to use other extensions, e.g. with "--metainfo=info" file.tar.info would be read for descriptions.

Junio C Hamano:

That's a good idea. I just made a simple "-m" switch to enable the new code. I'll change it into a "--metainfo" and post an updated patch later.