Re: [PATCH] fast-export: Avoid dropping files from commits

From: Elijah Newren <newren@gmail.com>
Date: Mar 26, 2009, 04:02 UTC
Message-ID: <51419b2c0903252102v71de617aiffad2d4934d4d07f@mail.gmail.com>
In-Reply-To: <7vy6utt0op.fsf@gitster.siamese.dyndns.org>

Hi,

On Wed, Mar 25, 2009 at 9:36 PM, Junio C Hamano <gitster@pobox.com> wrote:

Show 11 quoted lines

> Hmm, does it?
>
> Shouldn't an export with a bottom commit be always considered an
> incremental?  Why special case --import-marks?
>
> When you say "I want to export master~2..master", isn't the intention
> (unstated, because it is too obvious) that follows it "... because I do
> have master~2 already and I would want to replay the export on top of that
> state"?  If all of master~2, master~1 and master have a file "frotz" with
> exactly the same contents, I thought you wouldn't even have to have that
> same contents repeated in the export datastream.

No, without the --import-marks it doesn't make sense to make it an incremental. Think of it this way: git-fast-export converts all sha1's into integer identifiers (marks). When you export master~2..master, then when fast-export comes across master~1, it notices that there's nothing to convert it's parent sha1sum into since no information was output for master~2 in the export stream. The way around it would be to simply export master~2 anyway, and master~3, and... (An alternate way around it, I guess, would be using the sha1sum in the export stream instead of a mark...but do you have any way of knowing whether the destination repository you import into will have that commit? Using --import-marks seems like a nice way to specify whether you have that information or not.)

It's perhaps a gotcha, as I know I had the same initial assumption even after reading the manpage. Perhaps Documentation/git-fast-export.txt could be made a bit clearer on this point.

> Or am I (again) entirely misunderstanding the intended use case?

Exporting master~2..master without specifying --import-marks is a way of squashing history, essentially. Perhaps you don't like it serving this purpose, but it was intentional; looking at the master~2..master testcase in t/t9301-fast-export.sh, you'll see

   git fast-export master~2..master | sed "s/master/partial/" |
   <snip some other boilerplate> &&  test_must_fail git rev-parse partial~2