"clean" filter breaks git-svn

"clean" filter breaks git-svn

[Matt Wozniski]

I'm playing around with smudge and clean filters, and I've discovered
that they seem to completely break git-svn.  When trying to fetch
commits A and B from the SVN repos, it will fetch A, and then clean(A)
is committed to my repository.  Then when it tries to fetch B, it is
horribly confused - it complains of a checksum mismatch, since the
md5sums of the files in A in the SVN repos don't match up with the
md5sums of the files in the clean(A) commit in git land.  Is this a
known problem?  And, are there any work arounds other than just not
using filters when using git-svn?  Could git-svn be made to accept the
md5sum of *either* A or clean(A) instead?

Thanks!

~Matt

PS - Please CC me if possible; I'm not subscribed

Re: "clean" filter breaks git-svn

[Jonathan Nieder]

Hi Matt,

Matt Wozniski wrote:

>                                              When trying to fetch
> commits A and B from the SVN repos, it will fetch A, and then clean(A)
> is committed to my repository.  Then when it tries to fetch B, it is
> horribly confused - it complains of a checksum mismatch, since the
> md5sums of the files in A in the SVN repos don't match up with the
> md5sums of the files in the clean(A) commit in git land.

Interesting.  Yeah, that sounds like a bug.

I am not convinced cleaning fetched files is the right thing to do in
the first place (why not just trust the SVN repo?), but I assume
people with workflows involving such filters could make a better call.

> Is this a
> known problem?

There is no test for it in git.git, so in that sense no.

>                                    Could git-svn be made to accept the
> md5sum of *either* A or clean(A) instead?

I don’t think that would be right.  It’s not just the md5sums not
matching that is the problem; it is that in these situations git-svn
and the SVN server do not agree about the file’s current content.

Once the code makes the semantics clear, I assume it should Just
Work™.

Good luck,
Jonathan

Re: "clean" filter breaks git-svn

[Junio C Hamano]

Jonathan Nieder <jrnieder@gmail.com> writes:

> Interesting.  Yeah, that sounds like a bug.
>
> I am not convinced cleaning fetched files is the right thing to do in
> the first place (why not just trust the SVN repo?),...

My knee-jerk reaction was that it isn't even about _trusting_, but
mirroring what happened on the other side of the world faithfully.

If you clean what you got from the other side, and then build on top of
that result, how are you supposed to send your contributions back to the
other side which does not have the same clean-up?