On Fri, Sep 15, 2017 at 5:53 PM, Jonathan Nieder <jrnie...@gmail.com> wrote:
> Jason Merrill wrote:
>> On Fri, Sep 15, 2017 at 1:52 PM, Jonathan Nieder <jrnie...@gmail.com> wrote:
>> > Jason Merrill wrote:
>
>>>> Subject: Fix merge parent checking with svn.pushmergeinfo.
>>>>
>>>> Without this fix, svn dcommit of a merge with svn.pushmergeinfo set would
>>>> get error messages like "merge parent <X> for <Y> is on branch
>>>> svn+ssh://gcc.gnu.org/svn/gcc/trunk, which is not under the git-svn root
>>>> svn+ssh://ja...@gcc.gnu.org/svn/gcc!"
>>>>
>>>> * git-svn.perl: Remove username from rooturl before comparing to branchurl.
>>>>
>>>> Signed-off-by: Jason Merrill <ja...@redhat.com>
>>>
>>> Interesting.  Thanks for writing it.
>>
>> Thanks for the review.
>>
>>> Could there be a test for this to make sure this doesn't regress in
>>> the future?  See t/t9151-svn-mergeinfo.sh for some examples.
>>
>> Hmm, I'm afraid figuring out how to write such a test would take
>> longer than I can really spare for this issue.  There don't seem to be
>> any svn+ssh tests currently.
>
> Well, could you give manual commands to allow me to reproduce the
> problem?
>
> Then I'll translate them into a test. :)

Something like this:

git svn clone -s svn+ssh://user@host/repo
git config svn.pushmergeinfo yes
git checkout -b branch origin/branch
git merge origin/trunk
git svn dcommit

Thanks!

> FWIW remove_username seems to be able to cope fine with an http://
> URL.  t/lib-httpd.sh starts an http server with Subversion enabled,
> as long as the envvar GIT_SVN_TEST_HTTPD is set to true.  Its address
> is $svnrepo, which is an http URL (but I don't see a username in the
> URL).  Does that help?

I think the http transport handles the username separately, not in the URL.

I would expect that a dummy ssh wrapper like some of the tests use
would be sufficient, no need for an actual network connection.

> Alternatively, does using rewrite-root as in t9151-svn-mergeinfo.sh
> help?

Hmm, I'm not sure how rewriteRoot would interact with this issue,
whether it would be useful as a workaround or another problematic
case.

Jason

Reply via email to