Re: [PATCH 2/5] svn-fe: allow svnadmin instead of svnrdump in svn helper

From: Dmitry Ivankov <divanorama@gmail.com>
Date: Jul 6, 2011, 11:02 UTC
Message-ID: <CA+gfSn-mX-Uny1dy=1Xwo42r_pTYosWJibNGafV=sLrsmc-e+A@mail.gmail.com>
In-Reply-To: <CALkWK0=Lqg4vMOdD8zah5RM3GOkBorJ80kBfs4BHEjZrG+rNVg@mail.gmail.com>

On Wed, Jul 6, 2011 at 4:12 PM, Ramkumar Ramachandra <artagnon@gmail.com> wrote:

Show 15 quoted lines

> Hi,
>
> Dmitry Ivankov writes:
>> svnrdump produces the same stream as svnadmin dump. svnrdump is able
>> to do it via any svn remote protocol while svnadmin needs access to
>> the repository filesystem and can't produce subdirectory dumps.
>> But svnrdump is a newer tool and may be unavailable on some systems.
>>
>> Try to use svnadmin dump for file:// repository urls if there is no
>> svnrdump in the PATH. First of all this is to be used in tests, where
>> the repository is indeed local most of the time.
>
> svnrdump's output is not identical to svnadmin's output.  You'd
> perhaps want to document the differences, and the impact they might
> have?

Hm, I don't see much differences.
Data format is the same SVN-fs-dump-format-version: 3. In my case I
observe following differences:
- svnadmin writes sha1 checksums along with md5 ones
- hashtable dumps can have different elements dump order
- svnrdump sets Prop-delta: true for empty props
- text deltas can be encoded differently (does it affect svn:ann?)
My svnadmin is version 1.6.16 (r1073529)
svnrdump is from r1135490

Anything else I'm missing in a simple dump of a repository root?

Show 19 quoted lines

>
>> diff --git a/contrib/svn-fe/git-remote-svn-alpha b/contrib/svn-fe/git-remote-svn-alpha
>> index 61c9b07..84d841e 100755
>> --- a/contrib/svn-fe/git-remote-svn-alpha
>> +++ b/contrib/svn-fe/git-remote-svn-alpha
>> @@ -12,10 +12,32 @@ usage () {
>>        exit 129
>>  }
>>
>> +try_svnrdump () {
>> +       command -v svnrdump >/dev/null &&
>> +       echo "svnrdump dump --non-interactive --username=Guest --password= \
>> +               --quiet --incremental" ||
>> +       true
>> +}
>
> I saw this '--username', '--password' in other patches too.  I'm
> probably missing context, but I'm curious to know why this is required
> and what you're planning to do to fix it in the future.

For repos with no read restrictions this isn't necessary, like those in the test script supplied. It's a placeholder for quick local patch if anyone wants to test a repo with permissions. In future it should be allowed to configure credentials in corresponding remote section of git.config and pass all these options to helpers. Using ~/.subversion/ configuration is a valid option too.

Show 5 quoted lines

>
> Thanks.
>
> -- Ram
>