Johan Corveleyn wrote on Wed, 29 Apr 2020 12:23 +0200:
> On Wed, Apr 29, 2020 at 12:11 PM Daniel Shahaf <d.s_at_daniel.shahaf.name> wrote:
> > Johan, what do you mean by 'a very old format which isn't supported well
> > by "git svn"'?
> For instance an svn repository that was created by svnadmin 1.0, with
> BDB backend. Or a pretty old version of FSFS or something like that.
> > How would you tell, over the svn:// protocol, whether
> > a repository is of a 'very old format'? I'd expect the server to hide
> > these details from the client — assuming the version of svnserve is new
> > enough, at least. [Well, one thing the server won't hide is the
> > "mergeinfo" capability, which depends on the backend format number, but
> > that's probably not related to the case at hand.]
> I assumed, because Chaya said the "SVN project repo is actually file
> based", that she would have control over it as an admin, and perhaps
> would know more about its history and / or exact version, setup, etc.
> But maybe I was a bit too terse there, so thanks for chiming in and
> adding some concrete commands which could give us some more
> information :-).
I don't think you were too terse; I just didn't understand what failure
mode your recommendation to upgrade the repository was meant to rule out.
In the meantime I've managed to eyeball all callsites of the error
message in the subject on our end, and I see three scenarios (for the
specific case of git-svn over svn://):
- if the repository uses the BDB backend (+ other conditions that I haven't determined)
- during a 'commit' operation
- during a 'blame' operation (on a 1.1.0 (sic) or newer server)
Given this, the advice given upthread applies.
> > In any case, Chaya, could you run `svnserve --version -q` and `svnadmin
> > info /path/to/repos` and post the output here? In addition to Johan's
> > direction, there's a svn_txdelta_parse_svndiff() call in BDB that I'd
> > like to rule out. In fact, I wish I had a backtrace of the error (it's
> > generated by us: see subversion/libsvn_delta/svndiff.c:write_handler),
> > but I don't even know whether to ask you to debug svnserve(8) on the
> > server or svn(1) or git-svn(1) on the client… as Johan said, we really
> > need the git-svn folks to look into this first and come back to us with
> > a reproduction recipe in terms of our APIs.
> > Also, could you try creating an svnsync mirror of the repository, and
> > let us know whether that succeeded or errored?
> > Cheers,
> > Daniel
Received on 2020-04-29 12:42:20 CEST