Re: SourceForge migration has broken HTTP access to some hg repos

[Olaf Till]

On Tue, Feb 20, 2018 at 12:42:46PM -0800, Mike Miller wrote:
> On Tue, Feb 20, 2018 at 10:58:21 -0800, Mike Miller wrote:
> > I am unable to pull hg changes from 3 of the Octave Forge mercurial
> > source repositories over HTTP. I *am* able to pull over SSH. I think
> > it's likely that the repositories are fine, but something is wrong with
> > the HTTP hosting of hgweb, probably having to do with the recent
> > migration.
> 
> Looking into this a little deeper, the problem only affects hg
> repositories that have bookmarks. These are the 4 repositories geometry,
> image, signal, and statistics.
> 
> I don't know much about hg internals, but I do know how to trace HTTP
> transactions. The hg pull issues a few GET commands, and this sequence
> is getting weird results back from the SourceForge server. Using curl to
> mimic what hg is doing:
> 
>     $ curl -H "vary: X-HgArg-1" -H "X-HgArg-1: namespace=bookmarks" 
> "http://hg.code.sf.net/p/octave/signal/?cmd=listkeys";
>     @ 46ba861187455a5414bd15d671664dd210b27e09
> 
>     $ curl -H "vary: X-HgArg-1" -H "X-HgArg-1: namespace=phases" 
> "http://hg.code.sf.net/p/octave/signal/?cmd=listkeys";
>     @ 46ba861187455a5414bd15d671664dd210b27e09
> 
>     $ curl -H "vary: X-HgArg-1" -H "X-HgArg-1: namespace=phases" 
> "http://hg.code.sf.net/p/octave/signal/?cmd=listkeys";
>     publishing        True
> 
> Notice that the last two commands are identical, but the second call
> returns a different (correct) value. Maybe there is some server-side
> caching going on here.

Our mails crossed...

Maybe you could put the above information directly into the tracker at
the address I posted?

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: PGP signature