[isabelle-dev] http://isabelle.in.tum.de/testboard/

[Alexander Krauss]

Quoting Makarius <makarius at sketis.net>:

> The end of Python vomiting has this:
>
> <class 'mpatch.mpatchError'>: patch cannot be decoded
>       args = ('patch cannot be decoded',)
>       message = 'patch cannot be decoded'
>
> This is the same dropout that we've had on  
> http://isabelle.in.tum.de/repos/isabelle last week.
>
> It could be an NFS corruption of the repository, or just the server  
> feeling too hot.  I've no real idea.

This seems to be happening regularly now, which is really very  
annoying. I do not know what the cause is. Since I'll be mostly  
offline starting from Saturday, here is how to fix it when it occurs.  
If it happens on the main repos, the mailing list should be contacted  
before attempting any manual repair:

* run "hg verify" on the repository. It shows some errors (that is,  
the repository is corrupted), and it also shows the revision number  
REV of the first damaged changeset.

* Strip the broken revisions (after making a backup):

   cp -a . /some/backup/location
   hg strip -n REV

(Note: This seems to fail with the newest version installed at some  
machines: here, strip also produces an exception. However, hg 2.1.1  
installed at lxbroy10 works).

* Re-push the changesests that got stripped from some other clone.

Googling the error message is rather inconclusive. After all, it is a  
repo corrruption, but we don't know how it arises and cannot reproduce  
it predictably.

Who pushed to testboard today? What machines
  were you using as gateway, i.e., what location did you push to (see  
.hg/hgrc [paths])?

Let's see if this occurs again with hte main repos, or if the recent  
re-coling eliminated the problem there.

Alex