pycurl.error: (18, 'transfer closed with outstanding read data remaining')

[Wouter van Heyst]

I'm not entirely certain yet where the problem lies, I started getting this error after submitting to pqm "Better connection sharing for pycurl" (http://bundlebuggy.aaronbentley.com/request/%<email address hidden>%3E)

So, this could be a problem exposed by that bundle, and/or result from me trying to pull when pqm was processing a merge request (and then some window within that, if it would fail everytime there was a merge going on we could never pull).

I'm attaching the bzr.log of when I was doing a pull in bzr.0.12, will try to determine if a difference is made between using bzr.dev and bzr-0.12 doing the pull. Also pycurl vs urllib is worth a shot.

[Wouter van Heyst (larstiq)]

Some more toying around, and it has nothing to do with PQM.

Doing
   bzr pull -v http+pycurl://code.aaronbentley.com/bzr/bzrrepo/nested-trees/

fails, whilst

   bzr pull -v http+urllib://code.aaronbentley.com/bzr/bzrrepo/nested-trees/

succeeds with current bzr.dev. vila can also reproduce it and is looking into it.
Branches that have no new revisions will not trip you up.

[Vincent Ladeuil (vila)]

Mea ultima culpa.

I goofed while solving the conflicts in the last merge from bzr.dev (ironically the conflict was with my own modification for the robust range handling).

[Vincent Ladeuil (vila)]

Marked critical as the current bzr.dev tip (2157) will make bzr pull http+pycurl://whatever fail.
A workaround is to either apply the patch or use bzr http+urllib://whatever

[Vincent Ladeuil (vila)]

Larstiq, look, it's neither your fault or mine !

It's the test suite !

The attached patch add the tests that should have failed.

[Wouter van Heyst (larstiq)]

The fix actually went into bzr.dev, not 0.13, which didn't suffer the problem anyway.