[MERGE] Fix bug #225020 by catching the CURLE_SEND_ERROR error more broadly.

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Vincent Ladeuil wrote:
> So that bug want to play more...
> 
> The previous fix was too restrictive.
> 
> This one catch additional cases and should be final.
> 
> I filed a bug upstream against libcurl for which a patch has been
> proposed.
> 
> I tested the upstream patch against bzr.dev without my previous
> patch and with this patch. This upstream patch fixed the bug.
> 
> Yet, this patch implements a workaround that fix our problem
> whether or not the upstream patch is applied.
> 
> To summarize:
> 
> - bzr without any patch, libcurl-7.18.0 without patches: 23 tests failing,

Is that without any patch, or without your *earlier* (403) patch? I wonder if
we should at least check that we actually got a 4XX error, rather than
assuming so. For example, what if we got a 200 OK, but the connection died
(network went down).

Anyway, I'm okay with this patch, but:

BB:tweak


> - bzr  without any patch, libcurl-7.18.0 with upstream patch: pass
> - bzr with this patch, libcurl-7.18.0 without patches: pass
> - bzr with this patch, libcurl-7.18.0 with upstream patch: pass
> 
> I'll update bug #225020 once I get feedback from upstream about
> which curl versions will include their fix.
> 
>    Vincent
> 
> 

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFIuBBaJdeBCYSNAAMRAojGAKDFQ65MuGlY9MnoogvM1FNcNWnt0wCgpcLU
g6IErMOCCteWdKRwKEZ7kdE=
=JU43
-----END PGP SIGNATURE-----