On 07/09/2014 03:38 AM, Nathan Scott wrote:
Hi mate,
I've pulled all your current work, but had some problems with a
few tests (see attached bad files). I found reverting the commit
below got us back into a healthy state, so pushed this for now -
when you figure it out, just "git revert" on my revert should do
a revert revert & then just apply your followup fix on top).
I don't think I have permission to do reverts on the dev branch of the
SGI repository, and I would be too afraid to mess it up anyway, so I
just redid the change in my pcpfans brolley/dev branch. That in addition
to the fix which was to add EINPROGRESS handling to a few callers of
__pmConnect().
Dave
----------------------------------
commit a705d6bb86d87fa0cdf0fb0accf5299212cef08d
Author: Dave Brolley <brolley@xxxxxxxxxx>
Date: Fri Jul 11 12:35:56 2014 -0400
Re-instate "Respect FNDELAY in __pmConnect() for secure connections."
This re-instates commit 75e40a1a033df79ca36b4034670bede4ba39ebdf
which was breaking tests 067, 110 and 156.
commit ff23f3d6ab3d5d82141761722afec591f15a4597 adds the required
error handling code to callers of __pmConnect.
commit ff23f3d6ab3d5d82141761722afec591f15a4597
Author: Dave Brolley <brolley@xxxxxxxxxx>
Date: Fri Jul 11 11:25:40 2014 -0400
Always handle EINPROGRESS from __pmConnect().
NSPR sockets appear to always have the FNDELAY flag set, so it
has always been possible for __pmConnect() to return -1 with
neterror() set to EINPROGRESS. Now that the NSPR implementation of
__pmConnect() uses PR_INTERVAL_NO_WAIT, this happens more often and
a few callers that did not handle this were exposed.
All callers of __pmConnect() now handle this situation.
|