Re: [naviserver-devel] nsthreadtest fails on both Linux and Windows

[Stephen <yo...@gr...>]

On Wed, Oct 15, 2014 at 8:24 PM, Andrew Piskorski <at...@pi...> wrote:
>
> On Wed, Oct 15, 2014 at 03:45:59AM -0400, Andrew Piskorski wrote:
>
> > In my own application on Windows, my connection thread tells a worker
> > thread to do something using nsv, mutexes, and a condition variable.
> >
> > This is old code that works fine on AOLserver 4.0.x (on a different
> > older Windows XP box), but now with Naviserver on Windows 7, the
> > worker thread seems to never wake up, and never do anything at all.
>
> There is definitely a real problem.  For debugging purposes in my
> application, I replaced that one use of ns_cond (there are others)
> with a simpler sleep-and-poll arrangement.  That works - the submit
> thread wakes up after sleeping, sees the data waiting for it in its
> nsv queue, and does its job.
>
> So I don't know what exactly, but something is seriously broken with
> ns_cond and/or ns_mutex in Naviserver, at least on Windows, possibly
> on Linux as well.
>
> Btw, for those who haven't used ns_cond before, it is a low-level
> wrapper around standard POSIX condition variable semantics.  By design
> it MUST be used in conjunction with ns_mutex.
>
> Thus I'm extra interested in the way nsthreadtest crashes, which I
> don't understand.  But failing feedback on that, next I'll probably
> try and write some simpler stand-alone Tcl code to reproduce the "My
> thread waiting on ns_cond never wakes up" behavior I'm seeing with
> Naviserver.

You have a known good revision: aolserver-4.0.10, a known bad: tip,
and an automated test to distinguish between the two states:
nsthreadtest. Try 'hg bisect ...' to figure out where the error was
introduced.