Re: [PATCH] gcc 2.95 fixes

On Fri, 26 Mar 2004, Bryan Kadzban wrote:

> [email protected] wrote:
> > The attached patch makes gcc 2.95 happy.  Also it is just good
> > practice to declare variables at the top of their scope.  I'd
> > appreciate it if it was applied to CVS.
>
> It may make gcc happy to declare variables at the top, but it'll make
> the system much, much happier to initialize variables where they had
> been initialized before.  *ESPECIALLY* when that initialization is going
> to affect inter-thread synchronization...
>
> The second patch (the one in http.c) isn't going to affect anything, but
> the first one (in critsection.c) could very easily deadlock.
>
> It's much better to actually wait on the critical section's semaphore in
> the same place that it's being waited on in the current CVS code (i.e.,
> *after* the entry and contention counts are atomically incremented, not
> before -- otherwise, if the atomic increment affects the signaled state
> of the semaphore, then your patch could cause a deadlock).  Something
> like the attached is going to be a lot safer (plus it applies more
> cleanly to current CVS).

That's better than my patch.  Obviously mine was just a quick hack without bothering to see what all was going on.

-- 

<Douglas Leonard>
<[email protected]>