httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: OPTIMIZE_TIMEOUTS
Date Wed, 11 Mar 1998 22:32:22 GMT
On Tue, 10 Mar 1998, Brian Behlendorf wrote:

> At 02:13 AM 3/10/98 -0800, Dean Gaudet wrote:
> >I've only looked at linux libc (libc5 and glibc2), and berkeley db-1.85 so
> >far... and they both have problems with EINTRs at the wrong time.  
> 
> Time to stop linking with libc then?  :)

Well, we can never make apache signal safe, and we can never safely use
longjmp() unless we know for certain that the libraries we link with
expect that sort of thing to happen anywhere.  i.e. we'll never have
reliable, bug free code. 

For that reason I'm in no big hurry to fix OPTIMIZE_TIMEOUTS, because even
with it fixed the code is broken.  But at least I'll make it as safe as
things were before.  i.e. subtly broken. 

And no, C++ exceptions don't help this one bit.  Just in case someone was
thinking about that. 

BTW, child_exit() exacerbates the problem.  Not only is it a redundant
routine that's not required (the functionality already exists via
cleanups); it suggests that we can safely exit a child after receiving a
signal... and that's nowhere near the case. 

Dean



Mime
View raw message