Re: [squid-users] simple questions

Simon White wrote:
> 26-Mar-02 at 03:35, Joe Cooper (joe@swelltech.com) wrote :
>
>>Squid does not exhibit problems with any 2.96 GCC from Red Hat I have
>>used in any environment in which I have used it.
>>
>
> But how complete can your testing be? I was not particularly worried about
> 2.96 when I read that mplayer wouldn't compile on my workstation, because
> I only wanted mplayer to watch movies, which is hardly a work pastime.

I may not maintain more production Squid caches than anyone on this list
(Duane might handle more than me with the nlanr caches), but I wouldn't
bet on it. Admittedly, they are mostly used in the same type of
environment (ISP), but that environment is the hardest load I can think
of for a web cache. So, my testing is pretty complete--moreso than most.

> However, seeing database corruption in MySQL, I thought that the two
> together were enough to shed doubt, especially when reports all state that
> 2.95 works fine.
>
> Now let's look at this: http://gcc.gnu.org/gcc-2.96.html - the actual GNU
> response:-
>
> "Please note that both GCC 2.96 and 2.97 are development versions; we
> do not recommend using them for production purposes. Binaries built
> using any version of GCC 2.96 or 2.97 will not be portable to systems
> based on one of our regular releases."

Only true of C++ programs, AFAIK, due to changes in name mangling.
Portability is generally fine for C programs. The problem of name
mangling in C++ is universal--many compilers have had pains in that
area, and it is discussed in Stroustrup's Design and Evolution of the
C++ Programming Language (I just read that chapter yesterday ;-).

> Now, who still wants to compile Squid on a production high-load cache with
> 2.96?

Ooh, ooh! Me! Me! Me!

I've got some quite loaded boxes...doing just fine on Squids compiled on
2.96.

Not to be ornery about it, but I'm just not willing to take reactionary
anti-Red Hat FUD (not that you, Simon, are necessarily a purveyor of
FUD, though you are repeating what FUDders have said) at face value. I
watched the compiler stink as it developed way back when we were
sticking with RH6.2 for our systems--we stuck with 6.2 until several
months after 7.1 was released, partly due to what we perceived was a
compiler and compatibility issue with 7.x and gcc 2.96. I continued to
do testing on those new Red Hat versions, just didn't trust them enough
to deploy them. But, after a while I began to realize that I was /not/
seeing the problems that everyone claimed were rampant. My programs
built without a hitch, I saw no compatbility issues, and I never traced
any problems in Squid to a compiler problem. If such problems are so
rampant, I must have the luck of the Irish, because I'm not seeing them.

Quite frankly, I believe Bero's explanation about the whole
thing...which can be summed up as "It's not the compiler that is wrong."
  He makes a convincing case that the programs that fail on 2.96,
/ought/ to fail, because they are wrong. I'd link to it here, but I
haven't the foggiest notion where it is on Red Hat's page.

BTW-Robert mentioned OpenBSD and Squid's DNS code when compiled O2. I
believe this was a 2.95 variant of GCC...and in fact, was known to be
broken. But I don't recal the specifics of all of that.

Anyway, that's my .02. Everyone else may wish to remain with 2.95 for
all eternity. It works fine, too.

-- 
Joe Cooper <joe@swelltech.com>
http://www.swelltech.com
Web Caching Appliances and Support