[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH] lower cap on max_cmd_len to fix hppa2.0w-hp-hpux11.00
From: |
Michael Elizabeth Chastain |
Subject: |
Re: [PATCH] lower cap on max_cmd_len to fix hppa2.0w-hp-hpux11.00 |
Date: |
Mon, 9 Jul 2001 10:46:53 -0700 |
> The bug report at
> http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&database=gcc&pr=3055
> shows that the problem is intermittent, the reporter verifies that
> --with-system-zlib allows him to build gcc 3.0. I am assuming that
> ltconfig is being run for at least one other library without error.
That's a bad assumption. On my machine, zlib is the only library
in the 3.0 distribution which gets configured with ltconfig.
I don't think it's an intermittent problem at all.
> If the case was that every hp 11.0 machine (or nearly every one)
> couldn't use libtool, then I would advocate the change to work around the
> problem.
I've got one report from the field + it happens reliably on my
machine. That's two reported machines, which is enough for me
to want a workaround.
> The reason you don't have any change in performance when you lower
> this limit is that your project (gcc) doesn't build large libraries.
Again, no.
On these natives:
hppa1.1-hp-hpux10.20
i686-pc-linux-gnu
sparc-sun-solaris2.5
sparc-sun-solaris2.6
sparc-sun-solaris2.7
sparc-sun-solaris2.8
... ltconfig, right now, reports a limit of 512K or less. ltconfig
reports the same limit with my patch. You know this; I reported it in
my first message.
On this native:
hppa1.1-hp-hpux11.00
... your system reports 790K and my system craps out. I don't know what
the performance difference is between 790K and 390K (the value I get
after lowering the max limit). Even if this difference is substantial,
it is not worth the hassle of telling people to upgrade their system
software to run libtool.
> Talk with the HP support people, and let me know what they say. Deal?
No, I have lots of issues like this, and I can't spend N round-trips
in e-mail with the likely outcome that our customers have to apply one
patch per problem. It's much better for us if our source tree builds
on a wide variety of systems without patching the vendor software.
So I'm going to propose to our engineering group that we make a local
change to our source.
Michael Elizabeth Chastain
<address@hidden>
"love without fear"