From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 14022 invoked by alias); 3 Oct 2002 04:38:05 -0000 Mailing-List: contact gcc-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Archive: List-Post: List-Help: Sender: gcc-owner@gcc.gnu.org Received: (qmail 14009 invoked from network); 3 Oct 2002 04:38:04 -0000 Received: from unknown (HELO inet1.ywave.com) (65.161.32.36) by sources.redhat.com with SMTP; 3 Oct 2002 04:38:04 -0000 Received: from there (dsl5097.ywave.com [208.10.179.97]) by inet1.ywave.com (Postfix) with SMTP id DE5952CC1E; Wed, 2 Oct 2002 21:36:50 -0700 (PDT) Content-Type: text/plain; charset="iso-8859-1" From: Tim Prince Reply-To: tprince@computer.org To: "Kaveh R. Ghazi" , gcc-bugs@gcc.gnu.org, gcc@gcc.gnu.org Subject: Re: Profiling on mips-irix6? (Testcase gcc.dg/nest.c failure) Date: Wed, 02 Oct 2002 21:54:00 -0000 Cc: David.Billinghurst@riotinto.com, echristo@redhat.com References: <200210021422.KAA10736@caip.rutgers.edu> In-Reply-To: <200210021422.KAA10736@caip.rutgers.edu> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Message-Id: <20021003043650.DE5952CC1E@inet1.ywave.com> X-SW-Source: 2002-10/txt/msg00143.txt.bz2 On Wednesday 02 October 2002 07:22, Kaveh R. Ghazi wrote: > When running the testsuite on mips-sgi-irix6.5, I get a failure in > gcc.dg/nest.c: > > http://gcc.gnu.org/ml/gcc-testresults/2002-09/msg01040.html > > > FAIL: gcc.dg/nest.c (test for excess errors) > > WARNING: gcc.dg/nest.c compilation failed to produce executable > > Its not just me: > http://gcc.gnu.org/ml/gcc-testresults/2002-09/msg01050.html > > This test passes in the -pg flag. Looking at gcc.log, I see: > > ld32: FATAL 9 : I/O error (/usr/lib32/mips3/gcrt1.o): No such file or > > directory collect2: ld returned 32 exit status > > I can't find gcrt1.o anywhere. Its not already installed on my system > nor is it provided by and built by gcc itself. As I used Irix-6.5 in a previous cloak, I'll take a stab: -pg was not the way to do profiling on Irix-6.5; I'm not surprised the support libraries are missing. It was done by building with -g and using speedshop. > > So before I proceed I was wondering whether there is some place to get > a gcrt1.o and/or libprof1.a for irix6. > If not, I can open a PR/change-request for someone to provide one, How about symlinking gcrt0.o to gcrt1.o, for example? > and > mark the testcase to only assemble on irix, not compile & run. I've noticed that most cygwin testers are getting a link failure; a different problem, caused by the typical (full) installation not supporting -pg without over-riding the library search order. Simplest cure there is to un-install the alternate mingw runtime. I doubt it's feasible for gcc maintainers to fix target-specific things like this. But you're right, this test fails for target-specific reasons which aren't the fault of gcc. Apparently, it was meant specifically to diagnose a power-pc specific problem; shouldn't that be noted without requiring web research, for example, by making it run only on targets where it is relevant? -- Tim Prince