Message-Id:	<199905101950.OAA04852@indy1.indy.net>
From:	"Steve Snyder" <ssnyder AT indy DOT net>
To:	"PGCC Mailing List" <pgcc AT delorie DOT com>
Date:	Mon, 10 May 1999 14:49:23 -0400 (EST)
X-Mailer:	PMMail 2.00.1500 for OS/2 Warp 4.00
MIME-Version:	1.0
Subject:	PGCC fails more g++ tests than EGCS v1.1.2
Reply-To:	pgcc AT delorie DOT com

I've been playing with the Mandrake PGCC SRPMS (pgcc-1.1.3-3mdk.src.rpm -
thanks for the heads-up, Bernhard Rosenkraenzer) and comparing it to EGCS 
(RedHat's egcs-1.1.2-12.src.rpm).

On my heavily upgraded RH v5.2 system (glibc 2.1.1, kernel v2.2.7 and all 
the updates needed for the v2.2 kernel), running the testsuite produces 
more g++ errors under PGCC than under EGCS.

Before I get to the testsuites, I should also note that the PGCC compiler
won't build with RPM_OPT_FLAGS="-O6 -mcpu=i686 -march=i686" but will build
if the -march=i686 switch is removed.  The build fails with the errors that
limits.h can't be found.  Hmm...

Now on to the testsuites:

EGCS was built with RPM_OPT_FLAGS="-O6 -mcpu=i686 -march=i686"
PGCC was built with RPM_OPT_FLAGS="-O6 -mcpu=i686"

The results of the gcc and g77 tests are identical, but the g++ test has 4 
more errors under PGCC than under EGCS when run on my PII/450-based system.

  < PASS: g++.eh/spec1.C  Execution test
  > FAIL: g++.eh/spec1.C  Execution test

  < PASS: g++.law/arm13.C (test for excess errors)
  < PASS: g++.law/arm13.C  Execution test         
  > FAIL: g++.law/arm13.C (test for excess errors)
  > XFAIL: g++.law/arm13.C  Execution test        
                                                
  < PASS: g++.mike/eh49.C  Execution test         
  > FAIL: g++.mike/eh49.C  Execution test         
                                                
  < # of expected passes          4222            
  < # of expected failures          85      
  > # of expected passes          4218            
  > # of unexpected failures         3               
  > # of expected failures          86      

Thinking that I may have been overambitious, I dropped the optimization
down from -O6 to -O2, rebuilt the SRPMS and reran the tests.  This reduced
the number of unexpected failures by 1 (the spec1.c test), leaving 3
failures unaccounted for.

I also tried building the compiler and running the tests on an i586
machine, with RPM_OPT_FLAGS="-O6 -mcpu=i586" and again with
RPM_OPT_FLAGS="-O2 -mcpu=i586".  In both cases the results were the same as
I see on my i686 machine.

Is anyone else seeing this test discrepency between EGCS and PGCC?

Thank you.



*** Steve Snyder ***

- Raw text -