Re: GCC updated in NetBSD!

[Fernando Oleo Blanco <irvise_ml@irvise.xyz>]

Am Sat, 06 Nov 2021 21:02:48 +0000
schrieb Simon Wright <simon@pushface.org>:

> Fernando Oleo Blanco <irvise_ml@irvise.xyz> writes:
> 
> >                 === acats tests ===
> > FAIL:   ce3404c
> >
> >                 === acats Summary ===
> > # of expected passes            6959
> > # of unexpected failures        1
> > Native configuration is x86_64--netbsd
> 
> That's three times* the number of test cases. I noticed this before --
> are you doing three parallel executions (the part of the script that
> runs parallel executions is particularly gnarly)? Maybe only one of
> the ce3404c's failed?
> 
> * more or less. Just ran for gcc version 12.0.0 20211021 with -j4,
> 
> 		=== acats Summary ===
> # of expected passes		2328
> # of unexpected failures	0
> 
> I don't think the ACATS in GCC reports all the outcomes the way it
> should: in particular, there were a considerable number of
> 'unsupported tests' (e.g. those where the test reports
> 'NOT-APPLICABLE'), and of course this was a .0.0 (work-in-progress)
> release. Perhaps x86_64-apple-darwin15 has one more unsupported test
> than x86_64--netbsd.

I have good news, but first things first.

Yes, I ran the tests in parallel, to be exact, with -j3. I am using the
pkgsrc infra, which means that I am not using gcc infra directly, which
could lead to errors.

The first run I believe did not generate any errors. I tried a second
time and the results are what I posted. I tried a third time and it was
chaos. All the times I ran the tests in parallel, they seemed to run
three times, instead of in parallel.

I decided to rebuild the entire thing from scratch and run the test in
serial (-j1). Annnnnddddd....... Here are the results!

GCC 10.3.0 native ACATS NetBSD-9.99.92 (serial execution)

                === acats tests ===

                === acats Summary ===
# of expected passes            2320
# of unexpected failures        0
Native configuration is x86_64--netbsd


                === gnat tests ===


Running target unix
FAIL: gnat.dg/align_max.adb execution test

                === gnat Summary ===

# of expected passes            3297
# of unexpected failures        1
# of expected failures          23
# of unsupported tests          9
/usr/pkgsrc/wip/gcc10-aux/work/build/gcc/gnatmake version 10.3.0

Much better!!!! I think I was right to say that the previous failing
tests where not because of our doing (or lack thereof), but because of
the compiler/OS.

I also ran ACATS 4.1X. Here are the results:

                === acats Summary ===
# of expected passes            2520
# of unexpected failures        24
# of expected failures          1487
# of unresolved testcases       11
# of unsupported tests          124
*** FAILURES: c324006 c350a01 c35503d c35503f c452003 c452005 c452006
c611a04 c650b04 c760a02 cdd2b03 cdd2b04 cxai001 cxai010 cxaia01 cxaib05
cxaib06 cxaib08 cxd1003 cxd1004 cxd1005 cxd2006 cxd3001 cxd3002
/home/fernando/ada_test/ACATS/run_all.sh completed at Sat Nov  6
19:45:05 CET 2021


Much better too! For comparison, here are the GCC 10.3 ACATS 4.1W
results in NetBSD 9.2

                 === acats Summary ===
# of expected passes            2516
# of unexpected failures        28
# of expected failures          1484
# of unresolved testcases       11
# of unsupported tests          124
*** FAILURES: c324006 c350a01 c452003 c452005 c452006 c52103x c52104x 
c52104y c611a04 c650b04 c760a02 cb1010a cb1010c cb1010d cdd2b03 cdd2b04 
cxai001 cxai010 cxaia01 cxaib05 cxaib06 cxaib08 cxd1003 cxd1004 cxd1005 
cxd2006 cxd3001 cxd3002
/home/fernando/acats/ACATS/run_all.sh completed at Wed Oct 20 22:13:57 
CEST 2021

We are getting 4 less failures (with c452003 know to stall gcc), which
is always nice :D

I think my work is mostly done. Jay Patelani has also successfully
built GCC 10.3 in NetBSD 9.2; racoon from IRC has also built GCC 10.3
in NetBSD 9.99.92, so great! Another user from the ML repoted issues
with RELRO in the Ada code using NetBSD 9.99.88, but with the patch 

diff --git a/gcc10-aux/Makefile b/gcc10-aux/Makefile
index 88cc2dc1ea..4b122d83d1 100644
--- a/gcc10-aux/Makefile
+++ b/gcc10-aux/Makefile
@@ -24,6 +24,8 @@ EXTRACT_ONLY=         ${DEFAULT_DISTFILES}
 # Relocations result in a linker error on AArch64, but not x86.
 MKPIE_SUPPORTED=       no

+RELRO_SUPPORTED=       no
+
 NOT_FOR_PLATFORM=      Interix-*-*

 USE_TOOLS+=            gmake makeinfo sed:run tar:build

He made it compile. I think this issue if mostly due to his older
system. I think I heard that there were some issues with RELRO in
-current, which seem to have already been fixed in newer version.

Regards,
Fer