Re: [Gm2] cvs latest changes and regression test results

[Michael Lambert]

Hi Gaius,

On 15 May 2009, at 15:05, Gaius Mulley wrote:

> below are the latest entries in the ChangeLog.  The latest regression
> test results are now (when run on Linux 2.6.27-9-generic #1 SMP x86_64
> GNU/Linux).
>
>                === gm2 Summary ===
>
> # of expected passes            7806
> # of unexpected failures        24
> # of unresolved testcases       6

I'm seeing some strange behavior on Debian Lenny/Alpha (Linux 2.6.18-6- 
alpha-generic #1 alpha GNU/Linux).

Regression results:

                === gm2 Summary ===

# of expected passes            7770
# of unexpected failures        72

One of the failures is in cycles:

$ gdb cycles.x0
GNU gdb 6.8-debian
Copyright (C) 2008 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html 
>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show  
copying"
and "show warranty" for details.
This GDB was configured as "alpha-linux-gnu"...
(gdb) run
Starting program: /home/lambert/gnu/obj.sav/gcc/testsuite/cycles.x0

Program received signal SIGSEGV, Segmentation fault.
0x00000200002a40a0 in ?? () from /lib/libc.so.6.1
(gdb) where
#0  0x00000200002a40a0 in ?? () from /lib/libc.so.6.1
#1  0x000002000029ebf4 in strtold () from /lib/libc.so.6.1
#2  0x000000012001735c in ldtoa_strtold (s=0x12004a790 "350",
    error=0x11f9af598) at ../../gcc-4.1.2/gcc/gm2/gm2-libs-ch/ldtoa.c: 
76
#3  0x000000012001ce7c in StringConvert_StringToLongreal (s=0x12004a6c0,
    address@hidden)
    at ../../gcc-4.1.2/gcc/gm2/gm2-libs/StringConvert.mod:790
#4  0x000000012001f7bc in FpuIO_StrToLongReal (a=
      {_m2_contents = 0x120025580 "350", _m2_high = 2}, address@hidden)
    at ../../gcc-4.1.2/gcc/gm2/gm2-libs/FpuIO.mod:94
#5  0x000000012001fc60 in _M2_cycles_init ()
#6  0x0000000120001884 in init ()
#7  0x0000000120001b9c in main ()
(gdb)

I could easily reproduce the problem in ldtoa.c and discovered that  
compiling ldtoa.c with the '--std=c99' switch caused things to work  
(I'm not sure whether that's a bug or a feature, but the manpage  
suggests the latter).  After adding that flag to all instances of  
ldtoa.c in Make-lang.in I get

                === gm2 Summary ===

# of expected passes            7776
# of unexpected failures        66

I'm looking at the other failures--it would appear that longstr and  
realstr segfault on (effectively) infinite recursion.

Regards,

Michael