Re: Results for haifa-enabled egcs-2.92.07 19980914 (gcc2 ss-980609 e xperimental) testsuite on mips-sgi-irix6.2

Re: Results for haifa-enabled egcs-2.92.07 19980914 (gcc2 ss-980609 e xperimental) testsuite on mips-sgi-irix6.2

[Martin Knoblauch]

---"Billinghurst, David (RTD)" <David.Billinghurst@riotinto.com.au>
wrote:
>
> g++ results look horrible.  Problem is not immediately apparent to the
> untrained eye.

 Those fail due to the problem with a missing external symbol
"__vt$9type_info"/"type info virtual table" that I reported first
last week and re-reported yesterday. This is a clear regression
from egcs-1.1. It has to do with the fact that gcc now issues
a ".weakext" directive, where it issued a ".globl" before.
As a result the above symbolare no longer seen by other object
modules. For the testsuite, all tests linking to libcc/libstdc++
will fail.

 I will look into the problem a little further later this week.
It may be actually a problem of the SGI assembler or linker not
making ".weakext" symbols globally visible. But of course, it also
could be a problem in egcs :-)

> g77 failure is a new test

 This fails due to "recursive IO"


> Will provide more detail if required
> 
> Native configuration is mips-sgi-irix6.2
> 
> 		=== objc tests ===
>

 I actually wonder why you pass the objc suite. On my irix-6.5
configuration all the execution tests fail (something like "class
not found"). 

> 
> Running target unix
> 
> 		=== objc Summary ===
> 
> # of expected passes		38
> /usr/local/imports/BuildTree.irix/egcs-19980914/gcc/xgcc version
> egcs-2.92.07 19980914 (gcc2 ss-980609 experimental)
> 

> configure flags:
>
--with-gcc-version-trigger=/usr/local/imports/egcs-19980914/gcc/version.c
> --host=mips-sgi-irix6.2 --enable-shared --enable-haifa
> --prefix=/usr/local/tmp/egcs-19980914 --norecursion
> 
> 

 If you do not use "--enable-shared", all the g++ related problems
move from "execution" to "compilation", as the missing symbol is
detected earlier in a "static" link.

Martin
===
------------------------------------------------------
Martin Knoblauch
email: knobi@knobisoft.de or knobi@rocketmail.com
www:   http://www.knobisoft.de

_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com

Re: Results for haifa-enabled egcs-2.92.07 19980914 (gcc2 ss-980609 e xperimental) testsuite on mips-sgi-irix6.2

[Mark Mitchell]

Jason --

>>>>> "Martin" == Martin Knoblauch <knobi@rocketmail.com> writes:

    Martin>  Those fail due to the problem with a missing external
    Martin> symbol "__vt$9type_info"/"type info virtual table" that I

I think this problem stems from this change:

  1998-09-03  Jason Merrill  <jason@yorick.cygnus.com>

	* decl2.c (import_export_vtable): Always make artificials comdat.
	(import_export_decl): Likewise.
	* pt.c (mark_decl_instantiated): Likewise.

The MIPS assemblers don't seem to do `.weakext' for data; only for
functions.  For example, on IRIX 6.5:

	  .data
	  .weakext i
  i:
	  .word   0

yields, after assembly:

  Symbols from test.o:

  [Index]   Value      Size    Type  Bind  Other     Shndx   Name

  [1]     |         0|       0|SECT |LOCL |DEFAULT  |4      |.data
  [2]     |         0|       0|SECT |LOCL |DEFAULT  |8      |.debug_info
  [3]     |         0|       0|SECT |LOCL |DEFAULT  |9      |.debug_abbrev
  [4]     |         0|       0|SECT |LOCL |DEFAULT  |10     |.debug_frame
  [5]     |         0|       0|SECT |LOCL |DEFAULT  |11     |.rel.debug_info

Note the lack of `i'.  

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com

Re: Results for haifa-enabled egcs-2.92.07 19980914 (gcc2 ss-980609 e xperimental) testsuite on mips-sgi-irix6.2

[Martin Knoblauch]

Hi,

 Mark is probably right. Specifying "-fno-weak" when running "check"
seems to solve the problem.

 Now the question is whether it is a problem in the MIPSpro
assembler or in g++.

Martin

Mark Mitchell wrote:
> 
> Jason --
> 
> >>>>> "Martin" == Martin Knoblauch <knobi@rocketmail.com> writes:
> 
>     Martin>  Those fail due to the problem with a missing external
>     Martin> symbol "__vt$9type_info"/"type info virtual table" that I
> 
> I think this problem stems from this change:
> 
>   1998-09-03  Jason Merrill  <jason@yorick.cygnus.com>
> 
>         * decl2.c (import_export_vtable): Always make artificials comdat.
>         (import_export_decl): Likewise.
>         * pt.c (mark_decl_instantiated): Likewise.
> 
> The MIPS assemblers don't seem to do `.weakext' for data; only for
> functions.  For example, on IRIX 6.5:
> 
>           .data
>           .weakext i
>   i:
>           .word   0
> 
> yields, after assembly:
> 
>   Symbols from test.o:
> 
>   [Index]   Value      Size    Type  Bind  Other     Shndx   Name
> 
>   [1]     |         0|       0|SECT |LOCL |DEFAULT  |4      |.data
>   [2]     |         0|       0|SECT |LOCL |DEFAULT  |8      |.debug_info
>   [3]     |         0|       0|SECT |LOCL |DEFAULT  |9      |.debug_abbrev
>   [4]     |         0|       0|SECT |LOCL |DEFAULT  |10     |.debug_frame
>   [5]     |         0|       0|SECT |LOCL |DEFAULT  |11     |.rel.debug_info
> 
> Note the lack of `i'.
> 
> --
> Mark Mitchell                   mark@markmitchell.com
> Mark Mitchell Consulting        http://www.markmitchell.com

-- 
+---------------------------------+
|Martin Knoblauch                 |
|---------------------------------|
| http://www.knobisoft.de          |
+---------------------------------+
|e-mail: <knobi@knobisoft.de>     |
+---------------------------------+