RE: [PATCH] Fix PR 32915 dllimport regression

RE: [PATCH] Fix PR 32915 dllimport regression

[Danny Smith]

Wrong subject heading.  The PR is 32915


> Hello
> 
> The problem here is that dllimported symbols generated by the compiler
> have NULL as a DECL_ASSEMBLER_NAME, so the comparision with the alias
> decl_assembler_name_equal segfaults.
> 
> This regression is fallout Richard Henderson's rewrite of dllimport,
> exposed here by the use of weakref in libgfortran
> 
> (As an aside why are you using a pthread_once weakref on mingw32?)
> 
> This fixes by giving the artifical dllimport decl an 
> assembler name that
> matches the identifier of the SYMBOL_REF. 
> 
> 
:ADDPATCH i386:
> 
> 2007-07-28  Danny Smith <dannysmith@users.sourcforge.net>
> 
> 	PR target/32915
> 	* config/i386/i386.c (get_dllimport_decl): 
> SET_DECL_ASSEMBLER_NAME
> 	of the generated dllimport decl. 
> 
> 
> Index: config/i386/i386.c
> ===================================================================
> *** config/i386/i386.c	(revision 126996)
> --- config/i386/i386.c	(working copy)
> *************** get_dllimport_decl (tree decl)
> *** 7611,7616 ****
> --- 7604,7610 ----
>     set_mem_alias_set (rtl, ix86_GOT_alias_set ());
>   
>     SET_DECL_RTL (to, rtl);
> +   SET_DECL_ASSEMBLER_NAME (to, get_identifier (name));
>   
>     return to;
>   }
> 

Re: [PATCH] Fix PR 32915 dllimport regression

[Paul Thomas]

Danny,
> Wrong subject heading.  The PR is 32915
>   
I am so pleased to see that somebody else does that!

OK

Paul
>
>>
>> (As an aside why are you using a pthread_once weakref on mingw32?)
>>
>>     
I don't know - FX, any ideas?

Re: [PATCH] Fix PR 32915 dllimport regression

[FX Coudert]

>> 2007-07-28  Danny Smith <dannysmith@users.sourcforge.net>
>>
>> 	PR target/32915
>> 	* config/i386/i386.c (get_dllimport_decl):
>> SET_DECL_ASSEMBLER_NAME
>> 	of the generated dllimport decl.

Thanks a lot for providing a patch so quickly!

>> (As an aside why are you using a pthread_once weakref on mingw32?)

It appears on my build for mingw + pthreads-win32; libgfortran/io/ 
io.h includes gthr.h, which in turns include gthr-posix.h. The only  
functions used are then __gthread_mutex_lock and  
__gthread_mutex_unlock, to make the I/O library (and the random  
number generator) thread-safe (courtesy of Jakub Jelinek).

FX

RE: [PATCH] Fix PR 32915 dllimport regression

[Danny Smith]

> From: François-Xavier Coudert
> Sent: Wednesday, 12 September 2007 4:43 a.m.
> 
> > Thanks for the clarification.  I wasn't sure whether my authority
> > extended to such parts of files like i386.[ch].
> 
> ping?

I'm not 100% that my patch is "the right thing" . Shouldn't
decl_assembler_name_equal 
test for a NULL_TREE (or at least gcc_assert (decl_asmname !=
NULL_TREE))?

Danny
 
Index: tree.c
===================================================================
*** tree.c	(revision 128359)
--- tree.c	(working copy)
*************** decl_assembler_name_equal (tree decl, tr
*** 312,317 ****
--- 312,320 ----
  {
    tree decl_asmname = DECL_ASSEMBLER_NAME (decl);
  
+   if (decl_asmname == NULL_TREE)
+      return false;
+ 
    if (decl_asmname == asmname)
      return true;
  

> 
> FX
> 

Re: [PATCH] Fix PR 32915 dllimport regression

[François-Xavier Coudert]

> Thanks for the clarification.  I wasn't sure whether my authority
> extended to such parts of files like i386.[ch].

ping?

FX

RE: [PATCH] Fix PR 32915 dllimport regression

[Danny Smith]

> -----Original Message-----
> From: gcc-patches-owner@gcc.gnu.org 
> [mailto:gcc-patches-owner@gcc.gnu.org] On Behalf Of Uros Bizjak
> Sent: Friday, 24 August 2007 7:04 a.m.
> To: François-Xavier Coudert
> Cc: gcc-patches; Richard Henderson; jh@suse.cz; Danny Smith
> Subject: Re: [PATCH] Fix PR 32915 dllimport regression
> 
> 
> François-Xavier Coudert wrote:
> > > 
> > > 2007-07-28  Danny Smith <dannysmith@users.sourcforge.net>
> > > 
> > > 	PR target/32915
> > > 	* config/i386/i386.c (get_dllimport_decl): 
> > > SET_DECL_ASSEMBLER_NAME
> > > 	of the generated dllimport decl. 
> >   
> 
> This is pure cygwin/mingw stuff, written by the maintainer of this 
> (sub-)target. This patch doesn't touch shared code, so it 
> doesn't need 
> any other approvals. It is OK to commit the patch to mainline.
> 

Thanks for the clarification.  I wasn't sure whether my authority
extended to such parts of files like i386.[ch].
Danny 

> Thanks,
> Uros.
> 

Re: [PATCH] Fix PR 32915 dllimport regression

[Uros Bizjak]

François-Xavier Coudert wrote:
> > 
> > 2007-07-28  Danny Smith <dannysmith@users.sourcforge.net>
> > 
> > 	PR target/32915
> > 	* config/i386/i386.c (get_dllimport_decl): 
> > SET_DECL_ASSEMBLER_NAME
> > 	of the generated dllimport decl. 
>   

This is pure cygwin/mingw stuff, written by the maintainer of this 
(sub-)target. This patch doesn't touch shared code, so it doesn't need 
any other approvals. It is OK to commit the patch to mainline.

Thanks,
Uros.

RE: [PATCH] Fix PR 32915 dllimport regression

[François-Xavier Coudert]

http://gcc.gnu.org/ml/gcc-patches/2007-07/msg02024.html

*ping*