[patch] Fix PR41764

[patch] Fix PR41764

[Rafael Espindola]

[-- Attachment #1: Type: text/plain, Size: 491 bytes --]

The attached patch fixes PR 41764. The problem was that the fortran FE
was creating fake variables and marking them public. They were not
written to the final assembly becouse the FE sets TREE_ASM_WRITTEN,
but they were in the IL symbol table.

OK if bootstraps and tests are OK?

2009-10-21  Rafael Avila de Espindola  <espindola@google.com>

	PR 41764
	* trans-common.c (create_common): Set TREE_PUBLIC to false on
	fake variables.

Cheers,
-- 
Rafael Ávila de Espíndola

[-- Attachment #2: fake.patch --]
[-- Type: text/x-diff, Size: 1084 bytes --]

diff --git a/gcc/fortran/trans-common.c b/gcc/fortran/trans-common.c
index 5b1952a..1fb3c40 100644
--- a/gcc/fortran/trans-common.c
+++ b/gcc/fortran/trans-common.c
@@ -680,7 +680,6 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
       var_decl = build_decl (s->sym->declared_at.lb->location,
 			     VAR_DECL, DECL_NAME (s->field),
 			     TREE_TYPE (s->field));
-      TREE_PUBLIC (var_decl) = TREE_PUBLIC (decl);
       TREE_STATIC (var_decl) = TREE_STATIC (decl);
       TREE_USED (var_decl) = TREE_USED (decl);
       if (s->sym->attr.use_assoc)
@@ -689,7 +688,9 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
 	TREE_ADDRESSABLE (var_decl) = 1;
       /* This is a fake variable just for debugging purposes.  */
       TREE_ASM_WRITTEN (var_decl) = 1;
-      
+      /* Fake variables are not visible from other translation units. */
+      TREE_PUBLIC (var_decl) = 0;
+
       /* To preserve identifier names in COMMON, chain to procedure
          scope unless at top level in a module definition.  */
       if (com

Re: [patch] Fix PR41764

[Rafael Espindola]

> OK if bootstraps and tests are OK?

Looks like this breaks debug generation for fortran:

gfortran.dg/debug/pr35154-dwarf2.f scan-assembler (DW_AT_name:
"__BLNK__"|"__BLNK__[^\n]*"[^\n]*DW_AT_name)
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler (DW_AT_name:
"__BLNK__"|"__BLNK__[^\n]*"[^\n]*DW_AT_name)
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler (DW_AT_name:
"label"|"label[^\n]*"[^\n]*DW_AT_name)
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler (DW_AT_name:
"label"|"label[^\n]*"[^\n]*DW_AT_name)
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler DIE[^\n]*DW_TAG_common_block
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler DIE[^\n]*DW_TAG_common_block
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler DIE[^\n]*DW_TAG_common_block
gfortran.dg/debug/pr35154-dwarf2.f scan-assembler DIE[^\n]*DW_TAG_common_block
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"__BLNK__",226
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"__BLNK__",228
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"label_",226
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"label_",228
gfortran.dg/debug/pr37738.f scan-assembler-times DIE[^\n]*DW_TAG_common_block 3
gfortran.dg/debug/pr37738.f scan-assembler-times DIE[^\n]*DW_TAG_common_block 3


Will try to debug, but could really use some help from someone that
knows fortran (and gfortran).

Cheers,
-- 
Rafael Ávila de Espíndola

Re: [patch] Fix PR41764

[Rafael Espindola]

[-- Attachment #1: Type: text/plain, Size: 454 bytes --]

> Looks like this breaks debug generation for fortran:

The attached patch reduces the problems to:

gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"__BLNK__",226
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"__BLNK__",228
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"label_",226
gfortran.dg/debug/pr35154-stabs.f scan-assembler .stabs.*"label_",228

Debugging...

Cheers,
-- 
Rafael Ávila de Espíndola

[-- Attachment #2: fake.patch --]
[-- Type: text/x-diff, Size: 1823 bytes --]

diff --git a/gcc/dwarf2out.c b/gcc/dwarf2out.c
index ba59251..cb3d790 100644
--- a/gcc/dwarf2out.c
+++ b/gcc/dwarf2out.c
@@ -15606,12 +15606,11 @@ fortran_common (tree decl, HOST_WIDE_INT *value)
   tree offset;
   int volatilep = 0, unsignedp = 0;
 
-  /* If the decl isn't a VAR_DECL, or if it isn't public or static, or if
+  /* If the decl isn't a VAR_DECL, or if it isn't static, or if
      it does not have a value (the offset into the common area), or if it
      is thread local (as opposed to global) then it isn't common, and shouldn't
      be handled as such.  */
   if (TREE_CODE (decl) != VAR_DECL
-      || !TREE_PUBLIC (decl)
       || !TREE_STATIC (decl)
       || !DECL_HAS_VALUE_EXPR_P (decl)
       || !is_fortran ())
diff --git a/gcc/fortran/trans-common.c b/gcc/fortran/trans-common.c
index 5b1952a..1fb3c40 100644
--- a/gcc/fortran/trans-common.c
+++ b/gcc/fortran/trans-common.c
@@ -680,7 +680,6 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
       var_decl = build_decl (s->sym->declared_at.lb->location,
 			     VAR_DECL, DECL_NAME (s->field),
 			     TREE_TYPE (s->field));
-      TREE_PUBLIC (var_decl) = TREE_PUBLIC (decl);
       TREE_STATIC (var_decl) = TREE_STATIC (decl);
       TREE_USED (var_decl) = TREE_USED (decl);
       if (s->sym->attr.use_assoc)
@@ -689,7 +688,9 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
 	TREE_ADDRESSABLE (var_decl) = 1;
       /* This is a fake variable just for debugging purposes.  */
       TREE_ASM_WRITTEN (var_decl) = 1;
-      
+      /* Fake variables are not visible from other translation units. */
+      TREE_PUBLIC (var_decl) = 0;
+
       /* To preserve identifier names in COMMON, chain to procedure
          scope unless at top level in a module definition.  */
       if (com

Re: [patch] Fix PR41764

[Rafael Espindola]

[-- Attachment #1: Type: text/plain, Size: 542 bytes --]

> Debugging...

The joys of duplicated code...

The attached patch fixes the PR, bootstraps and passes the regression
tests. Is it OK?

gcc/
2009-10-27  Rafael Avila de Espindola  <espindola@google.com>

	* dbxout.c (dbxout_common_check): Accept non public trees.
	* dwarf2out.c (fortran_common): Accept non public trees.

gcc/fortran/
2009-10-27  Rafael Avila de Espindola  <espindola@google.com>

	* trans-common.c (create_common): Set TREE_PUBLIC to false on
	fake variables.

Cheers,
-- 
Rafael Ávila de Espíndola

[-- Attachment #2: fake.patch --]
[-- Type: text/x-diff, Size: 2706 bytes --]

diff --git a/gcc/dbxout.c b/gcc/dbxout.c
index b5688d9..0bf5e3c 100644
--- a/gcc/dbxout.c
+++ b/gcc/dbxout.c
@@ -3189,7 +3189,7 @@ dbxout_common_check (tree decl, int *value)
   rtx sym_addr;
   const char *name = NULL;
   
-  /* If the decl isn't a VAR_DECL, or if it isn't public or static, or if
+  /* If the decl isn't a VAR_DECL, or if it isn't static, or if
      it does not have a value (the offset into the common area), or if it
      is thread local (as opposed to global) then it isn't common, and shouldn't
      be handled as such.
@@ -3198,7 +3198,6 @@ dbxout_common_check (tree decl, int *value)
      for thread-local symbols.  Can be handled via same mechanism as used
      in dwarf2out.c.  */
   if (TREE_CODE (decl) != VAR_DECL
-      || !TREE_PUBLIC(decl)
       || !TREE_STATIC(decl)
       || !DECL_HAS_VALUE_EXPR_P(decl)
       || DECL_THREAD_LOCAL_P (decl)
diff --git a/gcc/dwarf2out.c b/gcc/dwarf2out.c
index 129ba7d..732b6a6 100644
--- a/gcc/dwarf2out.c
+++ b/gcc/dwarf2out.c
@@ -15621,12 +15621,11 @@ fortran_common (tree decl, HOST_WIDE_INT *value)
   tree offset;
   int volatilep = 0, unsignedp = 0;
 
-  /* If the decl isn't a VAR_DECL, or if it isn't public or static, or if
+  /* If the decl isn't a VAR_DECL, or if it isn't static, or if
      it does not have a value (the offset into the common area), or if it
      is thread local (as opposed to global) then it isn't common, and shouldn't
      be handled as such.  */
   if (TREE_CODE (decl) != VAR_DECL
-      || !TREE_PUBLIC (decl)
       || !TREE_STATIC (decl)
       || !DECL_HAS_VALUE_EXPR_P (decl)
       || !is_fortran ())
diff --git a/gcc/fortran/trans-common.c b/gcc/fortran/trans-common.c
index 5b1952a..1fb3c40 100644
--- a/gcc/fortran/trans-common.c
+++ b/gcc/fortran/trans-common.c
@@ -680,7 +680,6 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
       var_decl = build_decl (s->sym->declared_at.lb->location,
 			     VAR_DECL, DECL_NAME (s->field),
 			     TREE_TYPE (s->field));
-      TREE_PUBLIC (var_decl) = TREE_PUBLIC (decl);
       TREE_STATIC (var_decl) = TREE_STATIC (decl);
       TREE_USED (var_decl) = TREE_USED (decl);
       if (s->sym->attr.use_assoc)
@@ -689,7 +688,9 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
 	TREE_ADDRESSABLE (var_decl) = 1;
       /* This is a fake variable just for debugging purposes.  */
       TREE_ASM_WRITTEN (var_decl) = 1;
-      
+      /* Fake variables are not visible from other translation units. */
+      TREE_PUBLIC (var_decl) = 0;
+
       /* To preserve identifier names in COMMON, chain to procedure
          scope unless at top level in a module definition.  */
       if (com

Re: [patch] Fix PR41764

[Rafael Espindola]

Adding fortran@gcc.gnu.org

> The joys of duplicated code...
>
> The attached patch fixes the PR, bootstraps and passes the regression
> tests. Is it OK?
>
> gcc/
> 2009-10-27  Rafael Avila de Espindola  <espindola@google.com>
>
>        * dbxout.c (dbxout_common_check): Accept non public trees.
>        * dwarf2out.c (fortran_common): Accept non public trees.
>
> gcc/fortran/
> 2009-10-27  Rafael Avila de Espindola  <espindola@google.com>
>
>        * trans-common.c (create_common): Set TREE_PUBLIC to false on
>        fake variables.
>
> Cheers,
> --
> Rafael Ávila de Espíndola
>

Cheers,
-- 
Rafael Ávila de Espíndola

Re: [patch] Fix PR41764

[Ian Lance Taylor]

Rafael Espindola <espindola@google.com> writes:

> gcc/
> 2009-10-27  Rafael Avila de Espindola  <espindola@google.com>
>
> 	* dbxout.c (dbxout_common_check): Accept non public trees.
> 	* dwarf2out.c (fortran_common): Accept non public trees.
>
> gcc/fortran/
> 2009-10-27  Rafael Avila de Espindola  <espindola@google.com>
>
> 	* trans-common.c (create_common): Set TREE_PUBLIC to false on
> 	fake variables.

The middle-end parts of this patch are OK.

This is OK if the Fortran maintainers are OK with the change to
trans-common.c.

Thanks.

> diff --git a/gcc/dbxout.c b/gcc/dbxout.c
> index b5688d9..0bf5e3c 100644
> --- a/gcc/dbxout.c
> +++ b/gcc/dbxout.c
> @@ -3189,7 +3189,7 @@ dbxout_common_check (tree decl, int *value)
>    rtx sym_addr;
>    const char *name = NULL;
>    
> -  /* If the decl isn't a VAR_DECL, or if it isn't public or static, or if
> +  /* If the decl isn't a VAR_DECL, or if it isn't static, or if
>       it does not have a value (the offset into the common area), or if it
>       is thread local (as opposed to global) then it isn't common, and shouldn't
>       be handled as such.
> @@ -3198,7 +3198,6 @@ dbxout_common_check (tree decl, int *value)
>       for thread-local symbols.  Can be handled via same mechanism as used
>       in dwarf2out.c.  */
>    if (TREE_CODE (decl) != VAR_DECL
> -      || !TREE_PUBLIC(decl)
>        || !TREE_STATIC(decl)
>        || !DECL_HAS_VALUE_EXPR_P(decl)
>        || DECL_THREAD_LOCAL_P (decl)
> diff --git a/gcc/dwarf2out.c b/gcc/dwarf2out.c
> index 129ba7d..732b6a6 100644
> --- a/gcc/dwarf2out.c
> +++ b/gcc/dwarf2out.c
> @@ -15621,12 +15621,11 @@ fortran_common (tree decl, HOST_WIDE_INT *value)
>    tree offset;
>    int volatilep = 0, unsignedp = 0;
>  
> -  /* If the decl isn't a VAR_DECL, or if it isn't public or static, or if
> +  /* If the decl isn't a VAR_DECL, or if it isn't static, or if
>       it does not have a value (the offset into the common area), or if it
>       is thread local (as opposed to global) then it isn't common, and shouldn't
>       be handled as such.  */
>    if (TREE_CODE (decl) != VAR_DECL
> -      || !TREE_PUBLIC (decl)
>        || !TREE_STATIC (decl)
>        || !DECL_HAS_VALUE_EXPR_P (decl)
>        || !is_fortran ())
> diff --git a/gcc/fortran/trans-common.c b/gcc/fortran/trans-common.c
> index 5b1952a..1fb3c40 100644
> --- a/gcc/fortran/trans-common.c
> +++ b/gcc/fortran/trans-common.c
> @@ -680,7 +680,6 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
>        var_decl = build_decl (s->sym->declared_at.lb->location,
>  			     VAR_DECL, DECL_NAME (s->field),
>  			     TREE_TYPE (s->field));
> -      TREE_PUBLIC (var_decl) = TREE_PUBLIC (decl);
>        TREE_STATIC (var_decl) = TREE_STATIC (decl);
>        TREE_USED (var_decl) = TREE_USED (decl);
>        if (s->sym->attr.use_assoc)
> @@ -689,7 +688,9 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
>  	TREE_ADDRESSABLE (var_decl) = 1;
>        /* This is a fake variable just for debugging purposes.  */
>        TREE_ASM_WRITTEN (var_decl) = 1;
> -      
> +      /* Fake variables are not visible from other translation units. */
> +      TREE_PUBLIC (var_decl) = 0;
> +
>        /* To preserve identifier names in COMMON, chain to procedure
>           scope unless at top level in a module definition.  */
>        if (com

Re: [patch] Fix PR41764

[Toon Moene]

Rafael Espindola wrote:

> Adding fortran@gcc.gnu.org
> 
>> The joys of duplicated code...
>>
>> The attached patch fixes the PR, bootstraps and passes the regression
>> tests. Is it OK?

It doesn't fix everything.  I now get (gfortran is /usr/snp/bin/gfortran):

gfortran hdf2asim.o 
/scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/hdfg.a 
/scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a 
/scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/gcod.a -o 
/scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/bin/hdf2asim.x -O3 
-use-linker-plugin -ffast-math -fbacktrace -march=native -mtune=native 
-flto -fwhole-program --param max-inline-insns-single=10000000 --param 
max-inline-insns-auto=10000000 --param large-function-growth=100000 
--param large-function-insns=10000000 --param 
large-stack-frame-growth=100000 --param large-stack-frame=1000000 
-fdump-ipa-inline  -llapack -lblas -lfftw3 -L/home/hirlam/hdf/lib 
-lmfhdf -ldf -ljpeg -lz -lm
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of casihl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of cmachl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of casihl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of cmachl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of casihl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of cmachl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of casihl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
multiple definition of cmachl_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/port.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/gcod.a: 
multiple definition of grbcom_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/gcod.a: 
previous definition here
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/gcod.a: 
multiple definition of grbcom_
/usr/snp/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../x86_64-unknown-linux-gnu/bin/ld: 
error: /scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/lib/gcod.a: 
previous definition here
collect2: ld returned 1 exit status
make[1]: *** 
[/scratch/hirlam/hl_home/EXP/lib/src/linuxgfortran/bin/hdf2asim.x] Error 1

So the common blocks themselves (not the variables in it) are still 
multiply defined.

-- 
Toon Moene - e-mail: toon@moene.org - phone: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
At home: http://moene.org/~toon/
Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html

Re: [patch] Fix PR41764

[Rafael Espindola]

> It doesn't fix everything.  I now get (gfortran is /usr/snp/bin/gfortran):
>

....

>
> So the common blocks themselves (not the variables in it) are still multiply
> defined.

Can you provide me with a testcase? I don't know fortran, so if the
testcase is already reduce it it would be very nice :-)

> --
> Toon Moene - e-mail: toon@moene.org - phone: +31 346 214290
> Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
> At home: http://moene.org/~toon/
> Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html
>

Thanks,
-- 
Rafael Ávila de Espíndola

Re: [patch] Fix PR41764

[Tobias Burnus]

Ian Lance Taylor wrote:
> Rafael Espindola <espindola@google.com> writes
>> gcc/
>> 2009-10-27  Rafael Avila de Espindola  <espindola@google.com>
>>
>> 	* dbxout.c (dbxout_common_check): Accept non public trees.
>> 	* dwarf2out.c (fortran_common): Accept non public trees.
>>
>> gcc/fortran/
>> 2009-10-27  Rafael Avila de Espindola  <espindola@google.com>
>>
>> 	* trans-common.c (create_common): Set TREE_PUBLIC to false on
>> 	fake variables.
>>     
> The middle-end parts of this patch are OK.
>
> This is OK if the Fortran maintainers are OK with the change to
> trans-common.c.
>   
Sorry for the delay. The Fortran part is also OK.

Tobias

>> --- a/gcc/fortran/trans-common.c
>> +++ b/gcc/fortran/trans-common.c
>> @@ -680,7 +680,6 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
>>        var_decl = build_decl (s->sym->declared_at.lb->location,
>>  			     VAR_DECL, DECL_NAME (s->field),
>>  			     TREE_TYPE (s->field));
>> -      TREE_PUBLIC (var_decl) = TREE_PUBLIC (decl);
>>        TREE_STATIC (var_decl) = TREE_STATIC (decl);
>>        TREE_USED (var_decl) = TREE_USED (decl);
>>        if (s->sym->attr.use_assoc)
>> @@ -689,7 +688,9 @@ create_common (gfc_common_head *com, segment_info *head, bool saw_equiv)
>>  	TREE_ADDRESSABLE (var_decl) = 1;
>>        /* This is a fake variable just for debugging purposes.  */
>>        TREE_ASM_WRITTEN (var_decl) = 1;
>> -      
>> +      /* Fake variables are not visible from other translation units. */
>> +      TREE_PUBLIC (var_decl) = 0;
>> +
>>        /* To preserve identifier names in COMMON, chain to procedure
>>           scope unless at top level in a module definition.  */
>>        if (com
>>     
>