Bug 459374

This ticket was created by LLNL and first forwarded to Etnus's support organization for TotalView but after analyzing the issue, they have come to believe that it is a problem with the way that gfortran emits DWARF information.

Gfortran emits bad DWARF for array parameters. Upper bound is missing.

rhel52-x8664:/home/seppo/Bugs/Bug_11350 > cat /etc/*-rel*
cat: /etc/lsb-release.d: Is a directory
Red Hat Enterprise Linux Server release 5.2 (Tikanga)
rhel52-x8664:/home/seppo/Bugs/Bug_11350 > uname -a
Linux rhel52-x8664.totalviewtech.com 2.6.18-92.el5 #1 SMP Tue Apr 29 13:16:15 EDT 2008 x86_64 x86_64 x86_64 GNU/Linux
rhel52-x8664:/home/seppo/Bugs/Bug_11350 >

> gfortran -v
gcc version 4.1.2 20071124 (Red Hat 4.1.2-42)

> gfortran -g -O0 -o a_g.out tx_f90_contained_sub.f90


> readelf -w ./a_g.out

..
     DW_AT_location    : 3 byte block: 91 f8 7d         (DW_OP_fbreg: -264)
 <2><12a>: Abbrev Number: 10 (DW_TAG_formal_parameter)
     DW_AT_name        : c_sub
     DW_AT_decl_file   : 1
     DW_AT_decl_line   : 22
     DW_AT_type        : <232>
     DW_AT_location    : 3 byte block: 91 f0 7d         (DW_OP_fbreg: -272)
 <2><13b>: Abbrev Number: 10 (DW_TAG_formal_parameter)
     DW_AT_name        : c_size
...
     DW_AT_type        : <1e0>
 <1><222>: Abbrev Number: 17 (DW_TAG_array_type)
     DW_AT_sibling     : <232>
     DW_AT_type        : <e2>
 <2><22b>: Abbrev Number: 18 (DW_TAG_subrange_type)
     DW_AT_type        : <7f>
     DW_AT_lower_bound : 0
 <1><232>: Abbrev Number: 16 (DW_TAG_const_type)
     DW_AT_type        : <237>
 <1><237>: Abbrev Number: 19 (DW_TAG_pointer_type)
     DW_AT_byte_size   : 8
     DW_AT_type        : <222>
 <1><23d>: Abbrev Number: 16 (DW_TAG_const_type)
 ....


=> the problem is that for array c_sub, only DW_AT_lower_bound is
defined in DWARF. DW_AT_upper_bound is missing in <22b> entry.

Reproducer is attached.

This is TotalView internal bug #11350


This event sent from IssueTracker by jnansi  [Support Engineering Group]
 issue 198958

Date problem reported: 8-12-08

Problem: Gfortran emits bad DWARF for array parameters.

This issue was originally escalated to Etnus by LLNL as a TotalView bug,
but, Etnus has come back and said that the problem appears to be with
gfortran instead.

To reproduce:

1) rename the attached program so that it has an .f90 extension, rather
than a .f one.  

2) compile the program with: 

gfortran -g -O0 -o a_g.out bad-dwarf-array-params.f90

3) readelf -w ./a_g.out

According to Etnus, the problem is that for array c_sub, only
DW_AT_lower_bound is defined in DWARF. DW_AT_upper_bound is missing in the
<22b> entry.

What is expected from SEG:  As with some previous gfortran problems we
submitted, these are kind of over our heads, and we need someone well
versed in gfortran to review this and determine if we have a bug here that
needs fixing.

This does appear to reproduce on my RHEL5.2 test system.

Thanks.




Issue escalated to Support Engineering Group by: kbaxley.
kbaxley assigned to issue for LLNL (HPC).
Category set to: Devel Tools
Internal Status set to 'Waiting on SEG'
Status set to: Waiting on Tech

This event sent from IssueTracker by jnansi  [Support Engineering Group]
 issue 198958

Created attachment 314457 [details]
reproducer

Currently, we do not have anyone available in SEG familiar with fortran and DWARF, hence escalating to engineering.

Patch http://gcc.gnu.org/ml/gcc-patches/2008-08/msg01599.html

This should be fixed in GCC 4.3/4.4, won't be fixed in 4.1.

Which version of gcc 4.3 has this fixed in. I'm still getting reports that the version that ships with 5.3 still manifests this problem. If it is a version subsequent to 5.3 we can probably live with that but if it was supposedly fixed in the 5.3 release then we need to look deeper.

The patch mentioned in #c6 really fixed the debuginfo for variable sized non-desc arrays, if you modify the testcase by moving the print statements from contained_sub to sub_test, it works with that patch just fine.

The extra problems are caused by the nested functions.  I guess one problem is type sharing between the contained and containing function:

  /* ??? We should be remapping types as well, surely.  */
  new_decl = build_decl (VAR_DECL, DECL_NAME (decl), TREE_TYPE (decl));

in get_nonlocal_debug_decl is one issue (we should be remapping the types in the nested function if they are variable length for debuginfo purposes, perhaps just at -O0), another issue is that get_local_debug_decl should replace (again, perhaps at -O0 only) vars in the VLA type ranges that were referenced by nested functions and last issue is that we don't allow DECL_BY_REFERENCE for VAR_DECLs (as the same bit is used for TREE_PRIVATE).

See
http://gcc.gnu.org/ml/gcc-patches/2009-04/msg00602.html
and
http://gcc.gnu.org/ml/gcc-patches/2009-04/msg00604.html

The only problem after these patches is that a_sub symbol isn't in contained_sub's debuginfo, only in the outer function.

See
http://gcc.gnu.org/ml/gcc-patches/2009-04/msg01718.html
Another patch on top of the two that solves the remaining issues, now
all of a_sub, b_sub and c_sub can be correctly printed both in both sub_test and contained_sub if they are referenced in both.

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on therefore solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2009-1375.html