Description of problem:

When debugging Fortran code, GDB will sometimes miscategorize a symbol token.

Consider the test case for gdb.fortran/array-slices.exp. I tested this on Fedora 35 with glibc debuginfo installed.

[kev@f35-1 gdb]$ ./gdb -q testsuite/outputs/gdb.fortran/array-slices/array-slices 
Reading symbols from testsuite/outputs/gdb.fortran/array-slices/array-slices...
(gdb) b show_str
Breakpoint 1 at 0x404586: file /ironwood1/sourceware-git/f35-master/bld/../../worktree-master/gdb/testsuite/gdb.fortran/array-slices.f90, line 41.
(gdb) run
Starting program: /mesquite2/sourceware-git/f35-master/bld/gdb/testsuite/outputs/gdb.fortran/array-slices/array-slices 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
 

Breakpoint 1, show_str (array=..., _array=26) at /ironwood1/sourceware-git/f35-master/bld/../../worktree-master/gdb/testsuite/gdb.fortran/array-slices.f90:41
41	end subroutine show_str
(gdb) p array
Attempt to use a type name as an expression

The reason for this behavior is that yylex() (which turns into f_yylex() in a gdb build) in gdb/f-exp.c searches, in sequence, STRUCT_DOMAIN, VAR_DOMAIN, and MODULE_DOMAIN for the symbol, stopping when it finds a match. However, there might be more than one match and some of the matches might be better than the others.  In this particular instance, it turns out that GDB is finding 'struct array' in glibc:

 <1><1ef36d>: Abbrev Number: 48 (DW_TAG_structure_type)
    <1ef36e>   DW_AT_name        : (indirect string, offset: 0x2c46f): array
    <1ef372>   DW_AT_byte_size   : 152
    <1ef373>   DW_AT_decl_file   : 5
    <1ef374>   DW_AT_decl_line   : 126
    <1ef375>   DW_AT_decl_column : 8
    <1ef375>   DW_AT_sibling     : <0x1ef38c>

This comes from glibc's nss/nss_files/file-hosts.c due to DYNARRAY construction.

Version-Release number of selected component (if applicable):

Problem observed on Fedora 35 with glibc debuginfo installed.  Until this bug is fixed, version number / release number of GDB is not important.  I.e. it should exist in all versions of GDB including and prior to gdb-11.1-2.

How reproducible:

Always, when conflicting symbols from some other library are present.

Steps to Reproduce:
1. 

From GDB's build directory, do:

make check TESTS="gdb.fortran/array-slices-bad.exp gdb.fortran/array-slices.exp"

2.

Debug the test case executable as shown above.

Actual results:

See above.

Expected results:

(gdb) p array
$1 = 'abcdefghijklmnopqrstuvwxyz'

Additional info:

An upstream commit contains changes which paper over the test suite failures that I observed while testing GDB 11.1 for Fedora 35.  There is still a bug in GDB, however; these changes to the test suite do not fix it.

I am, however, applying that patch to GDB 11.1 in order to obtain test results which are easier to compare than the alternative.