Re: [Patch, fortran] PR107900 Select type with intrinsic type inside associate causes ICE / Segmenation fault

[Harald Anlauf <anlauf@gmx.de>]

Hi Paul,

On 6/17/23 11:14, Paul Richard Thomas via Gcc-patches wrote:
> Hi All,
> 
> The attached patch is amply described by the comments and the
> changelog. It also includes the fix for the memory leak in decl.cc, as
> promised some days ago.
> 
> OK for trunk?

I hate to say it, but you forgot to add the testcase again... :-(

The patch fixes your "extended" testcase in

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107900#c2

but the original one in comment ICEs for me here:

% gfc-14 pr107900.f90
f951: internal compiler error: Segmentation fault
0x1025c2f crash_signal
         ../../gcc-trunk/gcc/toplev.cc:314
0x9d31d3 resolve_select_type
         ../../gcc-trunk/gcc/fortran/resolve.cc:9791
0x9cef5e gfc_resolve_code(gfc_code*, gfc_namespace*)
         ../../gcc-trunk/gcc/fortran/resolve.cc:12588
0x9d2431 resolve_codes
         ../../gcc-trunk/gcc/fortran/resolve.cc:18057
0x9d24fe gfc_resolve(gfc_namespace*)
         ../../gcc-trunk/gcc/fortran/resolve.cc:18092
0x9cf0ee gfc_resolve(gfc_namespace*)
         ../../gcc-trunk/gcc/fortran/resolve.cc:18077
0x9cf0ee resolve_block_construct
         ../../gcc-trunk/gcc/fortran/resolve.cc:10971
0x9cf0ee gfc_resolve_code(gfc_code*, gfc_namespace*)
         ../../gcc-trunk/gcc/fortran/resolve.cc:12596
0x9d2431 resolve_codes
         ../../gcc-trunk/gcc/fortran/resolve.cc:18057
0x9d24fe gfc_resolve(gfc_namespace*)
         ../../gcc-trunk/gcc/fortran/resolve.cc:18092
0x9b11f1 resolve_all_program_units
         ../../gcc-trunk/gcc/fortran/parse.cc:6864
0x9b11f1 gfc_parse_file()
         ../../gcc-trunk/gcc/fortran/parse.cc:7120
0xa033ef gfc_be_parse_file
         ../../gcc-trunk/gcc/fortran/f95-lang.cc:229

It hits an assert here:

9790          st = gfc_find_symtree (ns->sym_root, name);
9791          gcc_assert (st->n.sym->assoc);

My tree is slightly modified, but the changes should not have
any effect here.

Can you please have a look, too?

Thanks,
Harald

> Regards
> 
> Paul
> 
> PS This leaves 89645 and 99065 as the only real blockers to PR87477.
> These will take a little while to fix. They come about because the
> type of the associate name is determined by that of a derived type
> function that hasn't been parsed at the time that component references
> are being parsed. If the order of the contained procedures is
> reversed, both test cases compile correctly. The fix will comprise
> matching the component name to the accessible derived types, while
> keeping track of all the references in case the match is ambiguous and
> has to be fixed up later.