Re: g77 testsuite failures under hpux 10.20

["John David Anglin" <dave at hiauly1 dot hia dot nrc dot ca>]

>    For reasons too disgusting to describe storage for the new name
>    is allocated either on the saveable_obstack (released at function
>    exit) or on the permanent_obstack for things that can never change
>    (libcall names for example). */

[...]

> If I was to guess, the obstack must be popped at some point, possibly
> after the main program (this is where the @do_lio string is allocated).

Actually, that's what happens according to the previous comment.  As a
further note, see the f/ChangeLog entry:

Thu Dec 30 11:42:05 1999  Geoff Keating  <geoffk@cygnus.com>

        * com.c (ffecom_init_0): Make double_ftype_double,
	float_ftype_float, ldouble_ftype_ldouble,
	ffecom_tree_ptr_to_fun_type_void local.
	(tracker_head): New static variable.
	(mark_tracker_head): New, marker procedure for tracker_head.
	(ffecom_save_tree_forever): New procedure.
	(ffecom_init_zero_): Remove obstack use.
	(ffecom_make_gfrt_): Remove obstack use.
	(ffecom_sym_transform_): Remove obstack use, save appropriate trees.
	(ffecom_transform_common_): Remove obstack use, save appropriate trees.
	(ffecom_type_namelist_): Remove obstack use, save appropriate trees.
	(ffecom_type_vardesc_): Remove obstack use, save appropriate trees.
	(ffecom_lookup_label): Remove obstack use, save appropriate trees.
	(duplicate_decls): Remove obstack use.
	(finish_function): push & pop ggc context around
	rest_of_compilation when building nested function.
	(mark_binding_level): New function.
	(init_decl_processing): Mark all the GC roots.
	(ggc_p): Set to 1.

This may be what broke things.

Possible solutions:

1) Fix ENCODE_SECTION_INFO so that that string is allocated on the permanent
   obstack.

2) Use ggc_p and put newstr in gc memory.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)