[RFA:] DejaGNU patch: Change -I to -isystem for testsuite.

[Hans-Peter Nilsson <hp at bitrange dot com>]

This patch fixes, for uninstalled newlib targets in a unified
tree, the listed spurious GCC testsuite excess-errors.  They
occur because the supposedly-to-be-installed newlib limits.h
uses #include_next.  That directive makes gcc emit a warning
when used in a non-system-header context.  While that can be
"fixed" for non-installed headers by adding "#pragma GCC
system_header", it shouldn't be needed; it'd be better if gcc
treated it as a system header while running the test-suite.
Done by changing -I to -isystem for include directives for the
test-suite.  The -isystem option has been there since at least
gcc-2.7.2 so it should be safe for all uses.  This patch is
against the sources.redhat.com unified-tree repository but the
patch applied fine (with offset) to the savannah.gnu.org DejaGNU
repository.

FAIL: gcc.dg/c90-intconst-1.c (test for excess errors)
FAIL: gcc.dg/c99-intconst-1.c (test for excess errors)
FAIL: g++.dg/abi/bitfield4.C (test for excess errors)
FAIL: g++.benjamin/bool01.C (test for excess errors)
FAIL: g++.benjamin/p12475.C (test for excess errors)
FAIL: g++.jason/rfg10.C (test for excess errors)

BTW, does anybody have an idea why gcc/testsuite/site.exp is
completely ineffective?  Changing s/-I/-isystem/g in the
site.exp in gcc/Makefile.in has no effect; in fact it doesn't
seem to be used at all.

Ok to commit to sources.redhat.com/dejagnu?

dejagnu:
	* lib/libgloss.exp (newlib_include_flags): Use -isystem, not -I.
	(libio_include_flags, g++_include_flags, libstdc++_include_flags,
	winsup_include_flags): Ditto.
	* doc/user.sgml (Local Config File): Use -isystem, not -I, in
	example.

brgds, H-P

Attachment: is1.txt
Description: Text document