Bison-1.875, again! (Re: GCC-3.2.2 pre-release (second iteration))

[Paolo Carlini <pcarlini at unitus dot it>]

Gabriel Dos Reis wrote:

> I would like you run the testsuite again but for current CVS
> gcc-3_2-branch and report the results.  Thank you very much for your
> time.
Hi Gaby!

Everything is clear, now. I'm still seeing those spurious regressions, 
to wit

FAIL: gcc.dg/cpp/assert_trad2.c  (test for errors, line 13)
FAIL: gcc.dg/cpp/assert_trad2.c  (test for errors, line 16)
FAIL: gcc.dg/cpp/assert_trad2.c bad syntax (test for errors, line 23)
FAIL: gcc.dg/cpp/assert_trad2.c (test for excess errors)
FAIL: gcc.dg/cpp/direct2s.c not a comment (test for errors, line 38)
FAIL: gcc.dg/cpp/direct2s.c (test for excess errors)

FAIL: g++.dg/lookup/using2.C  (test for errors, line 28)
FAIL: g++.dg/lookup/using2.C  (test for errors, line 44)
FAIL: g++.dg/lookup/using2.C (test for excess errors)
XPASS: g++.dg/parse/angle-bracket.C  (test for bogus messages, line 7)
FAIL: g++.dg/parse/angle-bracket.C (test for excess errors)
FAIL: g++.dg/template/typename3.C (test for excess errors)

but they are *all* due to the Bison version I'm using: the very same issue
("parse error" -> "syntax error") that a few days ago led me to propose a
patch for mainline. For some reason, the current 3.2.2 branch when fed the
C testsuite triggers few more new posix "syntax error" from the 
Bison-generated
parser than mainline (and also some C++ ones which, of course are not 
relevant
for mainline). For instance:

--- 3_2_1       2003-01-31 02:13:21.000000000 +0100
+++ 3_2_2-bison-1_875   2003-01-31 02:13:08.000000000 +0100
@@ -4,7 +4,7 @@
assert_trad2.c:8: missing '(' after predicate
assert_trad2.c:10: assertion without predicate
assert_trad2.c:13: predicate must be an identifier
-assert_trad2.c:13: parse error
+assert_trad2.c:13: syntax error
assert_trad2.c:16: predicate must be an identifier
-assert_trad2.c:16: parse error
-assert_trad2.c:23: parse error
+assert_trad2.c:16: syntax error
+assert_trad2.c:23: syntax error

In my opinion, a testsuite fix very similar to mine of a few days ago should
be applied to those testcases. If you want, I can work on this tomorrow...

Paolo.