regression of egcs-1.0 on powerpc-ibm-aix4.2.1.0

regression of egcs-1.0 on powerpc-ibm-aix4.2.1.0

[Andrey Slepuhin]

Here are the results of egcs-1.0 g++ testsuite on powerpc-ibm-aix4.2.1.0.
There is a regression with crash19.C and crash48.C files.
Indeed, compiler error messages are valid but they are not the
same as expected in testsuite. I put a part of g++.log file
related to these fails at the end of the message.

Test Run By pooh on Fri Dec  5 04:23:51 1997
Native configuration is powerpc-ibm-aix4.2.1.0

		=== g++ tests ===

FAIL: g++.brendan/crash19.C candidate for bad call (test for errors, line 1418)
FAIL: g++.brendan/crash19.C list of candidates for bad call (test for errors,
line 1463)
FAIL: g++.brendan/crash19.C no matching fn for call (test for errors, line 1486)
FAIL: g++.brendan/crash19.C invalid type for default argument -- not expecting
const (test for errors, line 1491)
FAIL: g++.brendan/crash48.C candidates (test for errors, line 14)
FAIL: g++.brendan/crash48.C no match (test for errors, line 20)
XPASS: g++.jason/destruct3.C - (test for bogus messages, line 38)
FAIL: g++.jason/thunk3.C (test for excess errors)
FAIL: g++.law/profile1.C (test for excess errors)
XPASS: g++.mike/dyncast1.C  Execution test
XPASS: g++.mike/dyncast2.C  Execution test

		=== g++ Summary ===

# of expected passes		3390
# of unexpected failures	8
# of unexpected successes	3
# of expected failures		82
# of untested testcases		6

-------------------------------crash19.C-------------------------------
Executing on host: /home/pooh/src/egcs-1.0/gcc/testsuite/../xgcc
-B/home/pooh/src/egcs-1.0/gcc/testsuite/../
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C
-I/home/pooh/src/egcs-1.0/libio
-I/home/pooh/src/egcs-1.0/libio I/home/pooh/src/egcs-1.0/libstdc++ 
-I/home/pooh/src/egcs-1.0/libstdc++/stl
-I/home/pooh/src/egcs-1.0/gcc/testsuite/../include -c  -o /tmp/crash19.o
compiler exited with status 1
output is:
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C: In
method `int filebuf::do_flush()':
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:317:
member `_pbase' is from private base class
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:317:
member `_pptr' is from private base class
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:317:
member `_pbase' is from private base class
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C: At top
level:
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:956:
`bool' is now a keyword
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:1224:
warning: invalid type `void (*)(const void *, FILE *, FILE *)' for default
argument to `void (*)(void *, FILE *, FILE *)'
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C: In
method `dict::dict()':
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:1218:
no matching function for call to `dict::internal::internal (int (*)(const void
*, const void *), void (*)(void *, FILE *), void (*)(const void *, FILE *, FILE
*), const char * (*)(const void *))'
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:1194:
candidates are: dict::internal::internal(const dict::internal &)
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:1169:                
dict::internal::internal(int (*)(void *, void *), void (*)(void *, FILE *), void
(*)(void *, FILE *, FILE *), const char* (*)(const void *))
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash19.C:1218:
confused by earlier errors, bailing out

PASS: g++.brendan/crash19.C  (test for errors, line 317)
PASS: g++.brendan/crash19.C use of bool as identifier (test for errors, line
956)
PASS: g++.brendan/crash19.C candidate for bad call (test for errors, line 1169)
PASS: g++.brendan/crash19.C candidate list for bad call (test for errors, line
1194)
PASS: g++.brendan/crash19.C no matching fn call (test for errors, line 1218)
PASS: g++.brendan/crash19.C invalid type for default argument -- not expecting
const (test for errors, line 1224)
FAIL: g++.brendan/crash19.C candidate for bad call (test for errors, line 1418)
FAIL: g++.brendan/crash19.C list of candidates for bad call (test for errors,
line 1463)
FAIL: g++.brendan/crash19.C no matching fn for call (test for errors, line 1486)
FAIL: g++.brendan/crash19.C invalid type for default argument -- not expecting
const (test for errors, line 1491)
PASS: g++.brendan/crash19.C (test for excess errors)
----------------------------crash48.C--------------------------------
Executing on host: /home/pooh/src/egcs-1.0/gcc/testsuite/../xgcc
-B/home/pooh/src/egcs-1.0/gcc/testsuite/../
 /home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C  -ansi
-pedantic-errors  -I/home/pooh/src/egcs-1.0/libio
-I/home/pooh/src/egcs-1.0/libio -I/home/pooh/src/egcs-1.0/libstdc++
-I/home/pooh/src/
egcs-1.0/libstdc++/stl -I/home/pooh/src/egcs-1.0/gcc/testsuite/../include -c  -o
/tmp/crash48.o
compiler exited with status 1
output is:
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C:11:
warning: all member functions in class `bug' are private
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C: In
method `bug::bug(int)':
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C:15: no
matching function for call to `internal::internal (int)'
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C:6:
candidates are: internal::internal(const internal &)
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C:6:                
internal::internal()
/home/pooh/src/egcs-1.0/gcc/testsuite/g++.old-deja/g++.brendan/crash48.C:15:
confused by earlier errors, bailing out

PASS: g++.brendan/crash48.C candidates are (test for errors, line 6)
PASS: g++.brendan/crash48.C several errors (test for errors, line 11)
FAIL: g++.brendan/crash48.C candidates (test for errors, line 14)
PASS: g++.brendan/crash48.C no match.* (test for errors, line 15)
FAIL: g++.brendan/crash48.C no match (test for errors, line 20)
PASS: g++.brendan/crash48.C (test for excess errors)
--------------------------------------------------------------------------------

Addition: regression of egcs-1.0 on powerpc-ibm-aix4.2.1.0

[Andrey Slepuhin]

Andrey Slepuhin wrote:
> There is a regression with crash19.C and crash48.C files.

This regression is seen only if egcs is bootstrapped. If I simply
do "make; cd gcc; make -k check" I don't get fails with these files.
The effect is stable.

Andrey.

Re: Addition: regression of egcs-1.0 on powerpc-ibm-aix4.2.1.0

[Jeffrey A Law]

  In message < 348AED8C.546B8532@msu.ru >you write:
  > Andrey Slepuhin wrote:
  > > There is a regression with crash19.C and crash48.C files.
  > 
  > This regression is seen only if egcs is bootstrapped. If I simply
  > do "make; cd gcc; make -k check" I don't get fails with these files.
  > The effect is stable.
This would tend to indicate that egcs mis-compiled itself, much
like a comparison failure.

Can you debug the two cc1plus programs and see where they start to
behave differently -- odds are that will point us at a codegen bug
in the compiler.

jeff

Re: Addition: regression of egcs-1.0 on powerpc-ibm-aix4.2.1.0

[Andrey Slepuhin]

Jeffrey A Law wrote:
> Can you debug the two cc1plus programs and see where they start to
> behave differently -- odds are that will point us at a codegen bug
> in the compiler.

Well, trying to debug bootstrapped cc1plus on crash48.C I obtain
the following gdb output:
----------------------------------------------------------------------
(gdb) run crash48.ii -quiet -dumpbase crash48.cc -version -o crash48.s
Starting program: /home/pooh/src/egcs-1.0/gcc/cc1plus crash48.ii -quiet
-dumpbase crash48.cc -version -o crash48.s
GNU C++ version egcs-2.90.21 971202 (egcs-1.00 release)
(powerpc-ibm-aix4.2.1.0) compiled by GNU C version egcs-2.90.21 971202
(egcs-1.00 release).
testsuite/g++.old-deja/g++.brendan/crash48.C:11: warning: all member
functions in class `bug' are private
testsuite/g++.old-deja/g++.brendan/crash48.C: In method `bug::bug(int)':
testsuite/g++.old-deja/g++.brendan/crash48.C:15: no matching function
for call to `internal::internal (int)'
testsuite/g++.old-deja/g++.brendan/crash48.C:6: candidates are:
internal::internal(const internal &)
testsuite/g++.old-deja/g++.brendan/crash48.C:6:                
internal::internal()

Program received signal SIGSEGV, Segmentation fault.
0x100e1e18 in add_conversions (binfo=0x0) at search.c:3696
3696      SET_BINFO_MARKED (binfo);
(gdb) bt
#0  0x100e1e18 in add_conversions (binfo=0x0) at search.c:3696
#1  0x100df79c in dfs_walk (binfo=0x0, fn=0x20016e9c <__dbargs+5884>, 
    qfn=0x20016e18 <__dbargs+5752>) at search.c:2519
#2  0x100e1eb0 in lookup_conversions (type=0x200543a8) at search.c:3706
#3  0x10115e44 in build_user_type_conversion_1 (totype=0x2006a640, 
    expr=0x2004bcd8, flags=3) at call.c:4300
#4  0x101162e4 in build_user_type_conversion (totype=0x0,
expr=0x2004bcd8, 
    flags=536964632) at call.c:4448
#5  0x1009760c in build_type_conversion (code=ERROR_MARK,
xtype=0x2006a640, 
    expr=0x2004bcd8, for_sure=1) at cvt.c:1335
#6  0x10095170 in cp_convert_to_pointer (type=0x2006a640,
expr=0x2004bcd8)
    at cvt.c:152
#7  0x10095964 in convert_to_pointer_force (type=0x2006a640,
expr=0x2004bcd8)
    at cvt.c:379
#8  0x10097354 in convert_force (type=0x2006a640, expr=0x20016e9c,
convtype=95)
    at cvt.c:1252
#9  0x100a3510 in build_c_cast (type=0x2006a640, expr=0x2004bcd8)
    at typeck.c:5631
#10 0x101f374c in build_new (placement=0x0, decl=0x2004d248, init=0x83, 
    use_global_new=0) at init.c:2773
#11 0x101fefc4 in yyparse () at parse.y:916
#12 0x1000303c in compile_file (name=0x0) at toplev.c:2513
#13 0x10006620 in main (argc=8, argv=0x2ff22144, envp=0x20016e18)
    at toplev.c:4403
#14 0x100001c8 in __start ()
(gdb) cont
Continuing.
testsuite/g++.old-deja/g++.brendan/crash48.C:15: confused by earlier
errors, bailing out

Program exited with code 042.
----------------------------------------------------------------------------

comparing this with cc1plus built with gcc-2.7.2 I saw that it
never calls yyparse () at parse.y:916, but calls yyparse () at
parse.y:900
in similar situation. Hope this information can help.

Andrey.

Re: Addition: regression of egcs-1.0 on powerpc-ibm-aix4.2.1.0

[Andrey Slepuhin]

Sorry, I mean of course, that bootstrapped cc1plus never
calls build_new at parse.y:916 :)

Andrey.