Last known to work with: "Sat Jul 9 22:01:35 UTC 2005". Known to fail with: "Sun Jul 10 06:32:56 UTC 2005". With LAST_UPDATED: "Tue Jul 12 03:25:01 UTC 2005" I still get: Running /home/hp/combined/combined/libstdc++-v3/testsuite/libstdc++-dg/normal.exp ... FAIL: 23_containers/set/explicit_instantiation/3.cc (test for excess errors) With the message in the .log being (c-n-p): /home/hp/combined/crislinux-sim/cris-axis-linux-gnu/libstdc++-v3/include/bits/stl_set.h: In member function 'std::pair<typename s\ td::_Rb_tree<_Key, _Key, std::_Identity<_Key>, _Compare, typename _Alloc::rebind<_Key>::other>::const_iterator, bool> std::set<_K\ ey, _Compare, _Alloc>::insert(const _Key&) [with _Key = __gnu_test::NonDefaultConstructible, _Compare = std::less<__gnu_test::Non\ DefaultConstructible>, _Alloc = std::allocator<__gnu_test::NonDefaultConstructible>]':^M /home/hp/combined/crislinux-sim/cris-axis-linux-gnu/libstdc++-v3/include/bits/stl_set.h:318: internal compiler error: tree check:\ expected ssa_name, have var_decl in is_old_name, at tree-into-ssa.c:466^M Will attach preprocessed source.
Created attachment 9248 [details] Preprocessed source cc1plus -fpreprocessed 2.ii -march=v10 -quiet -dumpbase 2.cc -auxbase-strip 2.s -g -O2 -version -ffunction-sections -fdata-sections -fmessage-length=0 -o 2.s
This looks very much related to PR 22416.
Between LAST_UPDATED "Mon Jul 11 09:23:25 UTC 2005" and "Mon Jul 11 18:07:42 UTC 2005", I started getting an additional similar failure for cris-axis-linux-gnu: ... FAIL: 23_containers/set/explicit_instantiation/3.cc (test for excess errors) Same ICE. In addition, the 3.cc test now fails on mmix-knuth-mmixware as well. Even more, both 2.cc and 3.cc now fail on cris-axis-elf as well. (No preprocessed source for any of these.)
These failures have appeared for me on i686-pc-linux-gnu on 20050717.
the patch in PR 22483 should fix this.
Fixed.
I'm reopening this PR as the failure with mmix-knuth-mmixware and LAST_UPDATED "Fri Jul 29 18:30:36 UTC 2005" is identical to the original description.
Created attachment 9387 [details] Preprocessed source for mmix-knuth-mmixware for reopened PR. /home/hp/combined/mmix-regobj/./gcc/cc1plus -fpreprocessed 2.ii -quiet -dumpbase 2.cc -auxbase-strip 2.s -g -O2 -version -fmessage-length=0 -o 2.s
For the record, cross to cris-axis-linux-gnu started failing with the same log message for 1.cc and 3.cc with "Sun Jul 31 22:56:52 UTC 2005", last known passed with "Sat Jul 30 14:53:00 UTC 2005".
*** Bug 22416 has been marked as a duplicate of this bug. ***
*** Bug 22598 has been marked as a duplicate of this bug. ***
*** Bug 23310 has been marked as a duplicate of this bug. ***
I'm reducing the testcase from 23310 at the moment. Trying to produce both an ice-on-valid and a (smaller) ice-on-invalid (or ice-on-ice, whatever will come out of it ;)).
Created attachment 9462 [details] reduced testcase from PR23310 PR23310 reduced, ice-on-valid with -O2, not with -O2 -fdump-tree-alias1
Created attachment 9464 [details] more minimized 23310 testcase Even more reduced. Now one has to manually propagate types to reduce it maybe even more. Remember to check successful compile with -O2 -fdump-tree-alias1 and ICE with -O2.
Backtrace of the ICE #0 internal_error (gmsgid=0x882bbf8 "tree check: %s, have %s in %s, at %s:%d") at diagnostic.c:534 #1 0x085f4b11 in tree_check_failed (node=0x402892c0, file=0x87b4728 "../../../src/gcc-unpatched/gcc/tree-into-ssa.c", line=466, function=0x87b4fcd "is_old_name") at tree.c:5854 #2 0x0825a791 in is_old_name (name=0x402892c0) at tree-into-ssa.c:466 #3 0x0825aba0 in maybe_replace_use (use_p=0x4031c234) at tree-into-ssa.c:1383 #4 0x082560a2 in rewrite_update_stmt (walk_data=0xbfffeb70, bb=0x40285730, si= {tsi = {ptr = 0x4023c6c0, container = 0x4027cf00}, bb = 0x40285730}) at tree-into-ssa.c:1471 #5 0x08288101 in walk_dominator_tree (walk_data=0xbfffeb70, bb=0x40285730) at domwalk.c:196 #6 0x0828817c in walk_dominator_tree (walk_data=0xbfffeb70, bb=0x40285690) at domwalk.c:212 #7 0x0825650c in rewrite_blocks (entry=0x40285690, what=REWRITE_UPDATE, blocks=0x88f4750) at tree-into-ssa.c:1617 #8 0x08258bbd in update_ssa (update_flags=128) at tree-into-ssa.c:2799 #9 0x0861bb82 in execute_todo (pass=0x8851660, flags=151, use_required=0 '\0') at passes.c:701 (gdb) print *pass $1 = {name = 0x87b655c "alias", gate = 0, execute = 0x8270142 <compute_may_aliases>, sub = 0x0, next = 0x8854100, static_pass_number = 22, tv_id = 33, properties_required = 92, properties_provided = 604, properties_destroyed = 0, todo_flags_start = 0, todo_flags_finish = 151, letter = 0 '\0'} the ICE vanishes if -fdump-tree-original, i.e. dumping before the ICE, and even if dumping after the ICE with f.i. -fdump-tree-vars. GCAC checking doesn't complain about anything, so does valgrind.
Created attachment 9472 [details] even more minimized 23310 testcase
MMIX is not a primary of secondary platform; removing target milestone.
This shows up on more than mmix, I should have marked this as such.
This has been failing on x86/linux for the last two months. Any ETA on a fix? If not, is there a patch that can get reverted to fix this? -benjamin
(In reply to comment #20) > This has been failing on x86/linux for the last two months. Any ETA on a fix? The issue is that this is very dependent on what is built. > If not, is there a patch that can get reverted to fix this? I almost think this is a C++ front-end bug with some GC bug but I cannot prove it. In some cases 1.cc also fails too with the same ICE.
Note that the reduced testcase from pinskia no longer "works" if supplying -fdump-treee-alias1.
Note also that removing any of the (unused) typedef's is causing the failure to go away, so it looks like something may be overflowing, like a hashtable or an array. The symptom is a variable not marked for renaming during/after alias1 pass. The key seem to be UIDs -- deleting code from the testcase and sticking in enough typedef int FooXX at the right places makes the ICE stay. Very weird stuff.
I think this is memory-related as well. On 2gig RAM, this fails reliably. On 512Mb, no fail.
I get the same bug on darwin8.2.0, with 768MB of ram
Present on my SPARC/Solaris 8 machine, but neither on the SPARC/Solaris 9 nor on the SPARC/Solaris 10 machines, while the 3 compilers are essentially identical.
For cris-axis-elf, 1.cc and 2.cc fail with the same error at LAST_UPDATED "Tue Oct 11 22:12:26 UTC 2005".
Subject: Bug 22444 CVSROOT: /cvs/gcc Module name: gcc Changes by: dberlin@gcc.gnu.org 2005-10-17 02:28:44 Modified files: gcc : ChangeLog tree-ssa-alias.c Added files: gcc/testsuite/g++.dg/tree-ssa: pr22444.C Log message: 2005-10-16 Daniel Berlin <dberlin@dberlin.org> Fix PR tree-optimization/22444 * tree-ssa-alias.c (compute_flow_insensitive_aliasing): Assert that we don't hit something with subvars. (setup_pointers_and_addressables): Don't add things with subvars, because we'll already process the subvars for aliasing purposes. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.10167&r2=2.10168 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-alias.c.diff?cvsroot=gcc&r1=2.111&r2=2.112 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/g++.dg/tree-ssa/pr22444.C.diff?cvsroot=gcc&r1=NONE&r2=1.1
Should be fixed now