Re: [jakub@redhat.com: [3.3.2 PATCH] Backport invalid C++ tree sharing fix + testcase]
I'll prepare another upload on Sunday.
Daniel Jacobowitz writes:
> I'm pretty sure this is the patch we need.
>
> - Forwarded message from Jakub Jelinek <[EMAIL PROTECTED]> -
>
> Date: Wed, 1 Oct 2003 20:35:19 +0200
> From: Jakub Jelinek <[EMAIL PROTECTED]>
> Subject: [3.3.2 PATCH] Backport invalid C++ tree sharing fix + testcase
> To: Mark Mitchell <[EMAIL PROTECTED]>
> Cc: [EMAIL PROTECTED]
> Reply-To: Jakub Jelinek <[EMAIL PROTECTED]>
>
> Hi!
>
> The following testcase is miscompiled on IA-32 (foo has an endless loop)
> in 3.3.x.
> It is a regression from 3.2.x and fixed again on the trunk.
> I tracked it down to:
> http://gcc.gnu.org/ml/gcc-patches/2003-07/msg01063.html
> Below is backport of the patch to gcc-3_3-branch.
> Ok to commit?
>
> 2003-07-09 Mark Mitchell <[EMAIL PROTECTED]>
>
> * cp-tree.h (break_out_calls): Remove declaration.
> * tree.c (break_out_calls): Remove.
> * typeck.c (build_modify_expr): Avoid invalid sharing of trees.
>
> 2003-10-01 Jakub Jelinek <[EMAIL PROTECTED]>
>
> * g++.dg/opt/cond1.C: New test.
>
> --- gcc/testsuite/g++.dg/opt/cond1.C.jj 2003-09-15 15:40:47.0
> +0200
> +++ gcc/testsuite/g++.dg/opt/cond1.C 2003-10-01 14:39:32.0 +0200
> @@ -0,0 +1,24 @@
> +// { dg-do run }
> +// { dg-options "-O2" }
> +
> +struct D { int x; };
> +struct W
> +{
> + W () {}
> + D & operator * () { return d; }
> + D d;
> +};
> +
> +int
> +foo (int y)
> +{
> + W m;
> + (*m).x = (y > 1 ? y : 0);
> + return (*m).x;
> +}
> +
> +int
> +main ()
> +{
> + return (foo (6) != 6);
> +}
> --- gcc/cp/cp-tree.h.jj 2003-10-01 12:08:54.0 +0200
> +++ gcc/cp/cp-tree.h 2003-10-01 22:23:33.0 +0200
> @@ -4317,7 +4317,6 @@ extern tree build_min PARAMS
> ((enum t
> extern tree build_min_nt PARAMS ((enum tree_code, ...));
> extern tree build_cplus_new PARAMS ((tree, tree));
> extern tree get_target_expr PARAMS ((tree));
> -extern tree break_out_calls PARAMS ((tree));
> extern tree build_cplus_method_type PARAMS ((tree, tree, tree));
> extern tree build_cplus_staticfn_typePARAMS ((tree, tree,
> tree));
> extern tree build_cplus_array_type PARAMS ((tree, tree));
> --- gcc/cp/typeck.c.jj2003-09-16 17:25:55.0 +0200
> +++ gcc/cp/typeck.c 2003-10-01 22:23:06.0 +0200
> @@ -5502,44 +5502,8 @@ build_modify_expr (lhs, modifycode, rhs)
>if (newrhs == error_mark_node)
> return error_mark_node;
>
> - if (TREE_CODE (newrhs) == COND_EXPR)
> -{
> - tree lhs1;
> - tree cond = TREE_OPERAND (newrhs, 0);
> -
> - if (TREE_SIDE_EFFECTS (lhs))
> - cond = build_compound_expr (tree_cons
> - (NULL_TREE, lhs,
> - build_tree_list (NULL_TREE, cond)));
> -
> - /* Cannot have two identical lhs on this one tree (result) as preexpand
> - calls will rip them out and fill in RTL for them, but when the
> - rtl is generated, the calls will only be in the first side of the
> - condition, not on both, or before the conditional jump! (mrs) */
> - lhs1 = break_out_calls (lhs);
> -
> - if (lhs == lhs1)
> - /* If there's no change, the COND_EXPR behaves like any other rhs. */
> - result = build (modifycode == NOP_EXPR ? MODIFY_EXPR : INIT_EXPR,
> - lhstype, lhs, newrhs);
> - else
> - {
> - tree result_type = TREE_TYPE (newrhs);
> - /* We have to convert each arm to the proper type because the
> - types may have been munged by constant folding. */
> - result
> - = build (COND_EXPR, result_type, cond,
> - build_modify_expr (lhs, modifycode,
> - cp_convert (result_type,
> - TREE_OPERAND (newrhs, 1))),
> - build_modify_expr (lhs1, modifycode,
> - cp_convert (result_type,
> - TREE_OPERAND (newrhs, 2;
> - }
> -}
> - else
> -result = build (modifycode == NOP_EXPR ? MODIFY_EXPR : INIT_EXPR,
> - lhstype, lhs, newrhs);
> + result = build (modifycode == NOP_EXPR ? MODIFY_EXPR : INIT_EXPR,
> + lhstype, lhs, newrhs);
>
>TREE_SIDE_EFFECTS (result) = 1;
>
> --- gcc/cp/tree.c.jj 2003-05-16 00:06:44.0 +0200
> +++ gcc/cp/tree.c 2003-10-01 22:24:46.0 +0200
> @@ -375,97 +375,6 @@ get_target_expr (init)
> {
>return build_target_expr_with_type (init, TREE_TYPE (init));
> }
> -
> -/* Recursively perform a preorder search EXP for CALL_EXPRs, making
> - copies where they are found. Returns a deep copy all nodes transitively
> - containing CALL_EXPRs. */
> -
> -tree
> -break_out_calls (exp)
> - tree exp;
> -{
> - register tree t1, t2
Bug#128950: [Bug c/9072] -Wconversion should be split into two distinct flags
PLEASE REPLY TO [EMAIL PROTECTED] ONLY, *NOT* [EMAIL PROTECTED] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9072 Jay dot St dot Pierre at Colorado dot EDU changed: What|Removed |Added CC||Jay dot St dot Pierre at ||Colorado dot EDU --- Additional Comments From Jay dot St dot Pierre at Colorado dot EDU 2003-10-03 15:24 --- Note that this is still an issue with 3.3.1. --- You are receiving this mail because: --- You reported the bug, or are watching the reporter.
Bug#213994: gcc-3.3: Please provide an overflow protection in the stock compiler
Package: gcc-3.3 Version: 1:3.3.2-0pre4 Priority: wishlist Tags: upstream I would very much like stack/buffer overflow protection such as (StackGuard [1] or to Stack Smashing Protector [2]) be available in Debian's stock gcc-3.3 compiler. The people at Immunix are making an effort to integrate StackGuard's latest version (3) into gcc, so maybe this will come by itself. Even if Debian packages might not be compiled with this protection, it would make it easier for Debian-derived distributions to recompile packages if Debian's compilers already featured this option. Regards Javi [1] http://www.immunix.org/stackguard.html Used to build the Immunix distribution. [2] http://www.research.ibm.com/trl/projects/security/ssp/ Used to build the Adamantix Debian-derived distribution. pgpASJrnE21Vl.pgp Description: PGP signature
Bug#214024: g++: when configure scripts are over 32768 g++ fails to compile test files
Package: g++-3.3 Version: 1:3.3.2-0pre4 Severity: grave Adam Heath recommended I file this bug against gcc/g++ rather than autoconf. The issue is as follows: configure is over 32767 lines long configure generates a conftest.cc file in that conftest.cc file is the line '#line 32780 "configure"' since the number is greater than 32767 g++ fails saying: conftest.cc:2:7: line number out of range which causes the test to fail and then configure thinks option is unavailable. configure:32757: checking for getdomainname configure:32807: g++ -c -Wnon-virtual-dtor -Wno-long-long -Wundef -Wall -pedantic -W -Wpointer-arith -Wwrite-string s -ansi -D_XOPEN_SOURCE=500 -D_BSD_SOURCE -Wcast-align -Wconversion -DNDEBUG -DNO_DEBUG -O2 -fno-exceptions -fno-ch eck-new -pedantic-errors -DQT_THREAD_SUPPORT -D_REENTRANT conftest.cc >&5 conftest.cc:1:7: line number out of range configure:32810: $? = 1 configure: failed program was: | #line 32780 "configure" | /* confdefs.h. */ ... This happens on my new arts build which has updated autotools, before updating autotools the configure script was small enough that the 32767 limit didn't affect it. Chris Cheney -- System Information: Debian Release: testing/unstable Architecture: i386 Kernel: Linux calc 2.4.22-bk18 #1 SMP Mon Sep 15 12:55:07 CDT 2003 i686 Locale: LANG=en_US, LC_CTYPE=en_US Versions of packages g++-3.3 depends on: ii gcc-3.31:3.3.2-0pre4 The GNU C compiler ii gcc-3.3-base 1:3.3.2-0pre4 The GNU Compiler Collection (base ii libc6 2.3.2-8 GNU C Library: Shared libraries an ii libstdc++5-3.3-dev 1:3.3.2-0pre4 The GNU Standard C++ Library v3 (d -- no debconf information
Bug#214024: marked as done (g++: when configure scripts are over 32768 g++ fails to compile test files)
Your message dated Fri, 3 Oct 2003 20:09:23 -0500 with message-id <[EMAIL PROTECTED]> and subject line closing bug has caused the attached Bug report to be marked as done. This means that you claim that the problem has been dealt with. If this is not the case it is now your responsibility to reopen the Bug report if necessary, and/or fix the problem forthwith. (NB: If you are a system administrator and have no idea what I am talking about this indicates a serious mail system misconfiguration somewhere. Please contact me immediately.) Debian bug tracking system administrator (administrator, Debian Bugs database) -- Received: (at submit) by bugs.debian.org; 3 Oct 2003 23:57:47 + >From [EMAIL PROTECTED] Fri Oct 03 18:57:46 2003 Return-path: <[EMAIL PROTECTED]> Received: from (pico.surpasshosting.com) [64.5.56.18] by master.debian.org with esmtp (Exim 3.35 1 (Debian)) id 1A5Znu-0003xH-00; Fri, 03 Oct 2003 18:57:46 -0500 Received: from conr-adsl-cheney.txucom.net ([207.70.165.48] helo=calc) by pico.surpasshosting.com with esmtp (TLSv1:RC4-SHA:128) (Exim 4.24) id 1A5Znl-0002H1-Iz; Fri, 03 Oct 2003 18:57:37 -0500 Received: from ccheney by calc with local (Exim 4.22) id 1A5Znt-0005fZ-2W; Fri, 03 Oct 2003 18:57:45 -0500 Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit From: Chris Cheney <[EMAIL PROTECTED]> To: Debian Bug Tracking System <[EMAIL PROTECTED]> Subject: g++: when configure scripts are over 32768 g++ fails to compile test files X-Mailer: reportbug 2.32 Date: Fri, 03 Oct 2003 18:57:44 -0500 Message-Id: <[EMAIL PROTECTED]> Sender: <[EMAIL PROTECTED]> X-AntiAbuse: This header was added to track abuse, please include it with any abuse report X-AntiAbuse: Primary Hostname - pico.surpasshosting.com X-AntiAbuse: Original Domain - bugs.debian.org X-AntiAbuse: Originator/Caller UID/GID - [0 0] / [47 12] X-AntiAbuse: Sender Address Domain - conr-adsl-cheney.txucom.net Delivered-To: [EMAIL PROTECTED] X-Spam-Status: No, hits=-6.5 required=4.0 tests=BAYES_01,HAS_PACKAGE version=2.53-bugs.debian.org_2003_10_03 X-Spam-Level: X-Spam-Checker-Version: SpamAssassin 2.53-bugs.debian.org_2003_10_03 (1.174.2.15-2003-03-30-exp) Package: g++-3.3 Version: 1:3.3.2-0pre4 Severity: grave Adam Heath recommended I file this bug against gcc/g++ rather than autoconf. The issue is as follows: configure is over 32767 lines long configure generates a conftest.cc file in that conftest.cc file is the line '#line 32780 "configure"' since the number is greater than 32767 g++ fails saying: conftest.cc:2:7: line number out of range which causes the test to fail and then configure thinks option is unavailable. configure:32757: checking for getdomainname configure:32807: g++ -c -Wnon-virtual-dtor -Wno-long-long -Wundef -Wall -pedantic -W -Wpointer-arith -Wwrite-string s -ansi -D_XOPEN_SOURCE=500 -D_BSD_SOURCE -Wcast-align -Wconversion -DNDEBUG -DNO_DEBUG -O2 -fno-exceptions -fno-ch eck-new -pedantic-errors -DQT_THREAD_SUPPORT -D_REENTRANT conftest.cc >&5 conftest.cc:1:7: line number out of range configure:32810: $? = 1 configure: failed program was: | #line 32780 "configure" | /* confdefs.h. */ ... This happens on my new arts build which has updated autotools, before updating autotools the configure script was small enough that the 32767 limit didn't affect it. Chris Cheney -- System Information: Debian Release: testing/unstable Architecture: i386 Kernel: Linux calc 2.4.22-bk18 #1 SMP Mon Sep 15 12:55:07 CDT 2003 i686 Locale: LANG=en_US, LC_CTYPE=en_US Versions of packages g++-3.3 depends on: ii gcc-3.31:3.3.2-0pre4 The GNU C compiler ii gcc-3.3-base 1:3.3.2-0pre4 The GNU Compiler Collection (base ii libc6 2.3.2-8 GNU C Library: Shared libraries an ii libstdc++5-3.3-dev 1:3.3.2-0pre4 The GNU Standard C++ Library v3 (d -- no debconf information --- Received: (at 214024-done) by bugs.debian.org; 4 Oct 2003 01:09:25 + >From [EMAIL PROTECTED] Fri Oct 03 20:09:24 2003 Return-path: <[EMAIL PROTECTED]> Received: from (pico.surpasshosting.com) [64.5.56.18] by master.debian.org with esmtp (Exim 3.35 1 (Debian)) id 1A5avE-ZY-00; Fri, 03 Oct 2003 20:09:24 -0500 Received: from conr-adsl-cheney.txucom.net ([207.70.165.48] helo=calc) by pico.surpasshosting.com with esmtp (TLSv1:RC4-SHA:128) (Exim 4.24) id 1A5av5-0005KZ-Bw for [EMAIL PROTECTED]; Fri, 03 Oct 2003 20:09:15 -0500 Received: from ccheney by calc with local (Exim 4.22) id 1A5avD-0002DE-KQ for [EMAIL PROTECTED]; Fri, 03 Oct 2003 20:09:23 -0500 Date: Fri, 3 Oct 2003 20:09:23 -0500 From: Chris Cheney <[EMAIL PROTECTED]> To: [EMAIL PROTECTED] Subject: closing bug Message-ID: <[EMAIL PROTECTED]> Mime-Version: 1.0 Content-Type: multipart/signed; m
