From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 17113 invoked by alias); 22 Jul 2002 18:43:00 -0000 Mailing-List: contact gdb-patches-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sources.redhat.com Received: (qmail 17106 invoked from network); 22 Jul 2002 18:42:59 -0000 Received: from unknown (HELO jackfruit.Stanford.EDU) (171.64.38.136) by sources.redhat.com with SMTP; 22 Jul 2002 18:42:59 -0000 Received: (from carlton@localhost) by jackfruit.Stanford.EDU (8.11.6/8.11.6) id g6MIgwM25521; Mon, 22 Jul 2002 11:42:58 -0700 From: david carlton MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <15676.21042.125481.349851@jackfruit.Stanford.EDU> Date: Mon, 22 Jul 2002 12:02:00 -0000 To: gdb-patches@sources.redhat.com Subject: patch for PR gdb/574 Cc: carlton@math.stanford.edu X-SW-Source: 2002-07/txt/msg00437.txt.bz2 For various reasons, I decided to delve into GDB recently, so I picked bug 574 somewhat at random and decided to try to fix it. Here's a patch. The problem: GDB tries to access an improper memory location while trying to find run time type info on certain code compiled with G++ 2.something. In the situation in question, GDB is trying to find the vptr of an object that is a subclass of a class with virtual functions and static member data. It looks for the information about the vptr at a place where information about the static member data is actually stored. This is a data structure with a useless (and very large) bitpos field; adding this bitpos to a pointer causes that pointer to point to the middle of nowhere. Oops. The solution: Get rid of the check for if (VALUE_ENCLOSING_TYPE(v) != VALUE_TYPE(v)) and the surrounding code in gnu-v2-abi.c(gnuv2_value_rtti_type). Some comments: * gnu-v3-abi.c(gnuv3_rtti_type) doesn't have any such code, so I don't think this fix is likely to break anything that isn't also broken in gnu-v3-abi.c. * It doesn't seem to cause any testsuites to fail that didn't fail before. (In fact, when I ran it, it turned gdb.base/selftest.exp from a FAIL into a pass, for what that's worth; I don't know if my changes are relevant to that, though.) I'm not too used to working the testsuites; please rerun them to make sure. * In the test case in question, the only reason why that check matched was because a call to value_cast earlier in the function set the enclosing type to differ from the type. * There are other problems in the surrounding code that are unrelated to this. When I compiled my test program under g++-3.1, I got an unrelated bad memory access from inside gnu-v3-abi.c. Also (and this may be related to that bad memory access), there seems to be a difference of opinion among various pieces of code as to whose job it is to demangle symbols that you want to look up in symbol tables. I'll try to spend some time over the next week or two sorting this out. * I'm not convinced that the code I deleted wouldn't be useful in some circumstances (where we had a pointer to something in the middle of a class); I'll try to spend some time over the next week or two coming up with test cases to see cause the code to fail, and to see if there's a better fix around somewhere. Nonetheless, given that my patch does fix a known problem and, if there is a problem with enclosing_type, then the current gnu-v3-abi.c code probably has it as well, I think it should be committed. * I haven't contacted the submitter of the bug report, but I'd be happy to do so if that is appropriate. I'm including a sample program that demonstrates the bug (though the submitter of the bug report also submitted a more elaborate test case that has some other interesting features; hurrah for good bug reports), a ChangeLog entry, and the patch; all will follow my signature. David Carlton carlton@math.stanford.edu /* An attempt to replicate GDB bug 574 with a shorter program. g++ -g test.cc # no optimization, else we'd need to split into 2 files gdb a.out (gdb) b main (gdb) r (gdb) n (gdb) p *theB */ class A { public: virtual void foo() {}; // Stick in a virtual function. static const int bar; // Stick in a static data member. }; class B : public A { static const int bar = 1; }; int main() { B *theB = new B; } 2002-07-22 david carlton * gnu-v2-abi.c (gnuv2_value_rtti_type): Eliminate test for being enclosed. Fix PR gdb/574. Index: gnu-v2-abi.c =================================================================== RCS file: /cvs/src/src/gdb/gnu-v2-abi.c,v retrieving revision 1.6 diff -c -p -r1.6 gnu-v2-abi.c *** gnu-v2-abi.c 4 Jan 2002 18:20:19 -0000 1.6 --- gnu-v2-abi.c 22 Jul 2002 18:22:05 -0000 *************** gnuv2_value_rtti_type (struct value *v, *** 189,195 **** struct type *rtti_type; CORE_ADDR coreptr; struct value *vp; - int using_enclosing = 0; long top_offset = 0; char rtti_type_name[256]; CORE_ADDR vtbl; --- 189,194 ---- *************** gnuv2_value_rtti_type (struct value *v, *** 244,268 **** if (VALUE_ADDRESS (value_field (v, TYPE_VPTR_FIELDNO (known_type))) == 0) return NULL; ! /* ! If we are enclosed by something that isn't us, adjust the ! address properly and set using_enclosing. ! */ ! if (VALUE_ENCLOSING_TYPE(v) != VALUE_TYPE(v)) ! { ! struct value *tempval; ! int bitpos = TYPE_BASECLASS_BITPOS (known_type, ! TYPE_VPTR_FIELDNO (known_type)); ! tempval=value_field (v, TYPE_VPTR_FIELDNO(known_type)); ! VALUE_ADDRESS(tempval) += bitpos / 8; ! vtbl=value_as_address (tempval); ! using_enclosing=1; ! } ! else ! { ! vtbl=value_as_address(value_field(v,TYPE_VPTR_FIELDNO(known_type))); ! using_enclosing=0; ! } /* Try to find a symbol that is the vtable */ minsym=lookup_minimal_symbol_by_pc(vtbl); --- 243,249 ---- if (VALUE_ADDRESS (value_field (v, TYPE_VPTR_FIELDNO (known_type))) == 0) return NULL; ! vtbl=value_as_address(value_field(v,TYPE_VPTR_FIELDNO(known_type))); /* Try to find a symbol that is the vtable */ minsym=lookup_minimal_symbol_by_pc(vtbl); *************** gnuv2_value_rtti_type (struct value *v, *** 304,311 **** if (full) *full=1; } - if (using_enc) - *using_enc=using_enclosing; return rtti_type; } --- 285,290 ----