Affects versions down to 4.9 (configured with --enable-checking=yes) : $ cat z1.c void foo () {} #pragma weak foo = _foo int _foo = 0; $ gcc-7-20170226 -c z1.c z1.c:1:6: error: 'foo' defined both normally and as 'alias' attribute void foo () {} ^~~ z1.c:1:6: error: 'foo' alias in between function and variable is not supported z1.c:3:5: warning: '_foo' aliased declaration int _foo = 0; ^~~~ z1.c:1:6: internal compiler error: Segmentation fault void foo () {} ^~~ 0xbf9f9f crash_signal ../../gcc/toplev.c:337 0x7f25e2 cgraph_node::get(tree_node const*) ../../gcc/cgraph.h:1261 0x7f25e2 cgraph_node::analyze() ../../gcc/cgraphunit.c:630 0x7f5d93 analyze_functions ../../gcc/cgraphunit.c:1118 0x7f68da symbol_table::finalize_compilation_unit() ../../gcc/cgraphunit.c:2599
Started with r199577.
Forgot to set Target Milestone.
GCC 6 branch is being closed
The GCC 7 branch is being closed, re-targeting to GCC 8.4.
GCC 8.4.0 has been released, adjusting target milestone.
Does not happen on gcc 9.2. Does the gcc 8.5 branch have these functions, cgraph_function_or_thunk_node or varpool_variable_node as those were removed on trunk and I'm suspecting that's were the problem is.
This does happen even with 9.2 and current trunk, you just need to read the first line in #c0.
(In reply to Jakub Jelinek from comment #7) > This does happen even with 9.2 and current trunk, you just need to read the > first line in #c0. While I ran it as on a Ubuntu 9.2 Toolchain configured as: Using built-in specs. COLLECT_GCC=gcc COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/9/lto-wrapper OFFLOAD_TARGET_NAMES=nvptx-none:hsa OFFLOAD_TARGET_DEFAULT=1 Target: x86_64-linux-gnu Configured with: ../src/configure -v --with-pkgversion='Ubuntu 9.2.1-9ubuntu2' --with-bugurl=file:///usr/share/doc/gcc-9/README.Bugs --enable-languages=c,ada,c++,go,brig,d,fortran,objc,obj-c++,gm2 --prefix=/usr --with-gcc-major-version-only --program-suffix=-9 --program-prefix=x86_64-linux-gnu- --enable-shared --enable-linker-build-id --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --libdir=/usr/lib --enable-nls --enable-bootstrap --enable-clocale=gnu --enable-gcc -c test.c-debug --enable-libstdcxx-time=yes --with-default-libstdcxx-abi=new --enable-gnu-unique-object --disable-vtable-verify --enable-plugin --enable-default-pie --with-system-zlib --with-target-system-zlib=auto --enable-multiarch --disable-werror --with-arch-32=i686 --with-abi=m64 --with-multilib-list=m32,m64,mx32 --enable-multilib --with-tune=generic --enable-offload-targets=nvptx-none,hsa --without-cuda-driver --enable-checking=release --build=x86_64-linux-gnu --host=x86_64-linux-gnu --target=x86_64-linux-gnu Thread model: posix gcc version 9.2.1 20191008 (Ubuntu 9.2.1-9ubuntu2) like this: gcc -c test.c and this is my source code: void foo () {} 2 #pragma weak foo = _foo 3 int _foo = 0; and I get this: test.c:1:6: error: ‘foo’ defined both normally and as ‘alias’ attribute 1 | void foo () {} | ^~~ test.c:1:6: error: ‘foo’ alias between function and variable is not supported test.c:3:5: note: aliased declaration here 3 | int _foo = 0; | ^~~~ test.c:1: confused by earlier errors, bailing out So no it does not appear to segfault on 9.2 not sure about trunk.
Sorry source code is: void foo () {} #pragma weak foo = _foo int _foo = 0; Copy and Pasted line numbers by mistake.
--enable-checking=release (the default on release branches) is not --enable-checking=yes (which is needed to reproduce this checking error-recovery failure).
(In reply to Nicholas Krause from comment #8) > test.c:1: confused by earlier errors, bailing out > > So no it does not appear to segfault on 9.2 not sure about trunk. YES it does. Just GCC is being smart here when configured with --enable-checking=release (the default for releases) and just prints out "confused by earlier errors, bailing out" for all ICEs after an error has happened. This is why Jakub's comment #7: "This does happen even with 9.2 and current trunk, you just need to read the first line in #c0." The first line in #c0 was: Affects versions down to 4.9 (configured with --enable-checking=yes) Which means you need --enable-checking=yes (and not --enable-checking=release) to get the full ICE happening.
Sorry about not reading the comment carefully. I'm not if this helps but after looking at the change to this file: https://gcc.gnu.org/viewcvs/gcc/trunk/gcc/cgraph.c?r1=199577&r2=199576&pathrev=199577 n = cgraph_create_function_alias (alias, decl); was never changed for creating with the new function arguments as found here: struct cgraph_node * cgraph_create_function_alias (tree alias, tree target) and therefore it should be: n = cgraph_create_function_alias (alias, target); and not decl.
Please forget the previous comment in verify_edge_corresponds_to_fndecl we need to switch it to this: if (e->callee->former_clone_of != node->symbol.decl && (!n->symbol.cpp_implicit_alias || e->callee->former_clone_of != alias_node->symbol.alias_target )) from the current: if (e->callee->former_clone_of != node->symbol.decl as those other cases are missing in the patch posted after looking at it.
Created attachment 48052 [details] Patch for this bug
(In reply to Nicholas Krause from comment #14) > Created attachment 48052 [details] > Patch for this bug After tracking this down to the new function that replaced verify_edge_corresponds_to_fndecl, cgraph_edge::verify_corresponds_to_fndecl which is basically the same we run into this: if (callee->former_clone_of != node->decl && (node != callee->ultimate_alias_target ()) && !clone_of_p (node, callee)) Basically it appears to be checking we are a former cgraph clone at the same time as being a current clone or the same node with clone_of_p on the callee cgraph node. I'm attaching a test patch for this. Gerhard does this fix the issue?
That patch will generate a -Wparentheses warning, and more importantly, doesn't fix the ICE, which you could have tried by yourself.
Sorry about that. I've managed to track it down to a problem with a alias issue And from reading the code in the original cvs commit. I'm not able to read it currently but the original function mentioned seems to now be cgraph_edge::verify_corresponds_to_fndecl. I've tried to access the original commit but it complains about permissions. Maybe I'm wrong but this seems very odd to me: node = node->ultimate_alias_target (); is basically referring to itself meaning: if (callee->former_clone_of != node->decl && (node != node->ultimate_alias_target ()) && !clone_of_p (node, callee)) the && (node != node->ultimate_alias_target ()) is basically referring to itself when comparing yet its already being defined equal to the return value of ultimate_alias_target meaning its always true. Not sure if this fixes this bug or something else. Thoughts?
GCC 8 branch is being closed.
GCC 9.4 is being released, retargeting bugs to GCC 9.5.
GCC 9 branch is being closed
GCC 10.4 is being released, retargeting bugs to GCC 10.5.
GCC 10 branch is being closed.