https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92364
Bug ID: 92364
Summary: strict-overflow warning only when using "--coverage"
flag and -O2 or above, and inlining
Product: gcc
Version: 6.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: gmc at synopsys dot com
Target Milestone: ---
Created attachment 47170
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47170&action=edit
Source file to be compiled with g++ to demonstrate the bug
This is an issue which is hard to characterize, as it seems that a whole lots
of different things have to line up for it to occur.
When compiling the sample code with g++, a warning will be issued for
'strict-overflow' ONLY when building with -O2, ONLY when building with
--coverage, and only when using inlining. Furthermore, if I make any unrelated
changes to the code, the issue goes away.
This issue exists in GCC 6.2, 7.3, and 7.4, but GCC 8.1 and newer (I tested
8.2, 8.3, 9.1, 9.2) do NOT have the issue
The g++ compiler is configured as follows (GCC 6.2 shown, 7.3 etc. configured
with the same options):
Target: x86_64-redhat-linux
Configured with: ../src/gcc-6.2.0/configure --prefix=/depot/gcc-6.2.0
--enable-bootstrap --enable-threads=posix --disable-checking --with-system-zlib
--enable-__cxa_atexit --disable-libunwind-exceptions
--enable-languages=c,c++,fortran --disable-libgcj --with-cpu=generic
--build=x86_64-redhat-linux
Thread model: posix
gcc version 6.2.0 (GCC)
Compile the testcase as:
g++ -c foo.cc -o /tmp/foo.o -O3 -Wall -Werror -Wno-uninitialized
--coverage
With this reduced testcase, I can flip the compilation / pass / fail status
with a whole range of unlikely changes – e.g.
- If I compile without the "-Wno-uninitialized" flag, then I get THREE
warnings:
+ 1 warning about strict overflow
+ 2 warnings about uninitialized vars thingy.GridMap::Num_Y and
hingy.GridMap::Num_X
then if I initialize those vars in the constructor via
GridMap() { Num_X=0; Num_Y=0; }
then not only do the uninit warnings go away, but so does the strict
overflow one
- Similarly, commenting out unrelated code, makes the warning disappear:
if (my_y > Num_Y) { my_y = Num_Y; /* trim = true; */ }
even though the value of trim has no effect on the following line where the
strict-overflow warning is triggered
- Disabling inlining also makes the issue go away (-fno-inline or set
attribute on funcs)
