Ubuntu 16 i686: case value cannot be narrowed to type '__fsword_t' (aka 'int') ; result of comparison, 'uintptr_t' (aka 'unsigned int') is always false

[advancedwebdeveloper]

Hi.

Build failed on

$ uname -a
Linux mylaptop 4.15.0-122-generic #124~16.04.1-Ubuntu SMP Thu Oct 15 16:14:02 UTC 2020 i686 i686 i686 GNU/Linux

with

$ clang-11 --version
Ubuntu clang version 11.0.0-++20200721055954+cebd637c886-1exp120200721161335.13
Target: i686-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

and

$ cmake --version
cmake version 3.19.0-rc2

.
My generator is configured for

$ ninja --version
1.10.0.git.kitware.jobserver-1

So I got these errors:

$ ninja
[2/431] Building CXX object tools/tracegen/CMakeFiles/trace.dir/Port.cpp.o
FAILED: tools/tracegen/CMakeFiles/trace.dir/Port.cpp.o
/usr/bin/clang++-11 -DJ9X86 -DLINUX -D_FILE_OFFSET_BITS=64 -D_X86_ -pthread -fno-strict-aliasing -m32 -msse2 -std=c++0x -fPIC -Werror -Wall -MD -MT tools/tracegen/CMakeFiles/trace.dir/Port.cpp.o -MF tools/tracegen/CMakeFiles/trace.dir/Port.cpp.o.d -o tools/tracegen/CMakeFiles/trace.dir/Port.cpp.o -c /home/oceanfish81/omr/tools/tracegen/Port.cpp
/home/oceanfish81/omr/tools/tracegen/Port.cpp:691:7: error: case value evaluates to 4283649346, which cannot be narrowed to type '__fsword_t' (aka 'int') [-Wc++11-narrowing]
case 0xFF534D42: /* CIFS_MAGIC_NUMBER */
^
1 error generated.
ninja: build stopped: subcommand failed.

$ ninja
[76/408] Building C object port/CMakeFiles/omrport_obj.dir/linux/omrvmem.c.o
FAILED: port/CMakeFiles/omrport_obj.dir/linux/omrvmem.c.o
/usr/bin/clang-11 -DJ9X86 -DLINUX -DOMRPORT_LIBRARY_DEFINE -D_FILE_OFFSET_BITS=64 -D_X86_ -I/home/oceanfish81/omr/port -Iport -I/home/oceanfish81/omr/include_core -I. -I/home/oceanfish81/omr/port/linux386 -I/home/oceanfish81/omr/port/linux -I/home/oceanfish81/omr/port/unix -I/home/oceanfish81/omr/port/unix_include -I/home/oceanfish81/omr/port/common -I/home/oceanfish81/omr/port/include -I/home/oceanfish81/omr/port/../nls -pthread -fno-strict-aliasing -m32 -msse2 -fPIC -Werror -Wall -MD -MT port/CMakeFiles/omrport_obj.dir/linux/omrvmem.c.o -MF port/CMakeFiles/omrport_obj.dir/linux/omrvmem.c.o.d -o port/CMakeFiles/omrport_obj.dir/linux/omrvmem.c.o -c /home/oceanfish81/omr/port/linux/omrvmem.c
/home/oceanfish81/omr/port/linux/omrvmem.c:474:23: error: result of comparison of constant 18446744073709551615 with expression of type 'uintptr_t' (aka 'unsigned int') is always false [-Werror,-Wtautological-constant-out-of-range-compare]
if ((ULLONG_MAX == start) && (ERANGE == errno)) {
~~~~~~~~~~ ^ ~~~~~
/home/oceanfish81/omr/port/linux/omrvmem.c:487:23: error: result of comparison of constant 18446744073709551615 with expression of type 'uintptr_t' (aka 'unsigned int') is always false [-Werror,-Wtautological-constant-out-of-range-compare]
if ((ULLONG_MAX == end) && (ERANGE == errno)) {
~~~~~~~~~~ ^ ~~~
2 errors generated.
[91/408] Building CXX object gc/CMakeFiles/omrgc.dir/base/HeapMap.cpp.o
ninja: build stopped: subcommand failed.

[mstoodle]

Apologies for the late response here @advancedwebdeveloper : most of the folks working on OMR don't do much with 32-bit systems and we seem to have lost most of our active port library contributors :( .

32-bit builds using the recommended (ancient) compilers seem to work just fine (now that we've removed a misconfigured machine from the build list). But not so much on the system/build compiler you're using, apparently.

The first problem is because the file system type for CIFS_MAGIC_NUMBER (0xFF534D42) is an unsigned value but the field of the struct that holds the type is a signed type on your system, for whatever reason. The type of that field is an internal type, so I don't want to use it. The best solutions I could find for this problem are either to cast the field to an unsigned int or to cast the constant to a signed int. The fix I'll propose is to cast the constant to signed int as having the least impact.

The second problem is just a type mismatch. On your platform uintptr_t doesn't evaluate to unsigned long long. The values need to be cast to ADDRESS which, on 32-bit, needs to go through uintptr_t anyway, but for the comparison we should be using the raw unsigned long long value returned by strtoull().

I'll put a pull request together and see how people feel about it. I've confirmed that these two changes build just fine inside a 32-bit ubuntu docker container, but since I could not reproduce the original problem it would be helpful if @advancedwebdeveloper could help us validate the patch to make sure we've caught the problems.

[mstoodle]

@advancedwebdeveloper would you please provide the cmake command line you've used, just for reference?

[davidjmccann]

@mstoodle I can hit the warnings highlighted by this issue and can confirm that the two changes you propose gets rid of the warnings.

[mstoodle]

Thanks @davidjmccann! I'll refresh the pull request for cleaner history and then remove the WIP so it can be merged .