As discussed in community meeting, the goal was to have core parsing in Windows EAL 20.02. Given that there is a crash and a doubt on the imported getopt library, I think it's better to postpone this series to 20.05.
06/02/2020 07:41, Dmitry Kozlyuk: > > On 2/5/2020 4:39 PM, Pallavi Kadam wrote: > > > On 2/5/2020 11:54 AM, Dmitry Kozlyuk wrote: > > >> Crashes at argument parsing, WinDbg log attached. > > > This patch works fine with meson=0.49. > > > I was able to execute the 'helloworld' app. > > > > > > When I tried to build with meson=0.52 it failed with the error: > > > [9/25] Linking target lib/librte_kvargs-20.0.dll. > > > FAILED: lib/librte_kvargs-20.0.dll > > > clang @lib/librte_kvargs-20.0.dll.rsp > > > clang: error: no such file or directory: '/OPT:REF' > > > > Fixed this error with the patch you sent before. > > were able to execute parsing: > > > > $ ./build/examples/dpdk-helloworld.exe -l 4-8 > > EAL: Detected 20 lcore(s) > > EAL: Detected 2 NUMA nodes > > hello from core 5 > > hello from core 6 > > hello from core 7 > > hello from core 8 > > hello from core 4 > > > > > > > > Not completely sure, if this could be the reason for the crash. > > The crash does not happen with any arguments, "-l 4-8" works fine as do > "-cf" and "-v", but "--log-level=eal:8" crashes reproducibly. I assume > application should not crash on invalid options, but report an error and exit. > For the reference, I applied your patchset to the latest dpdk:master. > > Z:\>build\native\clang\examples\dpdk-helloworld.exe > EAL: Detected 4 lcore(s) > EAL: Detected 1 NUMA nodes > > OK, no lcores specified. > > Z:\>build\native\clang\examples\dpdk-helloworld.exe -cf > EAL: Detected 4 lcore(s) > EAL: Detected 1 NUMA nodes > hello from core 1 > hello from core 2 > hello from core 3 > hello from core 0 > > OK, 4 lcores enabled. > > Z:\>build\native\clang\examples\dpdk-helloworld.exe -cf --log-level=eal:8 > > Crash, WinDbg log attached to bug report. > > Z:\>echo %errorlevel% > -1073741819 > > Z:\> > > Attaching the build log, but I don't think it matters, it's something with > getopt implementation or it's use, judging by stack trace. > > P.S. I can also see "--syslog" option in the help message, you should > probably hide it via #ifndef, as you did with said option handling.
