February 2016 - devel - mailman3.cels.anl.gov

Re: [mpich-devel] mpirun fails if there is a void environment variable
by Дмитрий 18 Feb '16

18 Feb '16

Hello again, env command output is like this $ env ... XDG_SESSION_PATH=/org/freedesktop/DisplayManager/Session0 = DISTCC_SAVE_TEMPS=0 ... 2016-02-18 20:05 GMT+03:00 Balaji, Pavan <balaji(a)anl.gov>: > > Does your environment not have anything in it? What's the output of "env"? > > -- Pavan > >> On Feb 18, 2016, at 9:08 AM, Дмитрий <dmitrodem(a)gmail.com> wrote: >> >> Hello, >> >> i've found that on my system mpirun /bin/hostname causes segmentation >> fault. Inspecting it in gdb, i've seen the following stack trace: >> >> (gdb) run /bin/hostname >> Starting program: /usr/bin/mpirun /bin/hostname >> [Thread debugging using libthread_db enabled] >> Using host libthread_db library "/lib64/libthread_db.so.1". >> >> Program received signal SIGSEGV, Segmentation fault. >> 0x00007ffff69c484a in ?? () from /lib64/libc.so.6 >> (gdb) bt >> #0 0x00007ffff69c484a in ?? () from /lib64/libc.so.6 >> #1 0x0000000000430455 in HYDTI_bscd_in_env_list (env_name=0x0, >> env_list=env_list@entry=0x7fffffffd890) at >> tools/bootstrap/external/external_common.c:27 >> #2 0x0000000000425ae4 in HYDT_bscd_ssh_query_env_inherit >> (env_name=<optimized out>, ret=0x7fffffffd8e4) at >> tools/bootstrap/external/ssh_env.c:17 >> #3 0x000000000042557c in HYDT_bsci_query_env_inherit >> (env_name=<optimized out>, ret=ret@entry=0x7fffffffd8e4) at >> tools/bootstrap/src/bsci_env.c:18 >> #4 0x00000000004208d3 in HYDU_list_inherited_env >> (env_list=env_list@entry=0x66c680 <HYD_server_info+128>) at >> utils/env/env.c:55 >> #5 0x000000000040404c in main (argc=<optimized out>, argv=<optimized >> out>) at ui/mpich/mpiexec.c:258 >> >> The problem seems to be in strcmp call, which compares NULL (empty >> named environment variable) with something. env command shows string >> "=" for me. >> >> So, I guess it would be nice to check env_name variable before passing >> it to strcmp? >> >> -- >> Regards, >> Dmitriy >> _______________________________________________ >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/devel > > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel -- Regards, Dmitriy

1 0

Re: [mpich-devel] mpirun fails if there is a void environment variable
by Balaji, Pavan 18 Feb '16

18 Feb '16

Does your environment not have anything in it? What's the output of "env"? -- Pavan > On Feb 18, 2016, at 9:08 AM, Дмитрий <dmitrodem(a)gmail.com> wrote: > > Hello, > > i've found that on my system mpirun /bin/hostname causes segmentation > fault. Inspecting it in gdb, i've seen the following stack trace: > > (gdb) run /bin/hostname > Starting program: /usr/bin/mpirun /bin/hostname > [Thread debugging using libthread_db enabled] > Using host libthread_db library "/lib64/libthread_db.so.1". > > Program received signal SIGSEGV, Segmentation fault. > 0x00007ffff69c484a in ?? () from /lib64/libc.so.6 > (gdb) bt > #0 0x00007ffff69c484a in ?? () from /lib64/libc.so.6 > #1 0x0000000000430455 in HYDTI_bscd_in_env_list (env_name=0x0, > env_list=env_list@entry=0x7fffffffd890) at > tools/bootstrap/external/external_common.c:27 > #2 0x0000000000425ae4 in HYDT_bscd_ssh_query_env_inherit > (env_name=<optimized out>, ret=0x7fffffffd8e4) at > tools/bootstrap/external/ssh_env.c:17 > #3 0x000000000042557c in HYDT_bsci_query_env_inherit > (env_name=<optimized out>, ret=ret@entry=0x7fffffffd8e4) at > tools/bootstrap/src/bsci_env.c:18 > #4 0x00000000004208d3 in HYDU_list_inherited_env > (env_list=env_list@entry=0x66c680 <HYD_server_info+128>) at > utils/env/env.c:55 > #5 0x000000000040404c in main (argc=<optimized out>, argv=<optimized > out>) at ui/mpich/mpiexec.c:258 > > The problem seems to be in strcmp call, which compares NULL (empty > named environment variable) with something. env command shows string > "=" for me. > > So, I guess it would be nice to check env_name variable before passing > it to strcmp? > > -- > Regards, > Dmitriy > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel

1 0

mpirun fails if there is a void environment variable
by Дмитрий 18 Feb '16

18 Feb '16

Hello, i've found that on my system mpirun /bin/hostname causes segmentation fault. Inspecting it in gdb, i've seen the following stack trace: (gdb) run /bin/hostname Starting program: /usr/bin/mpirun /bin/hostname [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib64/libthread_db.so.1". Program received signal SIGSEGV, Segmentation fault. 0x00007ffff69c484a in ?? () from /lib64/libc.so.6 (gdb) bt #0 0x00007ffff69c484a in ?? () from /lib64/libc.so.6 #1 0x0000000000430455 in HYDTI_bscd_in_env_list (env_name=0x0, env_list=env_list@entry=0x7fffffffd890) at tools/bootstrap/external/external_common.c:27 #2 0x0000000000425ae4 in HYDT_bscd_ssh_query_env_inherit (env_name=<optimized out>, ret=0x7fffffffd8e4) at tools/bootstrap/external/ssh_env.c:17 #3 0x000000000042557c in HYDT_bsci_query_env_inherit (env_name=<optimized out>, ret=ret@entry=0x7fffffffd8e4) at tools/bootstrap/src/bsci_env.c:18 #4 0x00000000004208d3 in HYDU_list_inherited_env (env_list=env_list@entry=0x66c680 <HYD_server_info+128>) at utils/env/env.c:55 #5 0x000000000040404c in main (argc=<optimized out>, argv=<optimized out>) at ui/mpich/mpiexec.c:258 The problem seems to be in strcmp call, which compares NULL (empty named environment variable) with something. env command shows string "=" for me. So, I guess it would be nice to check env_name variable before passing it to strcmp? -- Regards, Dmitriy

1 0

Re: [mpich-devel] MPICH_ASYNC_PROGRESS
by Balaji, Pavan 11 Feb '16

11 Feb '16

It's still allowed. It's autogenerated, which is why you can't find it with git grep. Note that the supported option is MPIR_CVAR_ASYNC_PROGRESS. MPICH_ASYNC_PROGRESS is an alias that is maintained for backward compatibility reasons. In the future, that option might disappear. -- Pavan > On Feb 11, 2016, at 12:53 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > > README.envvar says you still support it but it appears nowhere in the source code. Are you parsing it via some special magic or are the docs out of sync with the implementation? > > I ask this now because I saw what appears to be a bug in the implementation, but it's stupid to track down that bug if the problem is that MPICH_ASYNC_PROGRESS is not supposed to work anymore. > > Thanks, > > Jeff > > jrhammon-mac01:git jrhammon$ git grep ASYNC_PROGRESS > src/mpi/init/init.c: - name : MPIR_CVAR_ASYNC_PROGRESS > src/mpi/init/init.c: if (MPIR_CVAR_ASYNC_PROGRESS) > src/mpi/init/init.c: if (MPIR_CVAR_ASYNC_PROGRESS) { > src/mpi/init/initthread.c: if (MPIR_CVAR_ASYNC_PROGRESS) > src/mpi/init/initthread.c: if (MPIR_CVAR_ASYNC_PROGRESS) { > (pamid stuff) > > -- > Jeff Hammond > jeff.science(a)gmail.com > http://jeffhammond.github.io/ > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel

1 0

MPICH_ASYNC_PROGRESS
by Jeff Hammond 11 Feb '16

11 Feb '16

README.envvar says you still support it but it appears nowhere in the source code. Are you parsing it via some special magic or are the docs out of sync with the implementation? I ask this now because I saw what appears to be a bug in the implementation, but it's stupid to track down that bug if the problem is that MPICH_ASYNC_PROGRESS is not supposed to work anymore. Thanks, Jeff jrhammon-mac01:git jrhammon$ git grep ASYNC_PROGRESS src/mpi/init/init.c: - name : MPIR_CVAR_ASYNC_PROGRESS src/mpi/init/init.c: if (MPIR_CVAR_ASYNC_PROGRESS) src/mpi/init/init.c: if (MPIR_CVAR_ASYNC_PROGRESS) { src/mpi/init/initthread.c: if (MPIR_CVAR_ASYNC_PROGRESS) src/mpi/init/initthread.c: if (MPIR_CVAR_ASYNC_PROGRESS) { (pamid stuff) -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Re: [mpich-devel] ABI Compatibility Initiative
by Balaji, Pavan 08 Feb '16

08 Feb '16

Thank you, Andrey. The const change in MPI_Address and friends should not affect ABI. In general, const changes are only related to the compiler and should not affect ABI. Are you seeing any failures? -- Pavan > On Feb 8, 2016, at 7:42 AM, Ponomarenko Andrey <andrewponomarenko(a)yandex.ru> wrote: > > The report has been updated: http://abi-laboratory.pro/tracker/timeline/mpich/ > > Thank you. > > 08.02.2016, 03:14, "Balaji, Pavan" <balaji(a)anl.gov>: >> Andrey, >> >>> On Feb 7, 2016, at 3:06 PM, Ponomarenko Andrey <andrewponomarenko(a)yandex.ru> wrote: >>> Could you please clarify whether MPIX_* functions are public/documented? Whether they can be used in applications? >> >> No, they are not documented and must not be used by end users. They are only meant for developers testing things out, or friendly users who are trying out things before we propose them to the MPI standard. >> >>> I will prepare and upload corrected version of the report tomorrow. >> >> Thank you. >> >> -- Pavan > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel

1 0

Re: [mpich-devel] ABI Compatibility Initiative
by Ponomarenko Andrey 08 Feb '16

08 Feb '16

The report has been updated: http://abi-laboratory.pro/tracker/timeline/mpich/ Thank you. 08.02.2016, 03:14, "Balaji, Pavan" <balaji(a)anl.gov>: > Andrey, > >> О©╫On Feb 7, 2016, at 3:06 PM, Ponomarenko Andrey <andrewponomarenko(a)yandex.ru> wrote: >> О©╫Could you please clarify whether MPIX_* functions are public/documented? Whether they can be used in applications? > > No, they are not documented and must not be used by end users. They are only meant for developers testing things out, or friendly users who are trying out things before we propose them to the MPI standard. > >> О©╫I will prepare and upload corrected version of the report tomorrow. > > Thank you. > > О©╫О©╫-- Pavan

1 0

Re: [mpich-devel] ABI Compatibility Initiative
by Balaji, Pavan 07 Feb '16

07 Feb '16

Andrey, > On Feb 7, 2016, at 3:06 PM, Ponomarenko Andrey <andrewponomarenko(a)yandex.ru> wrote: > Could you please clarify whether MPIX_* functions are public/documented? Whether they can be used in applications? No, they are not documented and must not be used by end users. They are only meant for developers testing things out, or friendly users who are trying out things before we propose them to the MPI standard. > I will prepare and upload corrected version of the report tomorrow. Thank you. -- Pavan

1 0

Re: [mpich-devel] ABI Compatibility Initiative
by Ponomarenko Andrey 07 Feb '16

07 Feb '16

Hello Pavan, Thank you for the review of the report and feel free to add a link to your page. Could you please clarify whether MPIX_* functions are public/documented? Whether they can be used in applications? I will prepare and upload corrected version of the report tomorrow. Thank you. 07.02.2016, 20:02, "Balaji, Pavan" <balaji(a)anl.gov>: > Thank you, Andrey. This is great and very helpful. > > One small comment -- the ABI compatibility initiative is only for MPI_ functionality (i.e., functionality that is a part of the MPI standard). So the removal of MPIX_ functionality (MPICH-specific extensions) is not considered ABI breakage in our case. If you could screen out MPIX_ functionality in your list, it might more accurately depict where we stand. > > In mpich-3.2, you are right that we added a bunch of new functions. If you need some notes on that, we moved from MPI-3.0 to MPI-3.1 in that release, which added some additional functions. > > Would you mind if we added this link to our ABI page? > > Thanks, > > О©╫О©╫-- Pavan > >> О©╫On Feb 7, 2016, at 1:32 AM, Ponomarenko Andrey <andrewponomarenko(a)yandex.ru> wrote: >> >>> О©╫О©╫The primary aim of the initiative is for all parties to agree on a schedule for necessary ABI changes, leading to a more stable release cycle and fewer surprises for developers. This project is not limited to just MPICH-derived implementations. Other collaborators are welcome to join and participate in the initiative for the benefit of users, application developers and system administrators alike. >> >> О©╫Hello, >> >> О©╫I've prepared ABI report for the recent versions of the library here: http://abi-laboratory.pro/tracker/timeline/mpich/ >> >> О©╫Also I've shared a special report for the latest MPICH 3.2 in order to navigate over the ABI structure in detail (on x86_64): http://abi-laboratory.pro/tracker/abi_view/mpich/3.2/f8b8b/symbols.html >> >> О©╫Hope this helps the Initiative and may be Linux maintainers of the library too. The source code of the analysis tools is available here: https://github.com/lvc >> >> О©╫Thank you.

1 0

Re: [mpich-devel] ABI Compatibility Initiative
by Balaji, Pavan 07 Feb '16

07 Feb '16

Thank you, Andrey. This is great and very helpful. One small comment -- the ABI compatibility initiative is only for MPI_ functionality (i.e., functionality that is a part of the MPI standard). So the removal of MPIX_ functionality (MPICH-specific extensions) is not considered ABI breakage in our case. If you could screen out MPIX_ functionality in your list, it might more accurately depict where we stand. In mpich-3.2, you are right that we added a bunch of new functions. If you need some notes on that, we moved from MPI-3.0 to MPI-3.1 in that release, which added some additional functions. Would you mind if we added this link to our ABI page? Thanks, -- Pavan > On Feb 7, 2016, at 1:32 AM, Ponomarenko Andrey <andrewponomarenko(a)yandex.ru> wrote: > >> The primary aim of the initiative is for all parties to agree on a schedule for necessary ABI changes, leading to a more stable release cycle and fewer surprises for developers. This project is not limited to just MPICH-derived implementations. Other collaborators are welcome to join and participate in the initiative for the benefit of users, application developers and system administrators alike. > > Hello, > > I've prepared ABI report for the recent versions of the library here: http://abi-laboratory.pro/tracker/timeline/mpich/ > > Also I've shared a special report for the latest MPICH 3.2 in order to navigate over the ABI structure in detail (on x86_64): http://abi-laboratory.pro/tracker/abi_view/mpich/3.2/f8b8b/symbols.html > > Hope this helps the Initiative and may be Linux maintainers of the library too. The source code of the analysis tools is available here: https://github.com/lvc > > Thank you. > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel

1 0