hpcprof and hpcprof-mpi dump core with bad measurements directory

[jmellorcrummey]

If provided a bad name for a measurements directory (e.g., a misspelling), both hpcprof and hpcprof-mpi dump core rather than terminating gracefully.

Example provided below:

[johnmc@mic lulesh]$ $p/hpcprof hpctoolkit-foo-measurements
reading profile 0
ERROR: While reading profile 'hpctoolkit-foo-measurements'...
        [../../../../src/lib/analysis/CallPath.cpp:241]
terminate called after throwing an instance of 'Diagnostics::FatalException'
Abort (core dumped)

[jmellorcrummey]

A proper fix for this issue would be having hpcprof and hpcprof-mpi try to open the specified "measurement-group" as either a measurement directory or a profile file. If it is neither, print the error

ERROR: unable to open specified measurement directory or file

and call exit with an error code directly rather than using one of the Diagnostic macros that print source code attributions like [../../../../src/lib/analysis/CallPath.cpp:241].

[unkadoug]

I just rebuilt and installed master, and I cannot reproduce this bug:
-bash-4.2$ hpcprof hpctoolkit-foo-measurements
ERROR: While reading profile 'hpctoolkit-foo-measurements'...
[../../../../src/lib/analysis/CallPath.cpp:180]
ERROR: [Diagnostics::FatalException] error opening file [../../../../src/lib/prof/CallPath-Profile.cpp:908]
[../../../../src/tool/hpcprof/main.cpp:105]

Yes, it's an error, but no core file is produced. Can anyone verify that this bug remains, and if so tell me how to reproduce it?

[jmellorcrummey]

Doug, I too saw that the SEGV seems to be gone. Before closing the error, I think that the error message should be improved to distinguish the following cases: - no file or directory exists with the given name - the file or directory exists but can’t be read - the file exists, can be read, but there is an error reading it, i.e. it can’t be parsed as a correct hpcrun file - the directory exists but doesn’t seem to contain hpctoolkit measurements The internal information mapping it back to source locations makes this look like it is a problem with hpctoolkit rather than user error. That is my principal concern now that the SEGV is gone. -- John Mellor-Crummey Professor Dept of Computer Science Rice University email: johnmc@rice.edu phone: 713-348-5179

…

On Apr 13, 2018, at 1:00 PM, Doug Moore ***@***.***> wrote: I just rebuilt and installed master, and I cannot reproduce this bug: -bash-4.2$ hpcprof hpctoolkit-foo-measurements ERROR: While reading profile 'hpctoolkit-foo-measurements'... [../../../../src/lib/analysis/CallPath.cpp:180] ERROR: [Diagnostics::FatalException] error opening file [../../../../src/lib/prof/CallPath-Profile.cpp:908] [../../../../src/tool/hpcprof/main.cpp:105] Yes, it's an error, but no core file is produced. Can anyone verify that this bug remains, and if so tell me how to reproduce it? — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#14 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADvoYN06f7lL4Nk6aN1jJ5GFqt0fL1Y8ks5toQPZgaJpZM4PSaTY>.

[unkadoug]

I've addressed the first two cases. The third case is addressed already. I'm not sure how the behavior you want for the 4th case differs from the current behavior. An example would be helpful.

[jmellorcrummey]

Post the results of a test for each of the cases and I can offer an opinion as to whether everything is resolved.

[unkadoug]

There is no file foo:
-bash-4.2$ hpcprof foo
ERROR: While reading profile 'foo'...
[../../../../src/lib/analysis/CallPath.cpp:180]
ERROR: [Diagnostics::FatalException] file missing [../../../../src/lib/prof/CallPath-Profile.cpp:909]
[../../../../src/tool/hpcprof/main.cpp:105]

There is an unreadable file foo:
-bash-4.2$ hpcprof foo
ERROR: While reading profile 'foo'...
[../../../../src/lib/analysis/CallPath.cpp:180]
ERROR: [Diagnostics::FatalException] file access denied [../../../../src/lib/prof/CallPath-Profile.cpp:911]
[../../../../src/tool/hpcprof/main.cpp:105]

There is a readable file foo with junk data:
bash-4.2$ hpcprof foo
msg: Directory 'hpctoolkit-database' already exists. Trying 'hpctoolkit-database-108018'
msg: Created directory: hpctoolkit-database-108018
ERROR: While reading profile 'foo'...
[../../../../src/lib/analysis/CallPath.cpp:180]
ERROR: [Diagnostics::FatalException] error reading 'fmt-hdr' [../../../../src/lib/prof/CallPath-Profile.cpp:947]
[../../../../src/tool/hpcprof/main.cpp:105]

There is a readable file foo that cannot be fopened for reasons other than 'permission denied':
I don't have easy repro for that. Write a loop to use up all file descriptors? Whatever. It's the same error message that was there yesterday when fopen failed.

There is an empty foo directory:
-bash-4.2$ hpcprof foo
msg: Populating Experiment database: /home/dougm/pkgs-src/hpctoolkit/BUILD/hpctoolkit-database