GitLab

Previously, the end of jobs' submissions was only detected when a
job were completed. Hence, if a submitter said goodbye after the
last job completion, Batsim was not able to detect it.

The 'id' field was supposed to be followed by a ',',
which is not always the case.

The date is written as a simple floating-point number in the protocol.
However, the size of this date buffer was 16, which caused simulations longer
than 1e8 to fail. This happened to me when I tried to simulate HPC2N.

This buffer is now SIZE=32-character long.
Furthermore, the simulation is now stopped whenever a (SIZE-1)-character long
date is encountered, as it is susceptible to lose information, which may cause
infinite loops.

This commit adds the --post_only option of exec1 script.
This option only computes the post_commands of the instance.

This commit also adds the --recompute_instances_post_commands of
the execN script. This option allows to only recompute the
post_commands of all the instances which were marked as done
before running execN.

-> Now you can avoid recomputing long instances while modifying
   your post_commands!

To improve clarity of results when one forgot to run a Redis server

Merge branch 'data_storage' of git+ssh://scm.gforge.inria.fr//gitroot/batsim/batsim into data_storage

for all the cases I could reproduce.

Merge branch 'data_storage' of git+ssh://scm.gforge.inria.fr/gitroot/batsim/batsim into data_storage

This commit fixes issue #2, thanks to pfdutot's patch.

Previously, the user-given was directly parsed via argparse.
This limited the user-given command to be a direct call to
Batsim. Furthermore, the script was directly aborted on
parsing error, which did not explicit the abort reason well.

Now, prefixed Batsim calls should be handled.
Furthermore, whenever argparse fails in parsing the command,
the error outputting should be way clearer.

The old SimGrid (f620ec2586) does not compile with Boost 1.61.0-2.
The new SimGrid (42a5c2c5fa) segfaults when Batsim wants to SIMIX_process_throw.

So, jobs are no longer killed on reaching walltime as long as this problem
is not fixed. Creating a Github issue for it.

The regex part of the C++11 standard was not implemented before june 2015.
To avoid forcing users to have a very recent compiler, switching to boost instead.

The problem is probably related to some IO buffering issue.
Batsim is now run in quiet mode in the tests, which hides the problem :(