Time estimate for event generation can be off by an order of magnitude

[Christian Speckner]

For the attached input file, the time estimate for event generation is too low by more than an order of 10.

[kilian]

The timer module can also produce arithmetic exceptions (could result from the nagfor -nan option, which would indicate uninitialized reals). This is not easily reproducable, and no obvious cause.

[kilian]

r2278 adds the option to switch on/off time estimates. Given that, I postpone the ticket to the next patchday.

[Juergen Reuter]

There is one further point which is trivial (also to implement I guess) which is that if you read in grids (i.e. not integrating again) the time estimate is nevertheless calculated, namely exactly to one. Either leave out the time estimate in that case (only the conditional read_grids has to be accessed for this) or store the time estimate also in the grid set-up (which is fishy to me).

[Juergen Reuter]

These two issues are really nasty for the user, should be done with high priority.

[kilian]

Resolved in r2326.

• Module 'clock' took wallclock time, replaced by 'cputime' which takes CPU time and is also much more lightweight.
• Time estimate was computed for integration#n_pass, should have been #n_it (hence the NANs for uninitialized)