Revise the Timer system to provide better time logging facilities

[Chris Richardson]

The present Timer system has some drawbacks, especially in parallel. Some features which would be nice:

1. Knowledge of nested timers, so the same piece of code is not counted twice
2. Machine readable output format, for plotting graphs
3. Aggregation of timers in parallel, to produce a single output
4. Max/min/average calculation for each timer

Both boost and Eigen (required dependencies) have nice timers available.

[Anders Logg (Chalmers)]

Yes, good points.

[Prof Garth Wells]

C++11 has some new timing facilities too.

I use Boost timer, but the downside is that it's not header-only which adds a little extra complication to the build and configure process.

[Chris Richardson]

This issue could be closed. Only one small thing - @blechta - the deprecation message for list_timings() in Python is not very helpful.

The replacement for list_timings() is list_timings(TimingClear_clear, [TimingType_wall]) ?

[Jan Blechta]

It is list_timings(TimingClear_keep, [TimingType_wall]) I think. Do you suggest to keep list_timings() executing list_timings(TimingClear_clear, [TimingType_wall]) or just improving deprecation message?

[Chris Richardson]

Your choice - it is just a bit confusing for Python users at the moment...

[Jan Blechta]

Remainder for me: doc is not generated for enum class TimingClear and enum class TimingType, both in C++ and Py.

Fix possibly here.

[Jan Blechta]

Fix in next 2575751

[Jan Blechta]

Should be resolved by 831ab0d

[Martin Sandve Alnæs]

Removing milestone: 1.6 (automated comment)