allegedly analysis/rt43 is failing for Davide about plumed2 HOT 10 CLOSED

This is one of the few regtests that (for me) fail on linux. I think that is due to numerical errors. Maybe we could check if writing the reference histogram with less digit helps.

from plumed2.

OK I added an option to specify the file format for histogram. It was always working on my computer so this made no difference to me. Can someone try it on a computer where it wasn't working please? Thanks

from plumed2.

I still get the error (with intel c++ on mac).
I tried with FMT=%8.1f and still the two compilers give different results.
With FMT=%8.0f I get a surprising error:
+++ Internal PLUMED error
+++ file Grid.cpp, line 170
+++ message: assertion failed indices[i]>=0 && indices[i]<nbin_[i]

Maybe this is related to #19 ?

from plumed2.

Firstly, the grid spacing is 0.005, if you output with anything less than %8.3f then there are always going to be problems. I therefore think the error with %8.0f is a non starter.

What precisely fails with this test. Do you get a mismatch for just histo, just analysis.0.histo or both?

from plumed2.

I agree that the output is a nonsense with %8.0f. Still the code should NOT crash.

Anyway, with %8.4f I get a mismatch in both files with intel c++

from plumed2.

The grid goes from 0.2 to 1.2 in the x direction. The grid is read as the second histogram output is the sum of the histograms from the first half of the simulation and the second half. If you put %8.0f the first x value is 0, which is outside the range. Hence the crash with that error. If we want to do something to fix the %8.0f error I think we have to make sure the grid spacing can be output with the format. I don't know how to do that though.

from plumed2.

Got it! I did not realize it was re reading it... Thanks! Still it would be nice to make the regtest pass with different compilers I will try to play with parameters tomorrow and ask you if they make sense

from plumed2.

I think I found it. Intel c++ with -O changes the stream of random numbers thus produce a different traj with simplemd. If I compile only tools/Random.cpp with -O0 the regtest is passed. I will try to fix it

from plumed2.

Hurah! As far as I am concerned that means I am not to blame :-)

from plumed2.

@davidebr can you check if it is ok also on your machine?

from plumed2.