Core Dump on local 32bit linux build? about openmalaria HOT 2 CLOSED

Ah. I believe I figured this out. As constructed, an ITN intervention may have 
a postprandial effect specified like this (copied from the test 
scenario2ITNs.xml):

<postprandialKillingEffect baseFactor="0.10" holeFactor="0" 
holeScalingFactor="0.1" insecticideFactor="0.55" insecticideScalingFactor="0.1" 
interactionFactor="0"/>

Then, when calculating the survivalFactor in ITN.cpp (line 466),

survivalFactor = (1.-killingEffect)*invBaseSurvival,

with the parameter specifications exactly from the scenario file given above, 
it turns out, 

invBaseSurvival = 1/(1.-killingEffect),

and survivalFactor _should be_ identically unity. 

In this case, the call to assert() should not fail, but it occasionally does. I 
expect this is due to the the division in invBaseSurvival, failing at some 
tiny, tiny precision. Occasionally this is evaluated as ever so slightly 
greater than unity. I believe this causes assert() to fail. 

The author of ITN.cpp saw this, I think, with the subsequent lines including,

if ...
else if (survivalFactor > 1.0)
        return 1.0;

I've commented out the call to assert() and rebuilt. I do not dump core anymore 
and the simulations appear okay. 

Maybe this will help another user out there. 

Thanks and any comments are of course welcome! 

Interesting. Sounds like another case where a small tolerance is necessary.

Richard, would it be possible to send an example scenario where this fails, or 
at least do some testing?

Tobi, can you fix?