flread loses information in the 69th year of run

[mouchet]

The following behavior is obtained with configuration OFF_MY_TRC & NEMO v3.4

I launched a 100 years experiment (no intermediate restart).
Suddenly (within one time step), in the 69th year of run all concentrations became NaN.

My understanding is that flread suddenly loses the information about the actual record time.

There is nothing more special that year than any other. I use climatological dynamical forcing (perpetual year with values every 5 days). Exactly the same thing happen whether tracers undergo sms or not.

Ch. Ethé obtained exactly the same behavior on a different computer with his own version of the model.

Many thanks for your help,
Anne

[smasson]

The problem is that fldread uses the number of seconds since the last restart and store it in an integer. By default you compile NEMO with integer 4, which means that the bigger number you can store is 2^(4*8-1)-1 and

2.^31/24./60./60./365. = 68.0963

=> you exceed integer 4 precision when moving to year 69 and you get NaN or other values according to the compiler or the compiler options you use.

Solutions are:

• you do a maximum of 68 years without restart
• you recompile everything with integer 8 instead of 4

A stop has been had in the trunk, see r4085 and in v3_4_STABLE, see r4086

[smasson]

typo correction in the trunk, see r4103 and in v3_4_STABLE, see r4104