Loading float numbers from external data CSV file fails.

[malinb73]

--- Context ---
Built fmusim for Windows:
Platform: x86_64-windows
CMake Generator: Visual Studio 16 2019
Build type: Debug

Git commit checked:
Revision: 6d3a981
Author: Torsten Sommer torsten.sommer@3ds.com
Date: 7/5/2024 11:19:21 AM
Message: Fix variable index in fmi3{Get|Set}Binary() (#531)

--- Description ---
The error happens when importing external data from a CSV file and amongst other different inputs there are also inputs having float types.

Example:
time,s8_in,s16_in,array1D_in,float32_in
0,1,2,3 4 5,12.34

The error happens randomly.
The error message is: "Failed to reallocate memory."
The error occurs in fmusim/FMIStaticInput.c in function:
FMIStatus FMIApplyInput(FMIInstance* instance, const FMIStaticInput* input, double time, bool discrete, bool continuous, bool afterEvent)
at line 276:
CALL(FMIRealloc((void**)&input->buffer, requiredBufferSize));
The reason is requiredBufferSize variable has sometimes a very huge value due to some wrong memory read which is based on row variable.
This row variable is calculated in the while block from lines 255, see: "while (row < input->nRows - 2) {...."
Changing the while condition to "while (row < input->nRows - 1)" seems to solve the problem.
Can you check please?

[t-sommer]

Can you reproduce the problem if you use an input file with at least 3 rows?

E.g.

time,s8_in,s16_in,array1D_in,float32_in
0,1,2,3 4 5,12.34
1,1,2,3 4 5,12.34
2,1,2,3 4 5,12.34

[malinb73]

Hi Sommer,

I made again 3 tests and this is the status:

1. tested with 0.0.31

   • for 1 line fails
   • for more that 1 line ok

2. tested with commit Revision: 6d3a981

   • for 1 line fails
   • for more that 1 line ok

3. tested with my fix

   • for 1 line ok
   • for more that 1 line ok

I am not saying that I understood completely the code of FMIApplyInput() function from FMIStaticInput.c but the fix suggested is based on next fact:
I compared the code which fails from:
if (continuous && variable->variability == FMIContinuous)
with the code which works fine from:
else if (discrete && (variable->variability == FMIDiscrete || variable->variability == FMITunable))
and beside the specific code which handles FMIFloat32Type and FMIFloat64Type the notable difference is
while (row < input->nRows - 2)
VS.
while (row < input->nRows - 1)
Playing around with this condition and changing it to "row < input->nRows - 1" proved that worked fine also for floats (variability = FMIContinuous)

[malinb73]

Hi Sommer,
I can confirm that now it works as expected.
Can you tell me when is scheduled the next release having this fix?
Thank you,
Bogdan

[t-sommer]

You can download the binaries directly from the CI: https://github.com/modelica/Reference-FMUs/actions/runs/10073277513/artifacts/1734054905