Fail loading fmu on Ubuntu with GLIBC>2.35

[LorenzoBottaccioli]

On systems with glibc ≥2.36 (e.g., Ubuntu 24), paths starting with double slashes (//) are interpreted as network paths. PyFMI generate commands like:

cp -f "//resources//variables.cfg" "."

which fails with:

cp: cannot stat '//resources//variables.cfg': No such file or directory

As a result, FMU instantiation fails on newer systems, while it works correctly on older versions of glibc (<2.36).

Steps to Reproduce

Extract an FMU containing a resources folder.

Observe the cp command in the logs uses //resources//variables.cfg.

On Ubuntu 24, the copy fails and instantiation errors out.

Impact

Blocks FMU usage on modern Linux distributions without any changes to the FMU package.

[modelonrobinandersson]

Hi @LorenzoBottaccioli. Thanks for your issue.
I just tried reproducing this on Ubuntu 24, GLIBC 2.39, Python 3.11 using one of the reference FMUs with external resources:
https://github.com/modelica/Reference-FMUs/tree/main/Resource

This however load and simulates for me without any problem. Note that the unzipping of the FMU is done automatically as part of the loading.

In [1]: from pyfmi import load_fmu
In [2]: fmu = load_fmu('reference_fmus/2.0/Resource.fmu')

In [3]: fmu.simulate()
Final Run Statistics: ---

 Number of steps                                 : 501
 Number of function evaluations                  : 505
 Number of Jacobian evaluations                  : 9
 Number of function eval. due to Jacobian eval.  : 9
 Number of error test failures                   : 0
 Number of nonlinear iterations                  : 501
 Number of nonlinear convergence failures        : 0

Solver options:

 Solver                   : CVode
 Linear multistep method  : BDF
 Nonlinear solver         : Newton
 Linear solver type       : DENSE
 Maximal order            : 5
 Tolerances (absolute)    : 1.0000000000000002e-06
 Tolerances (relative)    : 0.0001

Simulation interval    : 0.0 - 1.0 seconds.
Elapsed simulation time: 0.03800328799843555 seconds.
Out[4]: <pyfmi.fmi_algorithm_drivers.FMIResult at 0x7fc542641ad0>

Note:

user@f8b38fd99941:/opt# ldd --version
ldd (Ubuntu GLIBC 2.39-0ubuntu8.5) 2.39
Copyright (C) 2024 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Written by Roland McGrath and Ulrich Drepper.

Can you tell me what version of PyFMI you used and/or perhaps can you share the error message that might be part of the log file?

[LorenzoBottaccioli]

Hi,

This is the version of PyFMI that has been installed with conda from conda-forge.

test.zip

f_5Zone_Default_SETPOINT_log.txt

PyFMI @ file:///home/conda/feedstock_root/build_artifacts/pyfmi_1750956030067/work/dist/pyfmi-2.17.4-cp39-cp39-linux_x86_64.whl#sha256=b35608f91056a4d1da2fb4

Attached you find the FMU that I'm using, which is a FMU of EnergyPlus generated with EnergyPlustoFMU, a test script that only tries to laod and the logfile. I have faced the same issue with fmpy, on Ubuntu 20 with GLIBC==2.32 I face no issue.

cp: cannot stat '//resources//variables.cfg': No such file or directory
Traceback (most recent call last):
  File "/home/lorenzo/test.py", line 3, in <module>
    fmu=load_fmu('f_5Zone_Default_SETPOINT.fmu')
  File "src/pyfmi/fmi.pyx", line 180, in pyfmi.fmi.load_fmu
  File "src/pyfmi/fmi.pyx", line 280, in pyfmi.fmi.load_fmu
  File "src/pyfmi/fmi2.pyx", line 5228, in pyfmi.fmi2._load_fmi2_fmu
  File "src/pyfmi/fmi2.pyx", line 3608, in pyfmi.fmi2.FMUModelCS2.__init__
  File "src/pyfmi/fmi2.pyx", line 1141, in pyfmi.fmi2.FMUModelBase2.instantiate
pyfmi.exceptions.FMUException: Failed to instantiate the model. See the log for possibly more information.

[modelonrobinandersson]

Hi @LorenzoBottaccioli
Thanks for the additional input.

This seems a bit strange, neither PyFMI nor FMIL (which is a dependency to PyFMI) tries to copy any files, and in particular it wouldnt copy files arbitrarily to your home directory. It does however create a temporary directory and unzips the FMU there.

A couple of questions:

1. Are you sure that this copy-command is not from external code in your FMU?
2. The instance running your FMU does it have permissions to copy the file to the target directory (assuming it exists)?

[LorenzoBottaccioli]

I did the following.

Setup a new env with python 3.9 install pyfmi with conda forge and try to run the fmu. If this is done in ubuntu 20 it owrks in ubuntu 24 it gives this error. I belive the problem is with GLIB becouse in Ubuntu 20 I'm not able to run the FMU generated in Ubuntu 24 for an issue with GLIB version.

This is all I can say, I just report here the problem I faced. The fmu is from Energyplus generated using the EnergyPlustoFMU script that is available in github.

[modelonrobinandersson]

Hi @LorenzoBottaccioli thanks for the update. Yes, I believe the issue should be reported to the tool that has created FMU or the author of the model/library. I suspect that copying of the file is from external code inside the FMU this is where it goes wrong.

[jschueller]

they definitely use double slashes:
https://github.com/lbl-srg/EnergyPlusToFMU/blob/master/SourceCode/v20/EnergyPlus/defines.h#L49
so not a pyfmi bug