f2py doesn't update .mod file

Hello,

When f2py -c foo.f90 -m bar and a foo.mod exists. F2py doesn’t update the .mod file, and what worse, it used the .mod file implicitly. If I changed something in the .f90 file, the .mod file will not be consistent with the .f90 file.

The .mod file is an intermediate file, as the same as .o, so if we don’t use it explicitly on a command line, it should be regenerated, at least, it should not be used.

The module variable with parameter attribute will not be written in the .o file, but in the .mod file. I have just encounter an odd bug because f2py implicitly used inconsistent .mod file.

[bilderbuchi]

Thank you for the illuminating and extensive reply, I’m gonna bookmark that 😄

I think in my mental model until now, the compiler created the .mod and .o files during compilation (so I mentally filed them both under “object files”), and uses both during linking. I had thought that that aligned with what you point out in the Fortran wiki – compiling generates a .mod file, linking uses – in this sense I don’t see the difference between a .mod and a .o file. What I did not realise is that even the compiler does not always regenerate a .mod if already present (gfortran seems to detect an interface change and regenerate then).

Luckily, I’ll be using f2py as part of a build system, so if any regeneration needs to be done, I’ll let cmake figure it out. 😄 I just need to be careful during manual experiments.

[HaoZeke]

Build systems, as you mentioned have their own logic for figuring out when to regenerate .mod files, as you noticed it is typically on the basis of when the interface changes.

Unfortunately making np.distutils include such logic feels a bit out of scope. np.distutils proxies over to a compiler command, not to a build system (though support for different build systems is planned) at which point this will be fixed.

EDIT: The mod file contains all the code in a module, even if that is separated across source files --> the mod is the unit of program use defined in the standard.

Consider section 5.2.4 of the J3/SD-007 Fortran 202X interpretation document. Crucially the standard nowhere defines any source files and defines a program as:

Program units are the fundamental components of a Fortran program. A program unit is a main program, an external subprogram, a module, a submodule, or a block data program unit