Add module for handling version information of stdlib

[awvwgk]

Closes #576

[ivan-pi]

Suggested change

	PROJECT_VERSION_MAJOR=0
	PROJECT_VERSION_MINOR=0
	PROJECT_VERSION_PATCH=0
	PROJECT_VERSION_MAJOR=0
	PROJECT_VERSION_MINOR=1
	PROJECT_VERSION_PATCH=0

[ivan-pi]

Or is this overridden during the fypp processing?

[awvwgk]

This is supposed to be a stub, similar like the MAXRANK=3 argument above, which would would not allow to actually compile stdlib.

[ivan-pi]

Don't these macros get applied when FORD runs fypp to preprocess the sources for the documentation? At least that's what I understand from here: https://github.com/Fortran-FOSS-Programmers/ford/wiki/Project-File-Options

[awvwgk]

It does show up in the final docs for the compile time constant:

Otherwise the value set won't affect the output of the documentation. I would prefer to leave this as a stub, so we don't have to worry about it becoming outdated when we release a new version.

[ivan-pi]

Comments are not allowed in the FORD meta-data section. Is there somewhere we could document these values are stubs, and don't reflect the actual library?

[ivan-pi]

The CMake syntax to read a list is just terrible.

Pending the small issues, LGTM. Thanks.

[ghost]

@awvwgk many thanks for this PR. I tested it on Windows using CMake+Ninja.
Regarding the issue with $(shell cut ..), I'm trying to find a solution but no luck yet.

[ghost]

OK I found a combination that works for me on Windows

set FC=gfortran
set FYPPFLAGS=-DMAXRANK=4
gnumake -f Makefile.manual

Thanks @awvwgk

[ghost]

Thanks @awvwgk

[ghost]

I just noticed ubuntu-latest CI is failing at make -f Makefile.manual FYPPFLAGS="-DMAXRANK=4" -j. It is the same issue. If we remove FYPPFLAGS="-DMAXRANK=4" from the command line and set it as environment variable, it will work, I think. That's how I did on my Windows.

[ivan-pi]

Just wondering if a fixed-length string can also be used to collect the output?

[awvwgk]

The dummy argument is currently character(len=:), allocatable, intent(out), optional, since we are using a subroutine rather than a function. The rationale is that we want a stable API and ABI for this procedure, we could also go for multiple functions for getting the individual values, to make use of it in a more functional way.

[ghost]

@awvwgk looks good.

[ivan-pi]

@jvdp1, @milancurcic, can one of you give this a third approval before merging?

[jvdp1]

LGTM. Just a couple of minor comments.
Thank you!

[jvdp1]

Why not only allow the correct approach to do it?

[awvwgk]

For the compact version I can see an application to verify the compatibility of API and/or ABI at runtime by a construct like:

use stdlib_version, only : stdlib_version_compact, get_stdlib_version
implicit none
integer :: major
call get_stdlib_version(major=major)
if (major > stdlib_version_compact / 10000) then
  error stop "Incompatible version of stdlib used"
end if
end

Similarly, stdlib_version_string and get_stdlib_version(string=string) can be used to provide a printout to inform about the stdlib versions used at build time and run time.

Not sure whether this is a realistic application of this information, usually this kind of verification happens at build time and constraints in package managers can ensure compatible version, but not everybody is using those.

[jvdp1]

Ok. Thank you for the explanation. You can merge this PR from my point-of-view.

[milancurcic]

Very nice addition, thank you.

Would it be useful for this module to also provide compile-time constants of compiler version and options used to build it?

module m
  use iso_fortran_env, only: compiler_version, compiler_options
  character(*), parameter :: stdlib_compiler_version = compiler_version()
  character(*), parameter :: stdlib_compiler_options = compiler_options()
end module m

program p
  use m
  print *, stdlib_compiler_version
  print *, stdlib_compiler_options
end

It works nicely with GFortran, but ifort 2021.3.0 doesn't seem to allow these intrinsic in constant expressions:

$ ifort test.f90 
test.f90(3): error #6263: This intrinsic function is invalid in constant expressions.   [COMPILER_VERSION]
  character(*), parameter :: stdlib_compiler_version = compiler_version()
-------------------------------------------------------^
test.f90(4): error #6263: This intrinsic function is invalid in constant expressions.   [COMPILER_OPTIONS]
  character(*), parameter :: stdlib_compiler_options = compiler_options()
-------------------------------------------------------^
test.f90(8): error #7013: This module file was not generated by any release of this compiler.   [M]
  use m
------^
compilation aborted for test.f90 (code 1)

which I think is incorrect. Anyhow, just an idea, something we could tackle in another PR.

[awvwgk]

Good idea, I opened a new issue for this one.