neural-fortran VS fortran-wringer-tests

A few more:

https://github.com/modern-fortran/neural-fortran (disclaimer, I originally created this one)

https://github.com/BerkeleyLab/inference-engine

https://git.exeter.ac.uk/hepplestone/athena

check out this, it's not really a framework but good enough to start with

An ISO standard should promote portability across implementations, prevent breaking changes to the language, and coordinate design, prototyping, testing, and description of new features.

Fortran features since F'95 are not terribly portable (https://github.com/klausler/fortran-wringer-tests); the F'23 standard has a needless breaking change; and every revision contains incompletely thought through features jammed in without prototyping ("rank agnostic array indexing"). They won't fix bugs, either -- it is possible to write DO CONCURRENT loops that are completely conformant but cannot be parallelized (and cannot be determined at compilation time).

> OTOH, the existence of an ISO standard with multiple implementations can benefit the portability and longevity of code.

This is true for ISO standards that actually standardize features. Fortran's standard, since F'90, has instead been inventing features, and doing so without prototyping in actual implementations. And without supplying standardized test suites to guide those implementations. The results, in actual practice, have been at best mixed. There are features that are "standard" but not at all portable, due to spotty and divergent implementations, and there are portable features that are not standard. Some features have been in the language for >=20 years without yet appearing in popular compilers.

So yes, standardization (ISO or otherwise) can be a good thing. But it hasn't really been so for Fortran. And I think things are getting worse; F'2023 has changes in it that actually silently change the behavior of existing standard-conforming code, which would have been viewed as an abomination in earlier days.

References: see the LLVM Flang documentation on extensions, non-standard features, &c. in https://github.com/llvm/llvm-project/blob/main/flang/docs/Ex... and a suite of various incompatible feature tests in https://github.com/klausler/fortran-wringer-tests .

If "it" is F'23, then none. GNU Fortran has had the "new" degree-unit trig functions for a while, but no compiler, FOSS or otherwise, has the newly invented features of this revision.

Fortran doesn't prototype features with real implementations (or test suites) before standardizing them, which had led to more than one problem over the years as ambiguities, contradictions, and omissions in the standard aren't discovered until years later when compiler developers eventually try to make sense of them, leading to lots of incomplete and incompatible implementations. I've written demonstrations for many examples and published them at https://github.com/klausler/fortran-wringer-tests/tree/main .