Make FFTW flags "public"

[JeffFessler]

The documentation for AbstractFFTs makes explicit reference to certain FFTW flags, so it would be very desirable for those flags to be public in FFTW so that developers can count on them indefinitely and so that users (like me) who follow those docs do not get warnings about using "non-public" names from ExplicitImports.jl.

Perhaps more (all?) of the FFTW constants should also be public and you or I could amend this PR to be more inclusive. For now I am limiting the request to just those that are cited explicitly in public Julia package documentation because I am unsure which of the many const values in fft.jl should be public.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.64%. Comparing base (03ad4a8) to head (32ca781).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #329   +/-   ##
=======================================
  Coverage   73.64%   73.64%           
=======================================
  Files           5        5           
  Lines         535      535           
=======================================
  Hits          394      394           
  Misses        141      141           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[stevengj]

I think all of the flags here should be public, since they are all documented in the underlying FFTW library:

FFTW.jl/src/fft.jl

Lines 70 to 79 in 08e409c

	const MEASURE = UInt32(0)
	const DESTROY_INPUT = UInt32(1 << 0)
	const UNALIGNED = UInt32(1 << 1)
	const CONSERVE_MEMORY = UInt32(1 << 2)
	const EXHAUSTIVE = UInt32(1 << 3) # NO_EXHAUSTIVE is default
	const PRESERVE_INPUT = UInt32(1 << 4) # cancels DESTROY_INPUT
	const PATIENT = UInt32(1 << 5) # IMPATIENT is default
	const ESTIMATE = UInt32(1 << 6)
	const WISDOM_ONLY = UInt32(1 << 21)
	const NO_SIMD = UInt32(1 << 17) # disable SIMD, useful for benchmarking

[JeffFessler]

That's kind of what I guessed- thanks. Done.

[giordano]

In case it wasn't clear, public doesn't work in v1.10.

[JeffFessler]

Thanks for the reminder. I added a VERSION switch.

[JeffFessler]

Having public within an if statement did not work; i.e., this fails with the error "Expected end" in v1.12.3 on MacOS:

if true
    public foo
end

Fortunately, having an include within the if statement appears to work, so I isolated the public assertions to a separate file, public.jl, as a workaround. Suggestions welcome.

[devmotion]

You can use something like https://github.com/JuliaStats/StatsAPI.jl/blob/ee9906ddd8cab08b454d1f0088c5285ac7bed1fe/src/StatsAPI.jl#L5-L10. Or Compat.jl which provides a macro for this expression, but in StatsAPI it didn't seem worth adding an additional dependency (in particular due to the large number of packages that depend on StatsAPI).

[JeffFessler]

Thanks @devmotion for that very helpful suggestion. I'm giving it a try now.

[JeffFessler]

Now the Julia min (aka v1.10) CI errors seem to be with CoverageTools.jl - it seems that CoverageTools is trying to parse public.jl even though the include for public.jl is wrapped in an if VERSION ... statement. Is this erroneous behavior of CoverageTools that I should report elsewhere? Or am I doing something wrong?

[JeffFessler]

Now passes all the relevant tests. (The macOS-13 tests seem to be deprecated - unrelated to this PR.)
The use of Eval seems slightly unbeatiful to me, but it works across old and new versions, again thanks to @devmotion.
When Julia min becomes 1.11 or higher, then this block of code could be simplified.