Add a script that converts FLINT headers to julia types

[lgoettgens]

This is a first step towards #1875.

I used some techniques described in #1889, but extended them and automated everything.

The only file that is supposed to be edited is the FlintCTypes_template.jl. The script then searches for the mentioned header files in the FLINT_jll artifact, strips away C-specific things like include guards, and converts all structs, enums, and typedefs to equivalent julia constructs.

This change on its own is non-breaking, as I don't change any already existing types (that's planned in a next PR). The new file isn't even included yet.

cc @albinahlback

[codecov]

Codecov Report

Attention: Patch coverage is 64.28571% with 5 lines in your changes missing coverage. Please review.

Project coverage is 88.27%. Comparing base (82193e6) to head (29471dc).
Report is 2 commits behind head on master.

Files with missing lines	Patch %	Lines
src/flint/FlintCHelpers.jl	66.66%	4 Missing ⚠️
src/flint/FlintCTypes.jl	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2010      +/-   ##
==========================================
- Coverage   88.31%   88.27%   -0.04%     
==========================================
  Files          98      100       +2     
  Lines       36126    36195      +69     
==========================================
+ Hits        31903    31951      +48     
- Misses       4223     4244      +21     

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

[thofma]

Do you plan to handle unions? I noticed that fq_default is missing.

(And the _template should probably be moved to etc/ too.)

[lgoettgens]

Do you plan to handle unions? I noticed that fq_default is missing.

Didn't spot that, I'll look into it.

[lgoettgens]

Do you plan to handle unions? I noticed that fq_default is missing.

unions now work as well (at least typedef union {...} fq_default_struct now results in a julia struct with a single NTuple{sizeof-computation, UInt8} field with intransparent memory usage.

After talking to @albinahlback about an hour ago, I now include some more header files and start with all content, but strip away everything that does not match any of the regexes. This then gets rid of all preprocessor instructions, function signatures etc.
Furthermore, I already upgraded this branch to work only with FLINT 3.2.0 (or rather the rc1 of that) as there are some refactorings included that avoid the need to include gmp.h as well.

[thofma]

Looked at some of the more complicated things, and I did not spot problems. But I did not check everything.

[albinahlback]

This states that it was generated with FLINT v3.1.3-p1 (I suppose), but further down it says v3.2.0.

[lgoettgens]

FLINT_jll v300.100.301 is actually FLINT v3.1.3 (everything times 100), but with an additional bump when rebuilding with aarch-*-freebsd support (cf. JuliaPackaging/Yggdrasil#10029).

This will automatically fix itself once FLINT 3.2.0 is released and the corresponding jll is built. The way that Overrides.toml work in julia is that I basically inject the FLINT 3.2.0-rc1 build results into the FLINT_jll v300.100.301 package (locally). So this is just an intrinsic problem with how overrides work, and nothing for me to fix. (One should just not regenerate this with Overrides active)

[albinahlback]

Not sure whether you actually want these things. Perhaps you do. But if you would like to avoid these stuff, I would encourage you to only translate stuff that you really need. Unless it is documented, it can break.

[lgoettgens]

Filtering seems kind of hard. I would rather document the breakage possibilities in Nemo and not count this file into SemVer guarantee.

[fingolfin]

Just wanted to mention that I'd been thinking about doing something akin to this (separate from this PR) to allow more optimized code:

Suggested change

	uniondata::NTuple{maximum(sizeof, (
	fq_mat_t,
	fq_nmod_mat_t,
	fq_zech_mat_t,
	nmod_mat_t,
	fmpz_mod_mat_t,
	)), UInt8}
	entries::Ptr{Nothing}
	r::slong
	c::slong
	rows::Ptr{Nothing}
	uniondata::NTuple{maximum(sizeof, (
	fq_mat_t,
	fq_nmod_mat_t,
	fq_zech_mat_t,
	nmod_mat_t,
	fmpz_mod_mat_t,
	) - 32), UInt8}

(where 32 of course is 2*sizeof(Ptr{Nothing}) + 2*sizeof(slong)).

So that one can then write fast accessors to these fields. Similar for fq_default_poly.

Not sure how to fit this with this PR... but one option would be to automate this as part of Union handling: if forming the C union of C struct types $T_1, \dots, T_k$ then we should in principle be able to figure out if they have common entries at the start, and automatically "extract" them (with slight cheating to "merge" Ptr types).