Add a script that converts FLINT headers to julia types - #2010
Add a script that converts FLINT headers to julia types#2010lgoettgens wants to merge 17 commits into
Conversation
Codecov Report❌ Patch coverage is
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 Harness. 🚀 New features to boost your workflow:
|
|
Do you plan to handle unions? I noticed that (And the |
Didn't spot that, I'll look into it. |
fa03268 to
3fa24e7
Compare
unions now work as well (at least 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. |
as we could extract structs from large block comments by accident
thofma
left a comment
thofma
left a comment
There was a problem hiding this comment.
Looked at some of the more complicated things, and I did not spot problems. But I did not check everything.
| # Most of this file is generated from FLINT's header files. | ||
| # Do not edit manually, only the corresponding `etc/*_template.jl` file should be edited. | ||
|
|
||
| # This file was generated using FLINT_jll v300.100.301+0. |
There was a problem hiding this comment.
This states that it was generated with FLINT v3.1.3-p1 (I suppose), but further down it says v3.2.0.
There was a problem hiding this comment.
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)
| const n_poly_t = Ptr{n_poly_struct} | ||
|
|
||
| const n_fq_poly_struct = n_poly_struct | ||
|
|
||
| const n_fq_poly_t = n_poly_t | ||
|
|
||
| struct n_bpoly_struct | ||
| coeffs::Ptr{n_poly_struct} | ||
| alloc::slong | ||
| length::slong | ||
| end |
There was a problem hiding this comment.
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.
There was a problem hiding this comment.
Filtering seems kind of hard. I would rather document the breakage possibilities in Nemo and not count this file into SemVer guarantee.
| uniondata::NTuple{maximum(sizeof, ( | ||
| fq_mat_t, | ||
| fq_nmod_mat_t, | ||
| fq_zech_mat_t, | ||
| nmod_mat_t, | ||
| fmpz_mod_mat_t, | ||
| )), UInt8} |
There was a problem hiding this comment.
Just wanted to mention that I'd been thinking about doing something akin to this (separate from this PR) to allow more optimized code:
| 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 Ptr types).
4 participants
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