Update scripts for Clang v0.17

[odow]

Closes #530

@simonbowly you could try building with this. It has a few tweaks, like using Cchar instead of UInt8

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (db2d8e6) 90.87% compared to head (36bc252) 90.87%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #532   +/-   ##
=======================================
  Coverage   90.87%   90.87%           
=======================================
  Files           5        5           
  Lines        2422     2422           
=======================================
  Hits         2201     2201           
  Misses        221      221           

Files	Coverage Δ
src/Gurobi.jl	80.00% <ø> (ø)

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

[odow]

Seems reasonable?

[simonbowly]

Yeah, a few adjustments got it building for v10. The changes look minor, but there are some failing MOI tests on my mac that I haven't dug into yet.

[simonbowly]

It looks like some constraints silently fail to be added? Perhaps the CChar change could cause this, this update changes const GRB_LESS_EQUAL = '<' and friends to const GRB_LESS_EQUAL = Cchar('<').

[simonbowly]

Manually adjusting the character constants (6bdd625) resolved this for me. I don't see an option to fix this in Clang.jl. I guess worst case these symbols can go in the preamble?

[odow]

Hmm. It'd be nice to know why. You could just do a find-and-replace on the file once it has been built?

[odow]

Otherwise just a note in the gen.jl script is okay. It's a very small job that happens infrequently.

[simonbowly]

I might just add them to the ignore list and preamble. This is a set of symbols that isn't likely to change.

As for why it fails, you may be on the right track with the Char<->[U]Int8 mapping. Evidently this differs by processor type, not by OS? If you happen to have an ARM chip handy that might shed some light on it.

It's a bit worrying that that MOI test seems to only find an issue when it checks the number of constraints in a model though? Makes it seem like an error code check is not being done?

[odow]

I didn't look into why it failed yet, but it's probably something to do with lines like this:

Gurobi.jl/src/MOI_wrapper/MOI_wrapper.jl

Line 1363 in db2d8e6

_info(model, c).type == GRB_BINARY

The C API calls will convert properly, but

julia> 'B' == Cchar('B')
false