Refactor src/hpc/traverse.c

[fingolfin]

One motivation was to remove the need for putting knowledge about how weak pointer objects work in there, which poses a (minor) complication for PR #2092

It also correct some quirks in the traverse code (e.g. the traversal for immutable lists of cyclotomics and FFEs was not being set before, leading to suboptimal performance), and opens it up for use by kernel extensions.

This code might at some point become interesting for GAP, too. In particular, I'd be curious to test performance for copying objects with it vs. with our current copying code. Of course I expect the current copying code to be faster, but by how much? If it is not too much, I'd investigate switching GAP to this alternate copying code, simply because it is much simpler (and would allow us to get rid of the COPYING tnum trick).

[codecov]

Codecov Report

Merging #2102 into master will increase coverage by 15.34%.
The diff coverage is 77.77%.

@@             Coverage Diff             @@
##           master    #2102       +/-   ##
===========================================
+ Coverage    53.9%   69.24%   +15.34%     
===========================================
  Files         426      489       +63     
  Lines      224155   256303    +32148     
===========================================
+ Hits       120821   177485    +56664     
+ Misses     103334    78818    -24516

Impacted Files	Coverage Δ
src/hpc/thread.c	51.48% <ø> (ø)
src/precord.h	96.87% <ø> (+28.69%)	⬆️
src/plist.c	91.17% <100%> (+0.78%)	⬆️
src/precord.c	96.04% <100%> (+7.38%)	⬆️
src/hpc/traverse.c	94.97% <100%> (ø)
src/objects.c	78.73% <100%> (+5.04%)	⬆️
src/weakptr.c	95.42% <100%> (+31.31%)	⬆️
src/objset.c	81.71% <28.57%> (-2.9%)	⬇️
src/funcs.c	78.3% <0%> (-15.61%)	⬇️
lib/random.g	93.33% <0%> (-6.67%)	⬇️
... and 377 more

[markuspf]

Except for a minor comment on an "empty" TODO this looks good to me.

I assume we'll add proper tests in your later PR? (Or later full stop)?

[markuspf]

What is it that needs to be done?

[fingolfin]

Ah, I wanted to add a comment there explaining what the function does. I'll remove this here, and add that comment in a future PR (I already have a bunch of them which I wrote while trying to understand the traverse code, and fixing some of it)

[fingolfin]

I took care of the TODO comment. As to tests: well, I made sure that we at least test the existing weakptr.tst (it supposedly is flaky?!? resp. depends on "compiler behavior", but so far it seemed fine for me... if it starts causing trouble at some point, we can deal with it then).

And yeah, further improvements are planned for a future PR.