Revert #1571 “perf/factor ~ deduplicate divisors”

[nicoonoclaste]

It was a draft PR, not ready for merging, and its premature inclusion caused repeated issues, see 368f473 & friends.

This reverts commits 3743a3e, ce218e0, and b7b0c76.

Close #1841.

[nicoonoclaste]

I ran the tests in a loop for 45 minutes, and confirmed there's no remaining bugs exposed by the tests (or at least, the ones left are highly unlikely to be hit)

[nicoonoclaste]

Performance wise, this seems to be a ~3% regression, so the complexity/speed trade-off seems to be definitely there.
We can revisit that if/when I find a non-cursed way to formalise and implement that GCD tree idea.

[nicoonoclaste]

@sylvestre the CI failure seems unrelated to my changes, or factor at all

[sylvestre]

are you sure?
it seems clearly factor

PASS: tests/factor/t20.sh
gawk: cmd. line:1: (FILENAME=- FNR=168) fatal: node.c:415:make_str_node: r->stptr: can't allocate -2147483645 bytes of memory (Cannot allocate memory)

[nicoonoclaste]

it seems clearly factor

PASS: tests/factor/t20.sh
gawk: cmd. line:1: (FILENAME=- FNR=168) fatal: node.c:415:make_str_node: r->stptr: can't allocate -2147483645 bytes of memory (Cannot allocate memory)

Thanks! I'm not sure how I missed that

[nicoonoclaste]

@sylvestre It's also broken on master :

PASS: tests/factor/t20.sh
gawk: cmd. line:1: (FILENAME=- FNR=168) fatal: node.c:415:make_str_node: r->stptr: can't allocate -2147483645 bytes of memory (Cannot allocate memory)

Generally, there seems to be a lot of failures in that testsuite, so something is really wrong.

PS: I had a look, and the “Run GNU tests” task seems to just pull in whatever is in coreutils/coreutils, so pushes to the GNU repo can break it. Given that, and that it's broken on master anyhow, I would suggest getting rid of that task (at least until someone fixes it up) ; CI that we teach people to ignore, is worse than no CI.

[nicoonoclaste]

Force pushed to remove the empty commit I used to rerun CI