Search
10:41:15
stassats
can fold (defun foo (x y) (if (eq y x) (eq y x) t)) to T by extending constraint-propagation a bit
10:41:20
stassats
but again, where does that come up
10:51:10
stassats
ok, it comes in a few places when building sbcl
10:54:49
stassats
even found a place where to remove a line
10:56:32
stassats
loading mcclim it gets triggered 2 times
11:18:59
stassats
but have to disentangle EQL and EQ constraints, maybe add some new, like =
11:19:44
stassats
and then add transitivity to > <, use them, also improving check-bound elimination
17:28:30
pfdietz
Wrote some more RTP tests for equal, equalp. Not seeing any failures.
18:06:31
stassats
that's encouraging, i was going for no failures too
18:23:49
stassats
well, i found a bug in the EQ transform
18:24:24
stassats
(lambda (x) (declare ((array * (10)) x)) (let ((y (adjust-array x '20))) (declare ((array * (20)) y)) (eql x y))) incorrectly folds to NIL
18:26:11
stassats
ancient as in present in cmucl
18:26:31
stassats
should be an easy fix, since i wrote the EQUAL transform with that in mind
18:28:51
stassats
but not today, it's been there for 30 years, it might wait a week or two (expecting the looming freeze)
18:31:51
joshe
I'm seeing a strange failure in ir2-optimize-comparisons under clisp
18:33:24
joshe
while compiling target-hash-table.lisp it seems xfloat-zerop is being called on a #<SB-C:TN t1[RAX] :NORMAL>
18:33:39
stassats
well, i see a potential problem
18:34:19
stassats
but not likely to be the cause
18:34:35
stassats
i'm calling nreverse where i shouln't
18:34:54
joshe
I have a branch with parallel make-host-2 under clisp, if you were thinking of trying a clisp build
18:34:58
stassats
i'll remove that anyway
18:35:08
stassats
thinking again, no, it might cause problems
18:36:19
stassats
the danger of nreverse, when i put it there it was safe, but then after modifying some other code it stopped being safe
18:45:01
joshe
and it's in genesis-2, which is further than before
18:52:35
stassats
that adjust-array could be smart, not just punt, but that's hard