freenode/#clasp - IRC Chatlog
Search
13:55:59
drmeister
But I think we have to ditch Mac for development or ditch ld64 and use object files directly.
14:28:16
Bike
i'm not sure if that can just be moved out of the conditional, because stabs seem to be ordered
14:40:07
drmeister
I'd have to build this code and put some print statements in to see what's going on.
14:40:43
Bike
this linker works on a finer grain than files, it reduces each file to a series of definitions (and stuff) and works with those.
14:41:31
Bike
if you put those structs in it or something, it might generate atoms but not ones with debug info, so they won't show up in this loop.
17:53:10
Bike
drmeister: is apply0 like apply, except (a) its function argument must be a function, and (b) it only takes one more argument, which is a list or valist or whatever?
17:54:05
Bike
apply still seems like kind of a mess of cases. i could throw in some compiler macros for obvious cases
17:54:16
Bike
like knowing the number of fixed arguments ahead of time, and making the function an actual function
17:55:14
Bike
i guess if the list is nil it can just be rewritten as funcall. kind of unlikely to matter of course
21:05:38
kpoeck
Better reference w/o errors: https://common-lisp.net/project/cl-test-grid/library/array-utils.html
21:28:57
Bike
so clasp's sequence functions work with "iterators" that are kind of like the sequence extension's. i found that if we go a little farther with that and have make-seq-iterator return "pre-dispatched" functions to work with them, iteration functions go twice as fast ish
21:32:29
Bike
of course going to a specialized version based on type inference would still be better
0:13:55
Bike
i had another thought - we might be able to reorganize loading clos to be less side effect oriented. we can just make method and generic function objects directly (i.e. not through make-instance) and set up a discriminating function that's not through the usual method but still perfectly good
0:27:04
drmeister
Bike. Lang came back with a response that I forwarded to you. Could you come up with an ld64 patch?
0:29:03
Bike
i don't know whether i can just write in another N_OSO stab or what, and i can't imagine they'll accept a patch to a core system program based on just any dumb experiments i do
0:32:23
Bike
i was hoping we could get in touch with someone who understands the linker and mach-o and stuff, so they could give us some guidance, or just fix it themselves
0:58:15
Bike
if you want me to take some time to understand ld64 i can try, but i really doubt i'll be able to do anything very quickly
1:32:03
drmeister
I talked with Martin about this. He thinks it would be good to give them something that illustrates the problem if we could. They’d certainly rewrite it.
1:33:16
drmeister
Then compile it three times with different defines that turn on one part at a time.