libera/clasp - IRC Chatlog
Search
15:57:43
yitzi
bike: On M1/bytecode-faso I am getting strange errors like the following after the image is loaded. `COMMON-LISP::TYPE-ERROR (:DATUM #<$CORE::SIMPLE-FUN @0x1078c05b9$> :EXPECTED-TYPE CORE::SIMPLE-CORE-FUN)`
16:05:58
bike
sometimes those type errors are caused by fasls having old stamps, but i'm guessing you're doing a clean build
16:07:59
bike
the fact it specifically wants a simple-core-fun seems unusual. do you have any idea where this type error is coming from? like, backtrace and so on
16:08:05
yitzi
I also the equivalent for "" literals in get-builtin-target-triple-and-data-layout which go away if I use `(make-string 0)`
16:14:51
yitzi
How were you thinking that the native FASL would work in build? Or how do you target it versus FASO?
16:18:14
bike
the native fasls aren't really in place enough to be causing your problems - or are you asking that as a separate question
16:21:20
bike
initially i was hoping the build could essentially be a bytecode build, and then once cleavir is loaded it would go through the fasls and compile native code for everything to produce native fasls
16:21:43
bike
but that's harder than i thought, so it would probably be more like the faso build, i.e. we do the bytecode build, then go through and read and compile-file everything again, but with the native compiler on
16:23:23
yitzi
I like the JIT idea, but I am wondering if we could do a pure native FASL replacement for BYTECODE-FASO and FASO. Just curious if my problems are FASO related.
16:34:23
yitzi
paulapatience: Fixes for the mac test CI have merged to main in case you want to rebase.
16:39:11
bike
it's more a flag to cmpltv. if the flag is on, whenever cmpltv compiles a bytecode module it also does b2b compilation, gives the bytes of a native module, and sticks them in the fasl as an attribute of the bytecode module.
16:43:59
bike
i figure eventually we could expose it as a compile-file parameter, and then maybe make it more fine grained later
16:46:17
bike
the problem is that native code generation is currently very tied to the faso format, so i had to untangle a lot of stuff, and my replacement use of the LLVM JIT hits errors i don't understand
19:28:33
paulapatience
yitzi: Also, I was wondering if you'd seen #1551 and had any comments on it. I was trying to package Clasp for Chimera, which uses musl
19:31:48
yitzi
paulapatience: it seems fine to me. You might want to rebase it because I think some of the changes were made in main already. bike usually does the final review of PRs. Generally we set PRs to draft until they are ready, then I usually just ask bike to review.