libera/#clasp - IRC Chatlog

12:57:25 Bike oh, and i had the random thought that if mmtk scanners are anything like mps (which they may or may not be) we could jit new scanners for new unboxed structs, or something

12:57:28 Bike long way off though

13:54:22 drmeister The big hurdle for implementing mmtk is interior pointers. We should start thinking about it.

13:56:29 drmeister Boehm has a good scheme for identifying interior pointers.

13:57:04 drmeister I'm going to give you my book on memory management and ask you to start thinking about this.

14:08:31 Bike is it the handbook?

14:17:41 drmeister Yes

14:23:54 Bike ok, cool. i did read through temple's copy back when but that's different from having it around as a reference

14:26:45 Bike i figure for interior pointers it will boil down to 1) cutting out our own use of internal pointers throughout c++, and 2) convincing llvm not to optimize out exterior pointers

14:27:23 Bike do we actually use shutdown functions anywhere? like from startup-shutdown

14:28:22 Bike jit-finalize-repl-function ignores it

15:36:09 drmeister No, we don't use shutdown functions - but I leave them in as place holders.

15:36:19 drmeister Yes on "i figure for interior pointers it will boil down to 1) cutting out our own use of internal pointers throughout c++, and 2) convincing llvm not to optimize out exterior pointers"

15:36:45 drmeister but it would be unconventional - people assume C++ means you MUST support interior pointers. Can we do better?

15:37:15 drmeister I say we add interior pointer support and use it to try and catch interior pointers .

15:38:08 Bike add support so mmtk can understand interior pointers, you mean?

15:42:36 drmeister I think with MMTk it's on us to scan the stacks and identify interior pointers.

16:18:50 drmeister Did we talk about the XLSX system? I hit a problem with it again that I thought I fixed.

16:19:04 drmeister Maybe I got an update that overwrote it.

16:35:18 Bike i don't think we've talked about it ssince that header thing i worked out a while ago

18:20:09 Bike oh, i can't use initialize_gcroots_in_module because it wants a xep too... guh...

18:20:59 Bike and actually that goes all the way down to the GCRootsInModule constructor. so, i guess i can't make gcroots without a xep? and without gcroots i can't jit anything.

18:27:02 Bike i guess i can just make a dummy xep, but...

18:30:11 yitzi Bike: #1430 is done.

18:33:16 Bike oh, i'll take a look

18:33:21 Bike and eclector released even, cool

18:35:44 yitzi :D

18:58:36 Bike i'm not really sure what to do for this callback stuff now. i don't have as much control of the jit as i expected to

19:03:09 Bike putting in a whole xep just to satisfy gcroots seem wonky, and how do i know something won't get broken if there's no reference to the actual c function pointer i care about

19:10:49 Bike i mean the whole idea here is to jit something without being in a usual lisp compilation's llvm module

20:22:51 Bike what are the function pointers even used for? there' a lookup_function but i don't see anything calling it.

20:23:14 Bike no wait there it is