libera/#clasp - IRC Chatlog

19:36:11 drmeister Hello

19:36:17 drmeister Yes - we work it into the build.

19:37:30 drmeister It will be a temporary generated file - right? We can put it in build/<GC>/generated/

19:37:54 drmeister The same directory that the scraper writes out all of its files to.

19:50:52 Bike Sure, makes sense to me

19:51:11 Bike right now i just have a function that outputs a string, and i've been writing it into src/main/clasp_gc.cc, so i haven't had to adapt the build

20:01:45 Bike is there any way i could validate whether the deallocators/finalizers are working correctly? I think you said they're both sort of obscure features?

20:02:10 drmeister Uh

20:02:50 drmeister Only a few of them are active I think.

20:04:28 drmeister For the finalizers - you could generate a printf statement in every finalizer and then watch them print when you run clasp.

20:04:55 drmeister You can target it by looking for classes that use finalizers - like Stream_O (I think). Hang on.

20:04:55 Bike oh, do they just run in normal use?

20:05:15 Bike i'd be satisfied if things didn't just crash and burn then

20:05:32 Bike i don't know how these tables are accessed is all. they're jused for jump tables but i don't know if they're indexed by stamp or what

20:05:44 drmeister Yeah - the ones that have this:

20:05:48 drmeister https://www.irccloud.com/pastebin/BWKehG5H/

20:05:53 Bike but since i generate it from a hash table the order is kind of arbitrary

20:06:02 drmeister static bool constexpr NeedsFinalization = true;

20:06:43 drmeister Right - the jump table.

20:07:06 drmeister Digging deeper.

20:18:27 drmeister I only see obj_finalize being called from MPS code.

20:18:52 drmeister I believe gctools__deallocate_unmanaged_instance is for static vector and things like that.

20:19:02 drmeister No - not static vector.

20:19:10 drmeister That would still be managed right?

20:20:14 drmeister I don't think obj_finalize or obj_deallocate are currently being used.

20:20:47 Bike Alright, I guess that works. Are they stuff we want to keep around for later?

20:20:53 drmeister Yeah.

20:21:21 drmeister obj_finalize looks like it's used by finalizers in MPS - but not boehm - I'm not sure why.

20:23:37 Bike well, it does look like the table index is just the stamp

20:23:41 Bike with no wtag, i guess?

20:23:50 Bike in which case the entries really do need to be ordered correctly

20:23:53 Bike for obj_finalize anyway

20:24:14 drmeister I think I implemented that stuff before wtags

20:29:10 Bike nonetheless, this seems like it could be a really confusing bug later, so let me see if i can get those sorted

20:34:55 Bike done.