Search
Wednesday, 16th of May 2018, 8:29:38 UTC
13:24:30
drmeister
Bicyclidine: Did you get the cst changes?
13:24:45
loke
drmeister: I was thinking of taking another look at clasp
13:24:58
loke
drmeister: Can you tell me which branch I should be compiling?
13:26:11
drmeister
'dev' - but I haven't checked if it's working at the moment.
13:27:09
drmeister
I'll do a build and then push everything to 'preview' and you can use that.
13:27:13
loke
drmeister: Thanks, I'll try it.
13:28:06
drmeister
I'm adding custom allocators - and I haven't completely verified it.
13:28:27
loke
drmeister: Oh, for classes?
13:29:17
drmeister
No, lower level - to allocate specific C++ classes in custom MPS pools.
13:29:33
drmeister
C++ classes that are exposed to Common Lisp.
13:30:11
drmeister
So that they can be integrated seamlessly with everything else but be constantly compacted into their own small pool for good (hopefully) cache behavior.
13:30:45
Bike
what do you mean by custom allocators for slots, though?
13:33:49
Bike
drmeister: does cando compare lisp number objects with < etc. in C++? because i'm trying to delete those operators (clasp doesn't use them)
13:44:42
drmeister
Comment them out and try and build everything.
13:45:34
drmeister
Comment them out and I'll build cando
13:46:44
Bike
i deleted operator< etc from General_O, where i'm pretty certain they should not exist, but i can imagine them making sense for numbers
13:46:51
Bike
some kinds of numbers. not like Complex_O could work
13:47:25
Bike
in other news, slack ended support for the irc gateway literally yesterday
13:47:32
Bike
there's still bridge add ons by third parties,though
14:08:29
Bike
i forgot what to run to fix cleavir.py
15:00:36
Bike
with-debug-info-generator expands into code that copies the body like... six times... wtf
15:05:18
Bike
with-rtv needs to be bound around hoist-ast. which is kind of unfortunate, i think it does llvm stuff
15:56:20
Bike
seems to die after module-before-optimize teh same way
15:56:26
Bike
so maybe it's still an arrangement problem
16:08:41
drmeister
Are you building on your mac for the next hour?
16:08:58
drmeister
How about if I limit myself to 16 procs?
16:14:19
drmeister
So - what example crashes cst right now?
16:14:54
Bike
(lambda () (block nil ((lambda () (return)))))
16:15:20
drmeister
(funcall (clasp-cleavir:cleavir-compile nil '(lambda () (block nil ((lambda () (return))))))) ?
16:16:04
Bike
you don't need the funcall, it dies during compile.
16:34:23
Bike
note to self, multiple-vvvalue-bind fix is still not incorporated b/c of having to update sicl
17:04:51
DVSSA
drmeister: I've read through those papers you've sent me. When would you like to discuss them?
17:38:17
drmeister
Shinmera: what is your documentation system called?
17:41:14
drmeister
I have to read them and get the rest of the students to read them.
17:41:55
drmeister
I read them last six years ago
17:42:10
drmeister
I may be repeating myself
18:23:53
DVSSA
drmeister: Friday after 3pm works. What exactly are our goals for the discussion exactly?
19:45:50
Bike
foundation.cc has a bunch of string functions in it for some reason
19:46:01
Bike
like not even lisp ones, just random things that could be in the STL
20:00:16
drmeister
Those string functions are intrinsic to clasp lispification of C++ strings.
20:00:58
drmeister
We could move them - but we can't delete them unless their calls are replaced wherever they are made.
20:02:47
Bike
i deleted a few of them no problem
20:02:52
Bike
some of them are used in cando, though
20:02:57
Bike
they don't seem to be used in clasp
20:03:14
Bike
that is,none of them seem to be used in clasp
20:03:26
drmeister
Which ones are used in cando but not clasp?
20:03:46
drmeister
And yay on deleting them if they weren't used.
20:03:57
Bike
stripCharacters and trimWhiteSpace
20:04:37
Bike
there's also a couple #defines in core.h that should probably be moved to cando, or deleted if cando has moved on
20:04:42
Bike
and no, not sure what's going on with cst
20:04:54
Bike
gotta... stare hard at the ll i guess
20:05:21
Bike
i thought maybe i'd shared a variable between basic blocks, but the obvious case of that doesn't seem to be it
20:05:35
Bike
i don't suppose the validator could tell me the problem?
20:05:54
drmeister
You have the llvm-ir in a .ll file?
20:09:25
Bike
that's what the 'Human readable bitcode' switch does, right?
20:09:45
drmeister
It should generate .ll files rather than .bc files.
20:12:34
Bike
but i am curious about the validator
20:12:56
drmeister
The validator is supposed to catch all problems - but my experience is that it doesn't.
20:15:39
Bike
in this case it did though, right? that invalid record error.
20:16:37
kpoeck
(unintern t) seems to be one of the fastest way to get a segmentation fault: 11
20:16:41
drmeister
I thought that message was from trying to load it into the linker
20:17:17
drmeister
kpoeck: Can you give me an example that will segfault?
20:18:44
Bike
how does the linker know it's invalid?
20:18:47
kpoeck
trying a getReadOnly() in ool Package_O::unintern_no_lock
20:18:57
Bike
i don't think we have any kind of package locks to prevent uninterning
20:19:06
Bike
it's fine to unintern a constant
20:19:28
kpoeck
perhaps the same trick you did for makunbound works
20:20:23
kpoeck
so getreadonly() is the wrong test?
20:20:49
Bike
unintern is about symbols, anyway, not their bindings. it doesn't care if there's a constant binding or whatever
20:20:52
kpoeck
perhaps just not unintern something from common-lisp
20:21:00
Bike
it's just that, specifically, symbols in the CL package can't be uninterned, yes
20:21:03
Bike
probably also core: and stuff
20:21:25
kpoeck
than let we try to protect the internal packages
20:22:54
Bike
it's probably going to be more involved
20:23:02
kpoeck
at the end of the ansi-tests t prints as #:t, weird
20:23:35
kpoeck
I thought i is by a direct unintern, but no, must be something more complex
20:25:22
Bike
sbcl has a whole manual chapter http://sbcl.org/manual/index.html#Package-Locks
20:26:09
drmeister
kpoeck: Oh - sorry (unintern t) - got it.
20:26:49
drmeister
Bike: It doesn't know it is invalid - I think it fails while loading the bitcode. I've never been able to do anything with a bitcode file that includes bad bitcode.
20:27:12
Bike
but it doesn't crash or anything
20:28:42
drmeister
I don't know if it's the validator or the linker that is having a problem and if it's the linker I don't know if the linker is crashing. All I know is that I see that error message when I try to load bitcode files that have bad llvm-ir into any tool.
Wednesday, 16th of May 2018, 20:29:38 UTC