Search
Tuesday, 13th of November 2018, 14:35:19 UTC
14:35:44
pfdietz
It doesn't say that notinline may not be respected, it just says that the things you might use to truly test if it had been inlined can't be done.
14:36:25
pfdietz
Something that doesn't show up in external behavior doesn't exist, from my point of view.
14:53:12
Shinmera
Assuming a deterministic compiler, you could compile twice and compare size of the disassemble output :^)
14:58:19
stassats
some inlining may be smaller or of equal size to a full call
16:50:26
Bike
DISASSEMBLE seems to be broken again, and i'm not sure how it works
16:50:48
Bike
it tries to dump a module called :no-module, which is the sort of default value for a variable called saved-module-from-clasp-jit
16:50:58
Bike
which i suppose is supposed to contain a module, but nothing i can see would do that
17:46:22
Bike
drmeister: kevin needs to talk to you. also, what was the thing to get bclasp+cleavir exactly how the build does it?
17:53:36
drmeister
Kevslinger: I'm in my office.
17:53:59
drmeister
Bike: There is the src/lisp/kernel/cleavir/workspace-load.lisp file
18:05:15
Bike
"Could not find the class CLEAVIR-IR:ENCLOSE-INSTRUCTION." mysterious things continuing to happen...
18:06:17
drmeister
Huh - I saw that as well a few days ago and didn't deal with it. Is the class defined at that point?
18:06:33
Bike
i'll try loading instead of compilefiling
18:06:57
drmeister
Doesn't that load the source code in exactly the same way the build system does?
18:07:15
Bike
i mean, loading workbench-load itself
18:09:50
Bike
it's while "Loading/compiling" graphviz-drawing, which is after general-purpose--instructions
18:10:12
drmeister
It's a mystery then - why does the build system work?
18:10:17
Bike
but the class indeed doesn't seem to be defined
18:10:30
Bike
it should work, since the file was loaded and all
18:10:59
drmeister
Can you figure out why the class isn't defined?
18:11:13
drmeister
This sounds peculiar - like kpoeck's issue with MPS
18:20:52
Bike
other classes in the same file are still defined
18:20:58
Bike
something dumb is underway here
18:21:18
Bike
the enclose-instruction thing is reproducible, though. the mps problem didn't seem that way.
18:21:31
drmeister
One problem at a time.
18:22:46
Bike
i... the file doesn't have a definition for enclose-instruction
18:23:49
Bike
oh. right right i moved them
18:24:25
drmeister
Why does the build system work?
18:24:45
Bike
well whatever i'm doing now and whatever the build is doing are using different file lists.
18:24:58
Bike
the build's is correct but this bclasp build is out of whack.
18:25:14
Bike
missing a file i added a couple weeks ago.
18:26:07
Bike
this is going off of cleavir-file-list.lisp, whereas the build uses cleavir_file_list.py
18:26:20
Bike
i think i edited the latter manually because i forgot how to regenerate it properly
18:26:22
drmeister
Have you been editing them by hand?
18:26:27
drmeister
Ah - there is a script for that.
18:26:49
Bike
regenerate-cleavir-file-list.lisp
18:28:53
drmeister
The script segfaults?
18:30:53
Bike
ah, no. maybe it's intentional.
18:32:00
Bike
yes, i think it quits at the end, but i was running it in slime and that was rather unexpected.
18:40:30
Bike
ok, now i'm getting the error i expected to get.
18:40:50
drmeister
We get the errors we deserve
18:41:52
Bike
i dont deserve any errors. i'm as pure and sinless as the first snow of the year
18:44:51
drmeister
ACTION is not looking forward to shoveling that snow.
19:09:49
drmeister
shiho: How did the smarts stuff go?
19:47:40
shiho
scymtym: Are you online? We have the relate methods building a hierarchy - but it's inside out.
19:47:49
shiho
The tests that should be first are last and vice versa
19:48:12
shiho
Is there a simple fix that I'm not seeing?
19:48:39
shiho
The first relate is being passed (... nil #<atom-test>)
19:48:50
scymtym
did you use the method i suggested and it is still flipped?
19:49:15
shiho
Shiho says yes - but a little bit modified.
19:49:31
shiho
I need to see the method you suggested - I'll go to my computer and check the log
19:50:00
scymtym
https://techfak.de/~jmoringe/linearize-test.lisp
20:35:19
scymtym
are you still thinking? i will have to leave soon
20:43:54
drmeister
Sorry about that - someone came and needed me for "5 min" - 45 min later I'm back.
20:44:09
drmeister
Shiho got the order right following your instructions.
20:44:48
scymtym
i know what you mean (regarding colleagues needed 5 min of one's time) :)
20:45:10
scymtym
great. are there other open problems?
20:45:21
drmeister
I don't think so - I'm just processing what she did.
20:45:46
drmeister
She created a struct 'chain' and I am trying to figure out how to map that onto our 'chain'
20:46:38
drmeister
Thank you very much for your help - I'll get back to you on how it turns out.
20:49:37
scymtym
sorry, that it is kind of complicated
20:50:00
scymtym
oh, the chain struct was supposed to be a placeholder for your actual chain structure
20:50:24
scymtym
so that you would only need one cons of intermediate data
20:51:01
drmeister
I think we can use our chem:chain if we set up slot readers and slot writers properly
20:52:12
scymtym
that was my idea at least
20:58:03
drmeister
The order is correct now .
22:03:59
Colleen
kpoeck: drmeister said 7 hours, 31 minutes ago: cclasp-boehm is working again. It was an out of order c++ static constructors problem.
22:04:39
kpoeck
with latest dev, and the optimisation to bring fli.lsp and inline.lsp to front, I can't build cclasp
22:05:02
kpoeck
get a (ERROR "Missing ~a" #P"src/lisp/kernel/lsp/fli")
22:05:35
kpoeck
Why would fli always be in the list of compiled files?
22:12:08
kpoeck
I believe either move-to-front should not error, or it should be assured that target is in files-list
22:13:05
kpoeck
Otherwise we are limited to distclean builds
22:13:19
drmeister
It should always be part of the compiled files. Where did it go?
22:13:47
drmeister
Oh - right - it's a partial build.
22:13:56
drmeister
No good deed goes unpunished.
22:14:01
kpoeck
But fli could be compiled and not outdated
22:14:13
kpoeck
so you woudn't recompiled it
22:15:35
kpoeck
i believe inline is always compiled
22:16:06
drmeister
I'll change move-to-front to maybe-move-to-front and change it like this...
22:16:18
drmeister
https://www.irccloud.com/pastebin/iSEdi0lu/
22:16:44
kpoeck
I did nearly the same :-)
22:17:54
drmeister
I just pushed the change - thank you for catching that.
22:18:36
kpoeck
It would have happened to you sooner or later :-)
22:29:17
kpoeck
and build/clasp starts again!
22:30:29
drmeister
I can't quickload cl-unicode - dammit.
22:30:52
drmeister
kpoeck: Have you tried ppcre or cl-unicode?
22:32:25
Bike
i imagine we might have problems with cl-unicode since we kind of don't support unicodez
22:32:54
Bike
but maybe it's portable enough. i don't know
22:37:33
kpoeck
you need it to load first with sbcl
22:37:34
kpoeck
tht donwloads some files
22:37:34
kpoeck
that you can use it with clasp
22:38:10
kpoeck
I have that in my .cache of quicklisp
22:39:28
kpoeck
something in the cl-unicode.asd is not executed right in clasp
22:40:36
kpoeck
perhaps this: (load-op (o c) (symbol-call :cl-unicode '#:create-source-files))
22:41:14
kpoeck
yes the files in build are not downloaded with clasp
22:41:37
kpoeck
so I exectuted this with sbcl, files got downloaded and than it worked in clasp
22:42:00
Bike
probably we ought to fix that.
22:58:41
drmeister
Hi - I'm back . (on the train).
22:59:35
drmeister
I created this issue - but I couldn't reproduce the problem because (ql:quickload :cl-unicode) worked in clasp after I did it in sbcl.
22:59:43
drmeister
Ha! You saw the same thing!
23:01:00
kpoeck
my bet is on perform (load-op (o c) (symbol-call :cl-unicode '#:create-source-files))
23:01:14
kpoeck
in (defsystem :cl-unicode/build ....)
23:03:03
drmeister
So I can get cl-interop to load now.
23:03:25
drmeister
Dammit: cl-interpol!!!!!
23:07:44
drmeister
There's an issue with readtables - clasp doesn't like other people's readtables.
23:09:18
kpoeck
remember the issue: the standard reatable sucks!
23:09:34
drmeister
I’ll use cl-ppcre for now.
23:09:54
kpoeck
Since clasp was supposed to change to eclector, I didn't look at the reader at all
23:10:25
kpoeck
the standard readtable sucks I meant to type
23:11:21
kpoeck
https://github.com/clasp-developers/clasp/issues/391
Wednesday, 14th of November 2018, 2:35:19 UTC