Search
Saturday, 26th of November 2022, 16:38:22 UTC
16:38:23
yitzi
In src/scraper/tags.lisp on line 294 there is a #+sbcl
16:38:30
consolers
thx i shouldve checked
16:39:14
consolers
i'll have to audit those next
16:39:17
yitzi
You are gonna need equivalent handlers for the other implementations you are trying.
16:39:25
jackdaniel
burn asdf, burn it! :)
16:39:55
yitzi
Nobody asked you, jackdaniel you ASDF hater. :P
16:40:57
jackdaniel
asdf is the best thing since sliced bread, but bread is a poor choice when it comes to build systems
16:41:48
yitzi
Although, since it has become a monopoly, perhaps we should rename it to TSDF = The system definition facility
16:42:28
yitzi
Yeah, the feature expressions are bugged. I had write my own groveler to avoid them.
16:42:47
consolers
i dont want to publish this, but if i upload my copy to a private git repo will i be able to share it with you?
16:42:59
consolers
it's changes to the tarball base commit
16:43:31
yitzi
Whichever works for you. You can do a patchfile if you prefer.
16:43:34
yitzi
on Gist or something. Whatever.
16:44:52
jackdaniel
don't foget to encrypt it
16:45:00
jackdaniel
and send a physical letter with the key
16:45:40
consolers
i dont know whats possible with private repos on gh i'll find out another day
16:45:46
yitzi
CC jackdaniel so he can attempt a man-in-the-middle attack.
16:47:48
jackdaniel
ACTION dresses in black to hind in the shadow
16:48:00
jackdaniel
and puts his hacking hood on
16:48:39
yitzi
Nice. Does that look at all like a Halloween ninja costume?
16:49:41
jackdaniel
like zenzu in gintama
16:50:19
jackdaniel
or now, this is even better, like gintama himself: https://i.ytimg.com/vi/L8OSPrUt6Qo/hqdefault.jpg
16:51:02
yitzi
Perfect. I'll never be the wiser.
16:55:58
consolers
gh refused to let me upload because my access token didnt have permissions for ci
16:56:08
consolers
which file do i delete?
16:58:27
consolers
yitzi Can I invite you as a "collaborator" on github?
16:58:40
consolers
I cant send /msg to you because i'm not logged in
16:58:57
yitzi
My github handle is yitzchak
16:59:56
consolers
i expect to be rebasing all the time, but you can see that it's possible to breakfree from the tyranny
17:02:45
consolers
i'm off now - hopefully i can do more next week. - the notes are under the sysdcl/ directory
18:45:25
Bike
wow, it's like the exact opposite of the guix problems.
18:53:01
yitzi
I dunno. I fail to see the motivation.
18:59:41
Bike
right. getting shot out of a cannon is the opposite of falling off a cliff
18:59:56
Bike
it would be nice if we could use impls other than sbcl, though.
19:00:34
Bike
not a big priority of course
19:02:37
yitzi
That's true. ECL or CLISP would be especially nice. SBCL is booted off of one of them in most packages.
19:03:41
Bike
do you know what exactly we do that's sbcl-dependent? i know we've used the scraper with sbcl just about forever, but its basic tasks seem portable enough
19:07:19
yitzi
Nothing specific. There are some conditional stuff for SBCL that just needs the equivalent for other implementations. We clone ASDF for ourselves but aren't careful to make the booting impl not see that. It breaks when booting with ECL since they use a modified ABCL.
19:09:23
Bike
oh, so nothing too complicated. nice.
19:19:59
Bike
it's more blue-sky, but if we keep up with the VM stuff, i'm hoping it might at some point be possible to produce bytecode in sbcl/whatever, which we then load into the clsap c++ program, and that can be our build process. but there are plenty of intermediate steps that would need doing there.
19:26:40
yitzi
That would be pretty amazing.
3:37:15
consolers
so i hit the llvm orc bits. if i use gentoo it'll be 210MB and over 6 hours of compilation with OOM problems, i could download a release amd64 tarball thats just half a gig
3:37:47
consolers
gentoo is already dropping packages because they dont compile with clang 16
3:38:30
consolers
maybe shelf life for 14 is not going be long. i alrady had 7,10,11,13 locally
3:41:36
drmeister
Bike: Are you online?
3:41:41
drmeister
Did you make it back yet?
3:43:40
drmeister
I got the parser working and I can generate all the training molecules and their focus monomer contexts.
3:43:53
drmeister
https://www.irccloud.com/pastebin/oChqvRRu/
4:17:42
drmeister
First training molecule...
4:17:44
drmeister
https://usercontent.irccloud-cdn.com/file/Ysqoldea/image.png
4:17:55
drmeister
It takes 1.5 seconds
4:19:40
drmeister
So with 223 trainers assuming we need to generate 500 conformations for each trainer and it takes 1.5 seconds per conformation ...
4:20:57
drmeister
47 hours to generate all the conformations with just two functional groups.
4:23:15
drmeister
Each additional functional group is 18 minutes.
4:26:27
drmeister
The little ones are super fast
4:26:28
drmeister
https://usercontent.irccloud-cdn.com/file/Qdq0KmLy/image.png
4:27:36
Bike
what takes the calculations so long?
4:28:22
drmeister
It does a non-linear optimization from essentially random coordinates.
4:28:41
drmeister
It first optimizes the geometry where the atoms are bonded together but they repel each other.
4:28:52
drmeister
Then it dials down the repulsive force to zero.
4:29:44
drmeister
It uses three non-linear optimizers (1) steepest descent (2) conjugate gradients (3) truncated Newton
4:29:59
drmeister
It switches between them as the solution gets closer and closer to optimal.
4:30:07
drmeister
Lots of complicated math.
4:31:05
drmeister
I think I've nailed it. Now I just need to glue a bunch of the code together.
Sunday, 27th of November 2022, 4:38:22 UTC