freenode/#clasp - IRC Chatlog
Search
3:38:47
Colleen
drmeister: Bike said 7 hours, 48 minutes ago: i figure you've already seen and worked out a fix for the Reader_O removal problem- but i'm looking at it and thinking we shouldn't be parsing the arguments strings at startup anyway
3:39:55
drmeister
Because we need them to construct LambdaListHandlers when we define the functions.
3:40:25
drmeister
I got rid of Reader_O and didn't expect the bootstrapping problem that I subsequently ran into.
3:41:02
Bike
i mean, if we have that a function has arguments "(x y z)" we could somehow specify the list for startup, instead of rereading the actual string
3:41:50
drmeister
Yes, we could do that - it would be better generate code to construct the LambdaListHandler_O objects - but that would take some work.
3:46:29
drmeister
It's not a high priority right now - we can continue to parse the strings at startup and sort out a way to do it at scrape-time later.
3:49:11
drmeister
It might be - this morning I thought it would be better to just get the regular reader to parse the strings - now I'm several hours into it and it's still broken.
3:50:04
drmeister
So I made a change that calls reader macros directly when they aren't set up properly.
4:56:06
drmeister
::notify Bike The other reason to do it at runtime is that then I use the same approach for defining lambda lists in clbind. Now - we could come up with another way to write lambda-lists that uses C++ template programming... But that sounds - bleh.
8:29:22
heisig
And I have to admit that, in lack of a specializing client, I never actually ran the hoisting part of the transformation. So it may contain silly bugs.
12:41:16
heisig
drmeister: It replaces each non-immediate literal object in a flowchart (and each load-time-value form) by code that is evaluated at load time.
12:42:34
drmeister
Currently there are load-time-value ast nodes - are there now load-time-value HIR instructions?
12:45:36
Colleen
karlosz: beach said at 2018.07.16 07:40:25: In SFA, if you propagate only when the left-hand side of the assignment has a single defining occurrence, then you get the same result as if you use SSA.
12:50:55
Shinmera
General feedback regarding the notes by the way, would you guys prefer it if the notes were shown in the same channel only, rather than on any channel the same user is observed on like it is now?
12:52:20
karlosz
beach: i agree. i was thinking we have type1 and type2 assignment instructions. type1's output has a signle def, whereas type2 has multiple defs. many flow algorithms treat type1 trivially by just copying the information, whereas type2 will have to perform some meet operation
12:53:02
karlosz
so the different class would only be to allow using generic dispatch instead of a manual case statement