Search
Saturday, 14th of July 2018, 15:34:43 UTC
15:55:53
karlosz
beach: it occured to me that SCC could be generalized into a much more powerful algoirthm that deals with general constraints
15:56:27
karlosz
the current type inference algorithm doesn't keep track of executable branches, right?
15:56:51
karlosz
so perhaps instead of consstants in the lattice, types and other constraint information can be kept track of
15:56:52
beach
Correct. I had similar thoughts before today.
15:57:30
karlosz
when i had implemented it earlier i realized that as long as the universe has the finite chain condition it should terminate and work the same
15:57:41
karlosz
the universe for the lattice, i mean
15:57:58
beach
I am convinced that you are right.
17:56:40
drmeister
In preparation of GC'ing code and switching fully to MPS and saving/loading images I'm taking an inventory of objects created at startup.
17:56:50
drmeister
These are all of the ClosureWithSlots...
17:57:39
drmeister
https://usercontent.irccloud-cdn.com/file/7sc8juny/out-flame.svg
17:58:00
drmeister
I'm generating a backtrace for every allocation (about 15,000) and then a flame graph from those backtraces.
17:58:12
drmeister
That gives me a birds eye view of where what is happening
17:58:37
drmeister
I found several thousand redundant allocations using this that I eliminated already.
17:59:13
drmeister
fmakunbound allocates a closure. I was calling fmakunbound twice on symbols during bootstrapping.
18:01:17
drmeister
This is what is created at startup
18:01:19
drmeister
https://www.irccloud.com/pastebin/XRPyrmkd/
18:01:31
drmeister
That doesn't involve any Common Lisp code - that's just the interpreter
3:10:57
beach
Good morning everyone!
Sunday, 15th of July 2018, 3:34:43 UTC