freenode/#lisp - IRC Chatlog
Search
15:20:13
jmercouris
I would probably run some test to see on however many threads my application is most efficient
15:20:36
jmercouris
and then allocate or deallocate the necessary amount of threads, I would never directly introspect the hardware to answer questions about how my software will run
15:20:54
Xach
mfiano: I like it. I am running into trouble converting some code that was very single-threaded into lparallel because it relies on things like process cwd which is not thread-local.
15:21:39
Xach
And I think I might be running into sb-ext:run-program issues with threads too but I'm not completely sure.
15:21:59
jmercouris
also depending on the cache, fsb of the processor, well, more threads doesn't necessarily mean better performance, even for a workload that can be parallelized
15:23:13
scymtym
Xach: are the issues performance-related? there have been recent changes affecting the performance of spawning sub-processes in SBCL
15:23:59
Xach
scymtym: no, I'm getting errors about binding SB-<something>:FILENAME to a non-pathname and too many open files
15:24:08
pfdietz
I worry about overheating the CPU on my machine if I'm running on all the (hyperthreaded) cores. So maybe a cl-temperature package would be useful.
15:24:23
Xach
I don't have the exact messages handy, sorry, but I can get them sometime later today.
15:27:06
Xach
I also struggle with errors in workers and their interaction with slime - it means reverting more to a batchlike mode of debugging
15:31:14
Xach
Right now I build each quicklisp project one-by-one and see what succeeds or fails. This is done with a shared fasl cache so it is not completely clean, but with a clean cache on each run it would take forever. It already takes two hours. I am looking to build more projects at once, but there's an issue of clobbering each others' fasls.
15:31:43
Xach
So I was thinking about binning projects into 16 buckets with 16 fasl caches and using lparallel to build N at a time without fasl clashes.
15:32:34
Xach
This required working out a bunch of bad assumptions but I am not sure I worked them all out.
16:02:51
ck_
So it has come to this. The old nemesis, topological ordering, has returned for a final [and so on...]
16:43:55
_death
Xach: this sounds like the quickdoc talk on the last ELS.. though I remember feeling there had to be a better alternative than the solution presented
17:10:12
gibsonf1
Is anyone interested on collaborating to build a CL implementation of Numenta's Nupic Hierarchical Temporal Memory Machine Intelligence system? https://discourse.numenta.org/t/common-lisp-htm-implementation/6279
17:16:19
gibsonf1
I'm thinking of first capturing the structure of the system in a graph, cl-graph, and using that as the main way to work the data
17:19:43
shka_
gibsonf1: anyway, whole deal with HTMs is that you are building sort of bunch of perceptrons predicting each other activation
17:20:26
gibsonf1
I'm keen on the hierarchy part, taking the first set of results and processing those etc for higher abstraction
17:21:56
gibsonf1
I think they just haven't quite figured out how to get there yet, although the neuroscience supports it. That's why I'm taking a spatially correct graph approach work with
17:34:06
Xach
scymtym: "The value NIL is not of type <pathname stuff> when binding SB-KERNEL:FILENAME"
18:00:31
Xach
scymtym: Not sure. This is from within a lparallel worker and I don't have the backtrace.
18:01:28
Xach
Also getting "#<SB-IMPL::HANDLER INPUT on BOGUS descriptor N ... has a bad file descriptor."
18:57:56
Kurvivor
i have built it in cygwin, and to my horror it crashes due to stack overflow any time i try to compile lisp files
19:20:10
jackdaniel
Kurvivor: I'm going to sleep soon, so I won't be able to help immedietely, but please report the issue on ecl's issue tracker
1:49:27
fouric
does there exist a sockets library for CL that allows me to access raw sockets (https://en.wikipedia.org/wiki/Network_socket#Raw_socket)? IOLib and usocket appear to both only provide the higher-level datagram (UDP) and stream (TCP) sockets interface
1:49:46
fouric
context: i'm trying to replicate https://jvns.ca/blog/2014/08/12/what-happens-if-you-write-a-tcp-stack-in-python/