libera/#commonlisp - IRC Chatlog
Search
18:53:23
aeth
contrapunctus: In context, it's more about SBCL's STYLE-WARNING/NOTE functionality rather than its (fairly limited) type checking
20:25:41
White_Flame
and then there's ((a . 1) (a . 2)) vs ((a . 1) (a . 3)) which both return the same value for key A
20:26:38
gin
I don't need to deal with associations being in different order right now. I will use "equal" for now.
20:31:35
pve
etimmons: I'm playing around with making a very simple asdf extension that just loads files in alphabetical order. It works fine, except it can't detect when a file has been added or deleted (unless I use :force t).
20:31:57
pve
etimmons: I have written the code that scans the directory for changes, but it's not clear to me how I should hook it up to asdf.
20:32:35
pve
etimmons: My first thought was to run the code in operation-done-p for compile-op, but then I thought maybe I should define a new operation scan-components-op and have compile-op depend on it.
20:33:05
pve
etimmons: Do you happen to have any thoughts on when/where this directory scan should take place?
20:33:29
pve
etimmons: I ask because you mentioned wanting to do this yourself for you own extension.
20:34:48
pve
the code is here: https://github.com/pve1/alphabetical-asdf/blob/main/alphabetical-asdf.lisp
20:39:57
etimmons
pve: I haven't looked at it yet, but I think the place to hook in would be the define-op for the system
20:40:43
etimmons
maybe an operation-done-p method that checks the current list of files against a cached list in the system?
20:45:49
pve
I understand ASDF has some kind of cache where it stores the results of operations. I was wondering if I get caching for "free" if I define a new op
20:46:24
etimmons
Since that scan-op would need to be performed before the plan for the load-op is computed
20:47:32
etimmons
yeah, that's why I was going to see if I could get away with hacking operation-done-p first.
21:04:23
jeosol
etimmons I just got out of a rabit hold chasing bazel, bazelisp for parallel compilation of my project. Not sure you remember, we had a discussion over from reddit chat (that you said you weren't initiaally aware it had a chat option)
21:06:18
jeosol
I haven't looked at it personally, but now that I fully understand the requirements of bazel and bazelisp (for large orgs, 200+, need lots of machines, etc), and so many issues (from others that used it for non C++ or non Java projects),
21:07:09
jeosol
I plan to start taking a look again at the repo I cloned a while back, may be get it to work for current SBCL
21:08:20
etimmons
I'm pretty sure bazelisp is used internally by Google, so hopefully it wouldn't have tooooo many issues
21:09:20
etimmons
I haven't looked at it myself in a very long time, but I also imagine it may not be the most portable thing if that's important to you.
21:11:10
jeosol
ok, but I thought you have to use bazel and bazelisp together. I could not get the former to install. I admit I don't understand how these systems are integrated. I was trying to download bazel to test the C++ tutorial
21:11:51
jeosol
I have dome some research, and I am mistaken, to think it was something simple, I could just drop in to do my parallel builds. But that is not the case
21:12:12
jeosol
It also requires huge amount of memory to run and you often need many machines to really get the benefits
21:15:22
etimmons
I forget if it had happened yet when we last talked, but SBCL has gotten rid of the lock around compile-file.
21:15:59
etimmons
would be interesting to see if just using plain old threads to compile things would work
21:18:09
jeosol
I have been in touch with Fare, the author, on possibly resurrecting this in some fashion, but I'd have to spend time to study the internals
21:18:53
etimmons
Hmmm, I think it got released in 2.1.4, but I see no mention of it in the release notes
21:23:13
etimmons
It's super nifty since it creates a new process that's ~identical to the first. Some application servers use it to, for example, load all the code, do any other expensive startup things, and then be able to spawn workers with nearly zero cost
21:23:41
etimmons
but it takes great care to use correctly and can easily lead to dead locks or other unforeseen complications
21:25:17
etimmons
ECL had an issue on MacOS where spawning a new process (nominally fork() then exec()) caused streams to get corrupted in the parent process if the exec() failed (e.g., the executable didn't exist)
6:22:40
^[
leo_song: It depends on your threat model. Some might argue that it's not safe to run Quicklisp anywhere, because it doesn't use TLS or verify downloaded packages.