freenode/#sicl - IRC Chatlog
Search
7:09:25
no-defun-allowed
Just to check, would you say that the SICL style guide, and any other expectations of programming style, were derived from Norvig and Pitman's LUV slides?
7:10:33
no-defun-allowed
Today I crept closer to the end of the credit application process, as one of the course coordinators got back to me with the required information on their course. One more form to go!
7:11:48
jeosol
no-defun-allowed: seen a few or your posts? are you transfering schools? are you in the US or Europe?
7:12:16
no-defun-allowed
jeosol: I am transferring universities and live in Australia for the time being.
7:12:54
jeosol
beach: that can get reduntant; there is also the other style: e.g., preceeding with get- , e.g., get-first-name
7:14:53
jeosol
beach: I read some of your comments on guides on #lisp, using car vs. first, etc. Using some coding standard goes a long way and reduces mental strain for others reading code
7:18:58
no-defun-allowed
beach: My jokes about moving to Europe have been better received than jokes about growing vegetables by my (admittedly relevant) family.
7:30:39
beach
There is Bike and karlosz from the US. no-defun-allowed from Australia. And ebrasca is from South America, right?
7:32:30
jeosol
I have also seen a few CL related talks in Atlanta and other places but not sure those guys are on here.
7:32:33
beach
Yeah, but they don't hang out here much. In #lisp the do. And there you have some Chinese and Indian people too.
7:33:13
jeosol
That's interesting. I like the SICL/compiler effort and the goal to have a common base for implementation
7:34:50
jeosol
Having fought dependency hell in python (and still do), I never get the same issues working with CL. For my application, I even upgrade my SBCL compiler "almost" with each update and only get minor incomptibility issues or none at all.
7:35:03
beach
And it's not just the compiler. Things like Clostrum, Trucler, Eclector, Incless, etc. could be very useful as well.
7:35:45
no-defun-allowed
I have had only one incident of updating SBCL breaking code I have used, and that (cl-simd) is fair game as it uses the internal assembler framework.
7:35:47
jeosol
I think people working on systems don't underestimate the importance of a stable environment and not having to fight the language all the time
7:36:28
beach
And Xach has been known to use SICL LOOP to detect violations of the standard, because it is more strict than MIT LOOP.
7:36:29
jeosol
no-defun-allowed: it does happen. It's only recently I got some VM: package related issues and something with ironclad also, and there were quick fixes
7:37:11
no-defun-allowed
On the other hand, Common Lisp has an ANSI standard, and extensions have been informally standardised. This is not the case (to my knowledge) for Python, so it is more likely there for implementation changes to break things.
7:38:22
beach
A large part of my motivation is to cut down on the collective maintenance burden, but I don't think I have been very successful yet. There are quite a few Common Lisp implementations, and they all have specific code for most of the system, even though many "modules" could be shared.
7:39:20
jeosol
beach: while we are at it? are you aware of fare's PIOU https://gitlab.common-lisp.net/qitab/poiu for parallel compilation
7:39:56
jeosol
I haven't been able to get it to work, but he suggested it's lacking and I work to make it work with recent SBCL and asdf
7:40:19
no-defun-allowed
On the other hand, Ruby and JavaScript have standards, but they are still moving targets. Perhaps why we don't have as much for library version "pinning" (oh look, another domain in which we pin things!)
7:41:02
jeosol
Essentially, I am trying to speed up my compilation process (not for code development, but more like docker deployment to a new machine - this is experimental)
7:41:39
jeosol
beach: you said you haven't been "very" successful yet. SICL is a great/large effort if I understand correctly - to have a common base for these different implementations ...
7:43:01
beach
Yes, but I would like to see some existing implementations use some of those modules. I don't think that is likely to happen though.
7:43:44
jeosol
why not? we talked about your recent work with the callsite optimization - that's something I could use to gauge speed up.
7:44:19
jeosol
I'll probably tag my repo now so I can get a baseline for the current running case. I do lots of number crunching, so may be it will help
7:47:23
jeosol
my jobs run for weeks, and the weather issues in the last few months, and power issues have killed my runs. I could certainly use a 2x-3x speed up
7:48:01
fiddlerwoaroof
There still are rough edges, though: not everything has migrated to darwin/arm64
7:52:59
karlosz
i bought this laptop in 2016 - that statistic almost made me feel like Moore's law strikes back
7:53:08
jeosol
Looking at the M1 sheet on apple, it does to have impressive stats and benchmark performance
7:54:47
fiddlerwoaroof
Here's a benchmark I did a while ago, building abcl: https://twitter.com/fwoaroof/status/1330040967882268674?s=20
8:29:30
beach
My main mission today is the annual visit to the dentist. I'll be back in several hours.
14:38:50
ck_
?:/ I mean you both are on top of it, and if i recall you don't share the apsp, so that can't be the reason. there must be others :)
14:40:37
beach
OK, so I have a few hours left of the day. Last time I worked on HIR and MIR was when I no longer had to preprocess several instruction types that needed to be turned into function calls before closure analysis.
14:42:07
beach
Before that, in particular the ENCLOSE instruction was modified. So it no longer had the associated ENTER instruction stored in it. Therefore, in MIR-to-LIR, I had to find the ENTER-INSTRUCTIONs in a different way.