freenode/#lisp - IRC Chatlog

17:49:08 raynold ahh it's a wonderful day :D

19:04:30 phoe evening

19:04:38 phoe at least over here

19:10:10 astronavt anyone here have a good (or not good) experience using Clozure CL to do Mac GUI work?

19:13:10 phoe astronavt: #ccl might be able to help best

19:13:27 astronavt thanks

19:33:46 Xach attila_lendvai: http://report.quicklisp.org/2017-08-30/failure-report/hu.dwim.reiterate.html#hu.dwim.reiterate_iterate ??

20:20:02 attila_lendvai ACTION is looking

20:26:18 attila_lendvai Xach: I wanted to be conservative, not pushing some of the brand new patches, but apparently I ended up pushing something half-baked. I pushed everything I had, it should load fine now

20:39:24 malice Is the error signalled when accessing out-of-bounds element of array/vector well-defined?

20:39:37 malice I can see that SBCL signalls some SBCL error, is that so for each implementation?

20:43:32 _death yes, it's not defined

20:58:38 jfrancis I just upgraded my quicklisp packages and broke my code. Drakma is now puking on the self-signed cert I'm using in my test system. If I do (drakma:http-request uri :verify nil ...) shouldn't it ignore the validity of the cert? Or am I reading the drakma docs wrong?

21:18:31 scymtym jfrancis: the current quicklisp release includes a combination of cl+ssl and drakma in which :verify nil does not work (because some necessary changes in drakma did not make it into quicklisp). in the meantime, a drakma release to fix this has been prepared and the next quicklisp release is probably imminent

21:24:22 phoe I am preparing a writeup post about the current state of the UltraSpec

21:36:14 jfrancis scymtym: thanks. in the meantime, I rolled back the quicklisp upgrade from a backup and that got me back up and running.

21:37:38 phoe jfrancis: you can always clone a recent version into ~/quicklisp/local-projects

21:38:19 Shinmera Or use Quicklisp's own versioning to roll back to the previous dist state.

21:38:27 Shinmera Though I don't recall how to actually do that :(

21:38:29 jfrancis I'll probably just wait. Right now, making progress on my application is more important than having the latest libraries. Only so many hours in a day.

21:39:03 jfrancis Yeah, I know I can do that with quicklisp, but it was quicker to pull from a backup. Not enough time to learn everything I wish I could.

22:20:34 phoe https://blog.teknik.io/phoe/p/375

22:21:11 phoe I hereby announce the presence of #clus - IRC channel linked to Discord where I hope to discuss matters related to development of Common Lisp UltraSpec.

22:26:49 pillton beach: You're welcome.

22:28:52 drmeister Hey lispers - any CLOS implementation experts online?

22:29:03 phoe drmeister: ask the question

22:29:12 drmeister I'm implementing beach's fast generic function approach and running into metastability issues.

22:30:08 drmeister Nasty ones that I'm guess are due to when I'm compiling a dispatch function it invokes other generic functions with missing or incomplete dispatch functions and that so on.

22:30:46 pillton I think beach (and others) call that problem satiation.

22:30:50 drmeister There probably isn't a good answer for this but how does ECL avoid this.

22:30:51 phoe are you bootstrapping the new dispatch from an existing one?

22:31:04 drmeister I am bootstrapping the new dispatch from the existing one.

22:31:23 drmeister That doesn't seem to be enough.

22:31:42 pillton There is a bit in the AMOP book that talks about it I think.

22:31:53 drmeister Clasp starts up using the ECL dispatch method and then switches all of its generic functions over to the new method.

22:32:37 phoe all of them.

22:32:49 drmeister While running with the old ECL dispatch method each generic function accumulates a "call history" of specializer lists that it has been called with and effective methods for those specializer lists.

22:32:53 phoe So, in theory, you are not using any of the old dispatch's functions, correct?

22:33:22 drmeister The generic function dispatch compiler simply compiles that alist into native code alist search.

22:33:47 drmeister Well, I use this to convert over...

22:33:59 pillton phoe: It is not just the dispatch functions, you have to ensure that all of the method objects use the new meta classes too.

22:34:01 drmeister https://www.irccloud.com/pastebin/onuPO4jc/

22:34:14 drmeister Note the list of functions that are not converted over.

22:34:29 drmeister That list grows longer the more times I try it - until I gave up.

22:34:40 drmeister Clasp can run with both dispatch methods side-by-side.

22:35:20 pillton beach has said in the past that he has a paper on this problem.

22:35:21 drmeister This isn't an easy question and I don't expect easy answers. I'm kind of flailing for other perspectives.

22:36:46 drmeister He has a paper on the fast generic function method and another on bootstrapping . It's not clear how they are related and I've discussed this with him. I don't have a solution yet.

22:37:20 drmeister What really puzzles me is that ECL avoids this problem somehow.

22:37:56 drmeister I'm thinking maybe I'm doing something in the wrong order, which makes my code sensitive to the problem. (sigh)

22:38:10 phoe Most likely, now that I think of it.

22:38:28 phoe ECL has two dispatches? It bootstraps one from the other?

22:39:10 drmeister When the fastgf dispatch function detects a dispatch miss it calculates an effective method, updates the call-history, compiles a new dispatch function and calls the effective method and returns.

22:39:44 drmeister No, ECL has a hash table cache. I mimic their cache.

22:40:36 drmeister I built beach's method on top of Clasp and I can switch individual generic functions to fastgf any time.

22:42:01 drmeister It wasn't easy and it took weeks of work. Now I'm ready to switch over to it - but this metastability issue that I don't understand is keeping me from doing that.

22:43:55 drmeister The idea is during startup I would switch over from the old to the new one and from then on all new generic functions would use the fastgf code. I can switch over but as soon as I evaluate anything interesting it goes into an infinite recursive loop and blows the stack.

22:45:40 pillton What about hard coding some of the generic functions first then switching them over to generic functions?

22:48:20 drmeister I'm thinking I effectively have that already. Ecl does that and I get call histories from that startup time

22:53:29 pillton What is the infinite loop trying to do?

22:54:19 pillton mop compute-effective-method

22:54:19 specbot http://metamodular.com/CLOS-MOP/compute-effective-method.html

22:55:57 pillton Is the new generic function sharing meta classes of the old one?

22:57:11 pillton It has been a while since I read the fastgf paper. From what I remember it requires changing internals of standard-object.

23:29:17 drmeister Afk for a bit. Dinner

0:36:19 learning_ is there a goto library for oauth2?

0:36:43 learning_ google gives me several

2:02:04 _main_ ** NICK __main__

2:06:46 drmeister Hello

3:18:19 beach Good morning everyone!