libera/#commonlisp - IRC Chatlog

5:56:31 beach Also, the remarks by dbotton prompted me to work on my book project on CLOS programming (from which the chapter on protocols is extracted), but I am not sure how long I will keep up the enthusiasm now that my favorite coauthor has developed other interests.

5:59:58 beach Maybe I should write the book openly so that people here can contribute. I'll give that some thought.

6:09:34 paulapatience beach: I would certainly read any draft and proofread.

6:10:36 beach Thanks! I know you are a good proofreader.

7:33:10 dbotton beach: thanks! Just up but will read and incorporate

7:45:22 dbotton Beach yes!!! I would love comments and advice on CLOG's code!

7:51:19 dbotton Your advice and comments priceless

8:26:44 younder beach How is the boot protocol coming along?

8:28:24 beach younder: Just finished phase 4. Thanks for asking.

8:28:46 beach dbotton: You you WHEN in a context where the value is needed. IF should be used then instead.

8:29:09 beach dbotton: And you have several violations of the rules stated on page 13 of the LUV slides by Norvig and Pitman.

8:30:16 beach dbotton: It also looks like you have significant code duplication.

8:32:08 beach dbotton: And since you have the same name for a slot and its reader, you suggest that it is OK to use SLOT-VALUE and WITH-SLOTS on it.

8:34:49 dbotton I will do a full walk through after your comments to spruce it up. It for sure needs it, was my first CL program

8:35:20 beach I see. You have a method on PRINT-OBJECT with an incorrectly indented body.

8:43:47 dbotton Beach one should not count on when returning nil?

8:44:31 beach You can count on it, but in order to signal the intent to the person reading your code, you should use WHEN and UNLESS only in a context where the value is not used.

8:44:45 beach Like in a PROGN before the last form, or in a TAGBODY.

8:44:49 dbotton Ok

8:45:28 dbotton When have laptop will look up the slides

8:47:52 beach See for instance: https://irclog.tymoon.eu/libera/%23commonlisp?around=1713974717#1713974717

8:47:52 dbotton I think i am ok with slot-value etc in the case of CLOG's use

8:48:01 beach You shouldn't be.

8:48:12 beach It prevents the use of auxiliary methods.

8:48:42 dbotton Ok. Good point

8:48:57 beach dbotton: Also, slots are implementation details. They may disappear as the result of maintenance.

8:49:27 beach dbotton: As my document on protocols indicates, all access should be though (typically generic) functions.

8:49:50 beach This point has also been iterated many times.

8:50:01 dbotton I very much appreciate that idea now

8:50:43 dbotton Four years ago didn't understand. Part of my recent comments

8:51:05 beach Like here: https://irclog.tymoon.eu/libera/%23commonlisp?around=1704383532#1704383532

8:51:31 dbotton It should not have taken me that long to understand the idea coming to CL and near no documentation for it

9:00:19 dbotton Beach I hope your CLOS book becomes a reality. There is a need for a well written practice as we understand it today books for idiomatic CL programming. I asked from start when learned of anything existed

9:09:01 beach The problem is that there is so much to do, and not enough time to do it.

9:12:51 dbotton I know the feeling :) for me it is also how much I am not the best person to do

9:42:18 pranav beach: Thanks for the time you take out for this chatroom, not least for clarifying terminology, which I believe helps me read and understand CLHS and CLTL2 better.

11:02:23 chkhd ** NICK chkhd`

11:30:20 beach pranav: Pleasure!

11:46:51 Stenotrophomonas ** NICK brokkoli_origin

14:58:10 mwnaylor In CLOS, is it possible to add a slot to an existing instance of a class?

14:58:42 beach No, the slot has to be added to the class.

14:59:03 dlowe the slot in existing instances will exist but be unbound

14:59:34 beach Oh, if there is an initform?

14:59:47 dlowe I can't remember for sure

15:01:03 beach mwnaylor: Making it possible to add slots to individual instances would make it much harder to implement slot access efficiently. Have you seen this possibility in any other language?

15:02:09 dlowe testing on sbcl and it honors the initform instead of having it be unbound

15:02:22 beach That's what I thought.

15:12:34 bike you might get this in a more free-wheeling prototype-based object system, like you get in javascript

15:13:00 beach And Self, right?

15:13:06 bike Yes

15:13:49 beach So it is not impossible to create an efficient such a system, but a bit trickier.

15:14:40 beach But I was interested in the reason for mwnaylor asking this question.

15:16:31 bike being used to javascript, maybe, is what i was thinking

15:16:57 beach I see.

15:19:40 beach Sometimes a person will give as a reason something they heard from a friend, meant to be an argument against Common Lisp, without the person having considered the reason himself or herself.

15:22:05 beach mwnaylor: The good news is that you can create such an object system in Common Lisp if you like. In fact, I think the GUI library (the name of which I forget) had such a system.

15:22:42 beach ... for CMUCL.

15:24:38 mwnaylor beach: I was was wondering if it would be simpler to patch onto an existing object rather than extending a class. FWIW, I'm still new to CLOS.

15:24:44 bike i thought one of the JSON libraries had some MOP business to match javascript objects, but i can't find it now

15:25:33 beach mwnaylor: It is a somewhat strange thing to want to do in terms of software engineering. What is the use case here?

15:27:45 NotThatRPG Is it possible to make a lisp file for sbcl that can be loaded into lisp OR used as a shebang script? Seems like LOAD won't like the shebang and bash won't like `;;; #!` but maybe there's a fix for this?

15:28:09 mwnaylor Messing w/ internals of stumpwm. Guess I'll be better served by learing proper CLOS and delving into the stumpwm source code.

15:28:46 bike NotThatRPG: i believe SBCL defines a shebang reader macro for exactly this purpose.

15:28:59 bike but that won't help with the semicolons

15:29:15 beach mwnaylor: You can always stick the associated object in a hash table with the original object as a key.

15:29:21 bike NotThatRPG: https://sbcl.org/manual/#Shebang-Scripts

15:29:22 NotThatRPG bike: I tried to load such a file (inside a script for buildapp), and got an error...

15:29:55 NotThatRPG I don't see an option for `sbcl --load hello.lisp"` in the manual there.

15:30:15 NotThatRPG Not a huge problem: easy to split into a "body" file and a shebang wrapper.

15:30:17 bike yeah, i think you have to use --script instead of --load

15:30:51 bike --script ignores the shebang, in addition to other stuff it does

15:30:57 bike but it still loads the file.

15:39:42 edgar-rft_ ** NICK edgar-rft

15:40:56 paulapatience NotThatRPG: https://github.com/Shinmera/wg-manager/blob/master/wg-manager.lisp

15:41:12 mwnaylor beach: Interesting implementation. Clojure has defrecord, which allows adding new elements w/o changing the nature of the original object. Lisps dialects tend to be flexible about modifying data structures.

15:41:24 paulapatience With a #||# comment you can have exec be the second line. By default executable scripts go through /bin/sh

15:42:24 NotThatRPG paulapatience: Thanks so much! didn't know that!

15:44:31 bike mwnaylor: does it not just define a new record type with the same name?

15:45:04 bike mwnaylor: in any case this doesn't seem like it does anything to individual objects, given that it defines a class

15:47:10 mwnaylor bike: no, same type. The modified instance will still be acceptable to any functions related to a protocol. If the type changed, it would not be accepted as a valid parameter.

15:47:55 bike so it does modify an instance somehow? it really doesn't look it does from the docs

15:48:04 bike plus it's clojure, which is all immutable stuff

15:49:41 mwnaylor It creates a new object that has the new element(s).

15:51:26 bike does it not create a new _class_ rather than a new object?

15:52:33 mwnaylor No, not a new class.

15:53:24 bike should i not be looking at https://clojuredocs.org/clojure.core/defrecord

15:53:33 bike because this is all pretty much saying new class to me

15:53:39 beach To me too.

15:54:36 bike or new record type, but that's not the point

15:56:08 beach "Dynamically generates compiled bytecode for class [sic] with the given name..."

15:59:37 mwnaylor As you can see, the types are the same. https://pastebin.com/67h4yHFX

16:01:00 bike ...okay, so it's _assoc_ that's effectively adding a field here.

16:01:34 bike assoc is creating an object.

16:03:52 bike but sure, so you can add a field to an object this way.

16:06:42 beach mwnaylor: This is not hard to accomplish in a programming language. The hard part is to make it efficient. If you are not careful, each access to a field would require a hash-table lookup.

16:07:56 bike it kind of looks like the object is a hash. i'm kind of curious how this is implemented. an actual java class with an extra field for a hash table of extra fields?

16:08:03 beach mwnaylor: I am convinced that the designers of CLOS avoided this features because of such additional cost. But, again, an object system can be designed in, and added to, Common Lisp that has this feature. I wish I could remember the system that already did that.

16:08:32 bike the docs mention an "extension field map" and a special field __extmap, maybe that's what those are

16:09:41 bike https://cl-json.common-lisp.dev/cl-json.html#PROPERTIES-OF-FLUID-OBJECTS aha, this is the json thing i was thinking of

16:09:42 beach The GUI system was named Garnet.

16:10:33 beach Oh, I guess the object system was named Garnet as well.

16:10:56 bike https://github.com/sharplispers/cl-json/blob/master/src/objects.lisp#L96-L115 here's where setf on a missing slot is made to add a slot

16:11:58 beach bike: So it's a hash-table lookup?

16:13:09 mwnaylor beach: Agreed (re: adding and making it efficient). Clojure has been designed to make this somewhat easy. That's why I was asking about something like an add-slot function that would modify an object.

16:13:32 bike beach: in clojure? i mean, maybe, but it's not stated outright

16:13:44 beach bike: I see.

16:14:29 beach mwnaylor: OK, and CLOS has avoided it in the name of efficiency. You would have to use a different object system like Garnet to do that in Common Lisp.

16:14:55 beach ... but since you are working with existing code, that might not be possible of course.

16:15:17 beach mwnaylor: Then just use a hash table as I suggested initially.

16:19:21 beach mwnaylor: Or just create a subclass with a single instance.

16:20:44 mwnaylor beach: I'll keep the hash table option in mind, though learning proper CLOS is probably the better option for *me*. Even with that option, would still have to explore the stumpwm source code. A lot of stuff in there that uses its own macros to define classes rather than simpler defclass inheritance.

16:21:07 beach Hmm, yes, I see.