libera/#commonlisp - IRC Chatlog

14:34:08 dnhester26 dlowe: pardon my ignorance for I have never used do. what is the last statement doing there? when is it called? it's after the test form, form the spec I don't understand

14:35:54 dnhester26 is it evaluating only if the test-form fails? That's what it looks like to me, but from the spec I couldn't understand

14:38:09 thuna` dnhester26: That DO evaluates its body - which is the RESTART-CASE - until the result of (ALL-START-WITH-SAME-LETTER ICE-CREAM SAUCE TOPPING) is non-nil.

14:38:22 dnhester26 In this line in the spec "At the beginning of each iteration, after processing the variables, the end-test-form is evaluated. If the result is false, execution proceeds with the body of the do (or do*) form. If the result is true, the result-forms are evaluated in order as an implicit progn, and then do or do* returns." I don't see in the sytnax neither the body of the do nor the result forms

14:38:36 dnhester26 thuna`: thanks

14:59:38 chkhd ** NICK chkhd`

16:37:14 hayley jmercouris: Consider writing --quiet with two hyphens before you smartass about wHaTeVeR rEaSoN.

17:09:54 edgar-rft_ ** NICK edgar-rft

20:28:57 NotThatRPG Is there any way to assert types and get compile-time type-checking for generic functions in SBCL?

21:18:32 aeth maybe one of the ways to inline generics?

23:19:06 dbotton wrote some quick notes on mixing packages and clos classes, am I missing anything

23:19:07 dbotton https://plaster.tymoon.eu/view/4523#4523

0:46:38 thuna``` ** NICK thuna`

0:47:41 paulapatience dbotton: Maybe mention the use of % prefix for slots so that they cannot be accessible publicly via slot-value if the accessor shares the same name and is itself exported.

0:48:25 dbotton good point

0:48:32 paulapatience It's not used by everyone, but the convention exists

0:48:54 dbotton right I would mention it for sure

0:49:20 dbotton need to write it up better and the other notes been making

0:50:48 paulapatience Also, mention that one convention for accessor naming is that they should not be prefixed with the class name, because it is awkward when you have differently named subclasses (there are also other reasons). This is actually where package namespacing comes in handy, in case you want the same name for a generic function but a different lambda list.

0:51:36 paulapatience Again, that convention is not universally followed

0:52:57 aeth the class prefix is a struct convention (though the struct can configure the conc-name)

0:53:24 aeth (including setting it to nothing)

0:53:34 paulapatience Another thing I learned from yitzi recently is that you can make a generic predicate to tell if a specific class implements a certain protocol rather than having an abstract superclass

0:53:42 aeth so if I see a class prefix, I assume it was refactored from a struct

0:55:08 dbotton ya sort of defeats packages and classes to include prefixes

0:55:25 aeth well, no, it defeats the purpose of generic

0:55:52 dbotton that too

0:56:01 pillton You shouldn't think of generic functions as being associated with a class. Functions perform actions. Classes encapsulate state.

0:57:12 dbotton paulapatience - generic predicate to tell if a specific class implements a certain protocol - just adding something to indicate you implemented it?

0:57:51 dbotton pillton: the GFs in this context more about interfaces

0:59:07 dbotton interface to what is the question - but I understand your point which is more not seeing CL with a C++ eye

0:59:36 pillton Sure. The functions have nothing to do with the class though. They are the interface to the actions that need doing. They can be generic, macros or normal functions.

1:02:19 dbotton well they act on them, and if you "model" things can see it as if do

1:02:43 dbotton some early books for sure presented CL more C++y

1:02:48 dbotton for classes

1:02:48 paulapatience dbotton: Suppose someone has an instance of a class and wants to know if they can call a protocol function. You have two options: You can do typep to check for all the types you know implement the protocol, or you call protocol-implemented-p (where presumably it is specialized on the appropriate classes to return T).

1:03:12 pillton One thing I like about lisps in general is that they have one way to perform an invocation. None of this f(x,y) versus x.f(y) or y.f(x) confusion.

1:03:13 dbotton ok so a programmer indicator

1:03:36 paulapatience The problem with typep is that unless all classes implementing the protocol have a common superclass, you can't know if a third-party implementation of the protocol does implement it

1:04:12 paulapatience yitzi knows more about this. He believes for example that it was a mistake in gray streams to export the class hierarchy.

1:04:27 dbotton So far I have just avoided trying to emulate java interfaces

1:04:33 paulapatience I have not yet used this pattern, but I think it will simplify the design of some of my in progress libraries

1:04:45 dbotton certainly could

1:05:12 paulapatience Ok, actually, another thing you could add to your document is the client pattern

1:05:35 paulapatience It is widely used in s-expressionists libraries, like Eclector

1:06:11 paulapatience Basically, you should only specialize a generic function if you own it, or if you own one of the types you are specializing

1:06:24 paulapatience Otherwise they can get overridden accidentally

1:06:26 ldb interface is a thing to work with static typing

1:06:34 yitzi To clarify more, I believe that exporting the classes in Gray streams should have been done as "mixins." There is already generic STREAMP, INPUT-STREAM-P, etc. So having the classes that as implied "you have use these" is confusing. And there are some minor contradictions as a result.

1:07:36 ldb in dynmaic typing it always possible to apply a generic function until it signals error at runtime or something

1:07:37 dbotton make sense

1:07:49 paulapatience The client is basically a (sometimes empty) class which users define to not break that rule

1:08:23 paulapatience Without having to possibly subclass a bunch of other classes

1:08:41 paulapatience Trucler is a good example of this pattern, and where I learned it from

1:08:55 paulapatience I think the Trucler manual has a small paragraph explaining it

1:10:38 dbotton will take a look

1:11:28 ldb client pattern unifies eql based dispatch

2:11:30 beach dbotton: You have to export names of slot readers/writers only if they are part of the protocol. A class can have some slots with private accessors.

2:12:05 beach dbotton: Similarly, there may be private generic functions that apply to a class, and you don't want to export their names.

2:12:34 beach dbotton: More generally, I guess, "exposing an object" does not make much sense.

2:13:20 beach dbotton: And the term "class method" does not make sense.

2:13:49 beach dbotton: And I don't understand the convention of using meta-names of packages that start with a colon.

2:15:09 beach dbotton: It is definitely not necessary to use :IMPORT-FROM, and recommend not using it. It is much better to use an explicit package prefix.

2:15:25 beach dbotton: This advice has been given over and over here.

2:17:45 aeth I suppose "class method" could be interpreted to mean "a method automatically generated by DEFCLASS"

2:17:50 beach dbotton: Why do you use uninterned symbols for package names, but not for the symbol names they export? And then you sometimes use keyword symbols for package names as well as in (IN-PACKAGE :TPAC).

2:17:58 aeth though these automatically generated methods do not do anything special or different, afaik

2:18:42 beach They usually become instances of STANDARD-READER-METHOD and STANDARD-WRITER-METHOD, but that's it.

2:20:45 beach dbotton: Instead of using a name such as SET-MYSLOT define a method on (SETF MYSLOT).