freenode/#lisp - IRC Chatlog
Search
6:14:29
smokeink
how to muffle such notes? note: doing unsigned word to integer coercion (cost 20) to "<return value>"
7:14:50
White_Flame
I have a macro that generates `(locally (declare ....) ,@body) so I can easily wrap it around small operations
7:15:43
White_Flame
(declare (sb-ext:unmuffle-conditions optimization-note)) around the parameters of the subform also allows its whinings to not be muffled, just the singular form in question
8:41:49
trittweiler
White_Flame, "global binding" versus "thread-local binding" (non-standard, of course) versus "local binding" is decent terminology I would reckon
8:44:48
jackdaniel
isn't the terminology: deep binding and shallow binding (and a symbol value which is not a binding)?
8:46:14
jackdaniel
and a consequence of deep binding implementation technique are thread-local bindings, and shallow gives you bindigns which are shared across threads (n.b probably harder to synchronize )
8:51:17
beach
jackdaniel: The distinction between deep and shallow binding is just an implementation issue.
8:52:34
beach
jackdaniel: Correct implementation of shallow binding in a multi-thread context is that there must be a per-thread "value cell".
8:54:06
beach
From memory, SBCL assigns a number to a symbol as soon as it is used as a variable. This number is used as an index into a table in the reified thread.
8:54:12
jackdaniel
hm, thanks for correcting me. I'll re-read the relevant parts of LiSP then to improve my understanding
9:07:35
jackdaniel
sorry, I can't tell from top of my head (and I'm focused on McCLIM issue atm so I'm not eager to look up ECL's internals code)
9:09:07
jackdaniel
I think that the location is on the stack and the array element is updated to point at that location
9:10:49
beach
But I wasn't asking what happens when a binding is done or undone. Just what SYMBOL-VALUE does. Again, if you are busy, you don't have to answer.
9:15:57
beach
For what it's worth, my current hypothesis is to use deep binding for SICL. I.e., the dynamic environment stack contains binding entries, and there is no per-thread table. SYMBOL-VALUE then has to search the stack to find the most recent binding.
9:16:50
beach
So searching for the current value or setting that value is more expensive than with shallow binding, at least if the stack is somewhat deep. Though finding the table in the thread is not free either.
9:17:30
beach
On the other hand, binding/unbinding is cheaper in deep binding than in shallow binding.
9:19:00
beach
If it turns out that 1. Access to special variables is frequent, and 2. Those accesses are often to deeply nested bindings, then I may have to reconsider.
9:28:19
White_Flame
beach: in SBCL, the TLS is always in a register, so "finding the table in the thread" actually is free
9:30:46
LdBeth
currying is just a function (defun currying (fn) (lambda (x) (lambda (y) (funcall fn x y))))
9:36:19
beach
It is *so* inspiring to answer questions from someone who 1. does not acknowledge this fact, and 2. leaves instead.
9:42:01
phoe
beach: the channel is logged, so if they ever come again and/or ask again, you can point them towards the logs in the channel topic.
9:43:18
jackdaniel
"- hey, I remember your nick from 5months ago, you did not thank me - do you remember that? - no (leaves the channel)", pretty burdensome strategy
9:44:04
jackdaniel
and I doubt that would change the culture of a person who is reminded that they did not follow proper netiquette
9:45:44
phoe
I thought of something like more like "oh I happen to recognize that nickname, they asked a question that was answered but they left, the answer is in the channel logs"
10:40:58
ghard
Have asked this before, methinks, but anyone tried to use ZS3 with DigitalOcean with success (claims to be S3-API-compatible) ?
11:04:13
no-defun-allowed
Note that it won't work for arrays of arbitrary dimensions, but only for one-dimensional vectors.
12:21:18
pjb
selwyn: (let ((array #2A((1 2 3) (4 5 6)))) (coerce (make-array (reduce (function *) (array-dimensions array)) :displaced-to array) 'list)) #| --> (1 2 3 4 5 6) |#
12:32:33
_death
pjb: just the other day I wrote this kind of flatten-array operator.. useful if you're representing bayesian network conditional probability as multi-dimensional arrays
13:07:28
reepca
for example: (defparameter test-array (make-array 10 :adjustable t :fill-pointer 0 :element-type '(unsigned-byte 8)))
13:07:51
reepca
(let ((*print-readbly* t)) (print test-array)) => #A((10) (UNSIGNED-BYTE 8) 0 0 0 0 0 0 0 0 0 0)
13:13:44
pjb
reepca: but you cannot use print, because it is not specified whether print uses print-object or not, and you cannot define print-object methods on standard classes such as array.
13:16:03
bitmapper
debugger invoked on a COMMON-LISP:UNDEFINED-FUNCTION in thread #<THREAD "main thread" RUNNING {10005084C3}>: The function SB-C::%MORE-ARG-CONTEXT is undefined.
13:19:11
boeg
I am trying to build a piece of software (the next browser) but it seems a common lisp error happens and it errors out. I get the error message thats put into the Makefile to make sure i have the xclip binary installed as well as developer files for sqlite and fixposix which I have in /usr/bin and in /usr/lib and /lib. I'm not sure if it's not able to find one or all of them or if something else goes wrong. Would anyone mind
13:19:11
boeg
look at the output from make and see if they can figure out what is wrong? It's here: http://ix.io/27sw
13:24:12
galdor
the error is The value "The root of all modes." is not of type LIST also called "do not use crappy stuff such as cl-annot which is dead, repo archived and unmaintained (and useless to start with)" or "test your code"
13:51:32
seok
if the complexity of gethash is O(1), isn't it better than arrays or lists in all cases for large dataset?
13:53:45
seok
phoe: so would that mean arrays are better in smaller sizes but hash better in larger ones?
13:54:09
Bike
you're not always using them for the same things, so comparing them may not be sensible
14:01:33
Nilby
Amortized O(1) + C is slower than exact O(1) + 0, especially when C is related to (length key).
14:15:04
pfdietz
Had a case recently where using string-case was much faster than intern, on a set of common symbol names.
14:15:06
pjb
seok: and here, you have integers as keys… Any other object would take a lot more time to hash.
14:15:55
pjb
pfdietz: you can usually exclude a case on 1 char=, and there are a finite number of cases, so you can select the right branch in basically O(1).
14:16:49
pfdietz
And you can play games with multiple character comparisons using (logior (logxor ...) (logxor ...) ...), which string-case does.
14:30:33
Nilby
Thinking about readably printing arrays made me try this experiement. Is there anything I missed? https://termbin.com/82c2
14:35:33
Bike
if you have two arrays displaced to the same array and you have print-circle you can reconstruct their sharing
14:36:33
_death
Bike: but then you want an actual displacement reference, rather than printing the data again
14:36:45
Bike
also if it is displaced you don't need the data. course you might have to recursively total-array-whatever-print the underlying array
14:42:34
_death
in reepca case it was also strange because he has an octet vector.. which usually don't have fill pointers, and usually are printed/read as octets..
14:46:03
reepca
fill pointer + adjustable was because there isn't a non-blocking way to read more than one byte at a time in usocket
14:50:37
_death
for adjustable, usually you know (or arrange to know) the size before-hand... for fill pointer, it makes sense, but for octet vectors I would consider having it separate so that you have a simple-array
14:51:28
boeg
How does it work when you have function with an argument list like `(defun aname (a (&rest argv) &body body) ()`? Doesn't &rest and &body kinda do the same?
14:53:15
Bike
it means it takes at least two arguments. the second argument is a list and fills argv, while any further arguments fill body.
14:56:45
p_l
I think it would look in use somewhat like (aname first-arg (argv list goes here) body...) ?
14:57:19
pjb
Nilby: note that if there's a displacement, you will want to have *print-circle* set to t, and deal with references.
15:01:25
pjb
Also, it looks like adjusting a displaced array to another displaced array so that there is a circle leads to an infinite loop.
15:05:03
pjb
Nilby: The problem is if you want to read back the printed-readably structure. If you just want to describe the structure, you can give all the information (a copy of it). But if you want to load back, you'd have to identify existing objects that are referenced. You get the ORM or OORM problem, with caching, identification of objects, etc…
15:21:57
pjb
Well, I'm not sure if it's completely specified. But basically, it assumes the current readtable. So if you print a symbol, you might not be able to read it in a different readtable, because it's not printed fully qualified and cased |FOO|:|BAR| would be the most readably. However, since again it depends on the readtable, it depends on the reader macro that could exist on #\|.
15:22:21
pjb
So something printed readably is not really stand alone. You must specify along the readtable.
15:26:40
_death
with-standard-io-syntax binds *package* and *readtable* .. though that wouldn't solve all issues of course
15:33:48
pfdietz
I would like to see similar tricks used for intern itself, at least for standardized packages or packages that import from them. Perhaps even dynamic code generation there.
15:38:00
pfdietz
The special case being keywords are printed with : even if the current package is the keyword package.
15:44:02
pfdietz
I now am trying to remember if symbols in other packages can be imported into KEYWORD.
15:49:32
pfdietz
Does importing a symbol into a package count as "interning" it? From glossary: intern v.t. 1. (a string in a package) to look up the string in the package, returning either a symbol with that name which was already accessible in the package or a newly created internal symbol of the package with that name.
15:50:06
pfdietz
But this does not happen in SBCL when a symbol belonging to another package is imported into the keyword package
15:50:27
pjb
pfdietz: yes, in one case: "If any symbol to be imported has no home package (i.e., (symbol-package symbol) => nil), import sets the home package of the symbol to package."
15:51:37
pjb
This is why the homeless symbol imported into keyword becomes a symbol. (symbol-value (find-symbol "FOO21312" "KEYWORD")) #| --> :foo21312 |#
15:51:40
pfdietz
The standard makes a distinction between being interned in a package, and having a package as the symbol's home package.
15:53:55
pfdietz
No. Simply importing a symbol into a package counts as interning it in that package.
15:56:53
pjb
And when those homeless symbols are imported into keyword, they must become symbols (bound to themselve, become constant variables, exported from keyword). If the implementation doesn't do that in this case, it's a conformity bug.
16:00:09
pfdietz
What should happen is that the symbol becomes a keyword. For example, it should become a constant that evaluates to itself.
16:04:46
Nilby
Sadly, this is why my sbcl version will always come up as dirty: https://termbin.com/2i8d
16:05:01
Xach
pfdietz: i could see an argument that the magic of intern happens in the "or a newly created" branch
16:05:47
pfdietz
But "intern" as a verb includes both cases. So I don't see how that inrepretation is consistent with the standard.
16:11:42
Xach
pfdietz: intern is approximately (or (find-symbol name package) (let ((sym (make-symbol name))) (import sym package) sym))))) - you're arguing that the find-symbol step should also include the work of making the symbol constant, self-valued, and external, if it isn't already?
16:15:02
pfdietz
Regardless, SBCL fails to do the special keyword processing even if one subsequently calls intern to get that symbol.
16:16:04
Xach
it's kind of regardful, because to get that behavior you'd need to do a lot more work in the find-symbol branch
16:19:09
pfdietz
"unintern removes symbol from package. If symbol is present in package, it is removed from package and also from package's shadowing symbols list if it is present there."
16:19:39
pfdietz
This implies being present in a package means the symbol is interned in the package.
16:20:47
pfdietz
As opposed to being accessible merely by means of inheritance from another package via use-package.
16:22:21
pfdietz
So there's an inconsistency here between intern and unintern. A symbol could continue be returned by intern even if unintern had been called.
16:24:18
pfdietz
In any event, even if we take the more restricted defn of interning, that it doesn't include accessible but not present, there's a bug here.
16:27:40
Nilby
I'm thankful they could agree on the flawed package system so at least we're not stuck with problems like Elisp.
16:38:29
Nilby
Another example you've probably considered is implementing something like Emacs's buffer local variables in the CL package system.
16:42:54
beach
It appears to be a widespread sport to find flaws in the standard, and to wish for the standard to be updated. However, most of the suggestions are made by people (not targeting you Nilby) who know nothing about language design, so they do not know the consequences that their suggestions might have. And they certainly don't know the constraints in the form of historical decisions and time pressure the committee had to deal with.
16:44:59
aeth
As far as finding flaws in the package system (or anywhere), I'd look at hacks that people do to work around the lack of features which they wish they had. For packages, I think what's missing is hierarchical packages because the foo/bar naming idiom is so common. This would not be easy to design and add, though.
16:46:11
Nilby
beach: I agree. Every *read-X*, *print-x* variable has consequeces, problems, and trade offs that are difficult to understand.
16:55:05
pjb
Nilby: HOWEVER, those variables, the printer and the reader, must be understood as development tools, to help debugging (and basically implement the REPL, the debugger, and inspect). They are not industry streight tools to use in your applications. For your application, it is understood that you will have to implement your own I/O and validation functions!
16:56:25
beach
aeth: So for example, the existence of all those people who wish that SYMBOL-VALUE, EVAL, etc. had access to lexical variables would indicate a flaw in the standard. And we should fix it, thereby making compilation of Common Lisp essentially impossible.
16:58:45
beach
All I am saying is that there is often more to it than just lamenting the lack of a particular feature.
17:02:28
ralt
I often have this snippet in my projects for package aliases... https://www.irccloud.com/pastebin/yjVAmMZI/
17:05:14
beach
Package-local nicknames solve a real problem. But that does not imply that the standard is flawed.
17:07:22
beach
ralt: Exactly. And that is particularly clear when you notice that people use languages on a daily basis that don't even HAVE a standard.
17:08:36
ralt
something that _is_ a bit odd with Lisp, is that as opposed to many languages, anything can be outside of the standard. In many languages, if it's not supported by the language, you're SOL, so there is often a strong argument to put that in the main place. Not so for Lisp.
17:10:19
Xach
pfdietz: if you define an "interned symbol" as one that was created through the intern process I think you can rationalize the current behavior.
17:12:06
beach
_death: I must be tired (after a long day), but I don't know what joke you are referring to.
17:14:37
_death
beach: it's US-centric joke.. in the US "chapter 11" means bankcruptcy.. the CLHS Packages chapter is chapter 11.. during the package flamewars associations were made
17:17:26
beach
So anyway, there are very few things in the standard that prevent me from doing my work, and those that do are easy to get around. At the same time, there is a lot of work to be done with what the standard provides.
17:39:37
pfdietz
There are problems with the standard, but you have to have some experience to find the real ones.
17:42:29
pfdietz
Another large class of problems are things that are problems, but not significant ones. Edge cases that language lawyers enjoy but that have little practical importance.
17:46:57
pfdietz
If I had to choose between "fixing/expanding the standard" and "fixing/expanding available libraries", I go with the latter in a heartbeat. Admittedly, the boundary is a bit blurry.
17:52:56
galdor
do you consider introducing packages that all implementations implement the same way "expanding the standard" ?
17:53:44
beach
No, that's just "widely agreed-upon libraries". There are plenty of those, and they are good.
17:53:48
galdor
if you take bordeaux threads, I find it curious to end up with a library instead of having implementations follow the bordeaux thread pseudo standard, which would allow every one to use, e.g. threads:make-thread without having to import a library
17:55:14
beach
galdor: It is much easier to use a library than to convince all the maintainers of all the implementations to add the same thing.
17:55:54
Shinmera
not to mention convincing them to possibly change an interface if they already have users for it.
17:56:21
pfdietz
That's the blurry part. Are de facto standard libraries part of the standard? I'll go with whatever answer makes things easier.
17:56:53
pfdietz
Having a better social process for creating standard libraries, now that's interesting and useful.
17:56:57
Shinmera
galdor: I'm sure people would prefer it if all implementations offered all of the possible features.
17:57:36
pfdietz
Having better use of standard libraries (for example, encourage explicit import rather than :use) would be valuable.
17:58:10
galdor
in practice, the thing is that the vast majority of implementations won't change anything because they are in low maintainance mode and won't ever evolve
17:58:50
galdor
but at some point, who cares if only 2 or 3 implementations do the work, others are dead anyway
17:59:49
Xach
the conditions for change vary but I don't think it's the case that they won't change anything.
17:59:51
pfdietz
They keep extending C and C++ even though each of those has large numbers of effectively dead compilers.