freenode/lisp - IRC Chatlog
Search
6:33:27
verisimilitude
VECTORs are SEQUENCEs and ARRAYs, but other ARRAYs aren't SEQUENCEs, yes, seok.
6:34:04
verisimilitude
Is that the essay about the FORTRAN programmer complaining about how slow Lisp was at the time?
6:34:22
verisimilitude
Yet, this programmer also didn't understand how to use Lisp was part of the issue?
6:37:16
jackdaniel
http://www.cs.umd.edu/~nau/cmsc421/norvig-lisp-style.pdf here, it is one of examples of "bad code"
6:39:55
verisimilitude
He also thinks Python is a Lisp, so I'm inclined to be skeptical of his opinions instead of blindly following them.
6:41:11
jackdaniel
either way, I've said what I've wanted to say - advising to represent arrays as lists is harmful
6:41:53
verisimilitude
I can't download this PDF; I'm inclined to believe this server is ignoring Tor connections without rejecting them; I believe I may have had this issue with this server before.
6:46:05
verisimilitude
Oh, I believe I've already been recommended this document and read it before, actually.
7:04:23
aeth
jackdaniel: not just in norvig's essays... it's also in the Worse is Better essay, except with matrices as the example
7:06:08
aeth
As for lists being "convenient"... This is Common Lisp. 90% of what you would want to do with vector-like lists you can also just do with vectors without modifying your code because they're generic for sequences.
7:06:32
jackdaniel
aeth: you are right, thanks for a remainder (I think I would quote it if I had remembered)
8:02:57
pjb
seok: accessing the nth element of a list is an operation that takes O(n) memory accesses (and increment and tests). Accessing the nth element of a vector is an operation that takes O(1) memory accesses (and one addition, and one test).
8:03:28
pjb
seok: but notice that accessing the 1st element takes about the same time; when n=1, O(n) = O(1) !
8:04:36
pjb
seok: on the other hand, extending the size of a vector is an operation that takes O(n) (with n=(length vector)), since we have to copy all the old elements of the vector to the new bigger vector. while extending the size of a list if done at the head, is an operation that takes O(1), since we only need to add a single cons cell.
8:05:03
pjb
ecraven: indeed. O(n) is meaningful only when you consider big and bigger data structures.
8:05:22
pjb
seok: so if you only have a handful of elements, it doesn't matter whether you use a vector or a list.
8:06:43
pjb
seok: now, adding elements in vectors can be done in O(1), using vectors with fill-pointers, at the cost of pre-allocating the maximum number of elements you may have to store.
8:07:34
seok
I am specifically looking to store game object data like coordinates, velocity and such
8:08:21
pjb
seok: there's also the case where you need to insert new elements in the middle of the sequence. Here, list can be O(1) if you already have a reference to the cell in the middle of the list, while with a vector you will have to displace the remaining elements, so it'll be O(n).
8:08:28
seok
Probably would reserve empty indices to add or remove object to avoid changing vector length
8:09:48
pjb
Well, there are sorting algorithm that work on list directly, but AFAIK, for big sequences, if will always be faster to convert to vector, soft the vector, and store back the sorted elements in the list. So I would guess that all sane implementation do that. (but I suspect one or two use a list sorting algorithm).
8:12:31
pjb
Also, I would expect to have to access objects geographically. There are more sophisticated data structures that would help there. For example quad-trees. (or octo-trees in the case of 3D).
8:12:55
pjb
Of course, you can implement quad-trees using vectors (I did that once for a big tree, that was mmap'ed from disk).
8:18:53
shka__
having said all that, it is usually good idea to start with vectors is you simply need a sequence of elements with random access
8:23:01
pjb
seok: that said, if you multiply the size of the vector by a constant > 1 (preferably closer to or bigger than 2) when you need more space, then inserting elements remain O(n) instead of O(n²). http://www.cs.cornell.edu/courses/cs3110/2011fa/supplemental/lec20-amortized/amortized.htm
8:24:38
aeth
If it's for a game, though, it's perfectly fine to have some fixed upper limit in size. Most do somewhere.
8:25:06
aeth
And once you have a limit might as well allocate it all because RAM's cheap and it's probably tiny compared to the textures.
8:29:23
pjb
Definitely. For games other considerations enter the scene. For example, pre-allocating arrays lighten the load for the garbage collector, which may be a good thing for real-time games.
8:46:23
pjb
But we're talking into the million elements. A vector of 1 million fixnum takes 8 MB (64-bit); a list of 1 million fixnum takes 16 MB.
9:02:07
phoe
it's a copypaste from pgloader - my biggest question is if I got licensing right on that one
9:02:13
shka__
for now, i just wanted to point out that alexandria already has some of the functions you implemented
9:02:51
shka__
and i think it is wise to use alexandria tools whenever possible instead of defining your own
9:03:24
phoe
shka__: file a pgloader issue then, https://github.com/dimitri/pgloader/blob/master/pgloader.asd#L19
9:10:16
aeth
pjb: Two additional issues other than just the memory size. (1) the list might not be together in memory and (2) if the large vector has an element-type that's not T it afaik greatly simplifies the GC process
9:15:41
aeth
shka__: the list won't just be slower... I think (hope? I don't see why not) that the GC scanning the vector would be O(1) if the vector can only hold something like single-float or (unsigned-byte 8)
9:16:19
aeth
Use a 2D array instead of a vector-of-vectors and you could see some major GC improvements if that's the case, assuming there's an element-type that works.
9:26:47
phoe
as in (let ((solution (make-generational-gc))) (declare (considerable-extent solution)) ...)?
9:43:21
dimpase_
it seems that the only way to use ASDF if the compiler cache and the package you build is by doing (asdf:disable-output-translations)
9:47:15
jackdaniel
i.e :move-here "/opt/foo/" where /opt/foo is on /dev/sda1 and /home/dimpase/.cache/ is on /dev/sda2
9:48:00
jackdaniel
that is a copy operation, so it is not very suprising you can't "move" there (however inconvenient)
10:23:45
dimpase_
Unable to rename file #P"/home/dima/.cache/common-lisp/ecl-16.1.2-unknown-linux-x64/mnt/opt/Sage/sage-dev/local/var/tmp/sage/build/kenzo-1.1.7-r1/src/kenzo--all-systems.fasb" to #P"/mnt/opt/Sage/sage-dev/local/var/tmp/sage/build/kenzo-1.1.7-r1/src/kenzo--all-systems.fasb".
10:36:12
jackdaniel
well, not really. I'm ECL maintainer so I just happen to know these parts of asdf and ecl.
13:07:38
cryptopsy
how to make a function that takes a line and returns a local variable which is a number of the occurance a character occurs in that line?
13:08:01
cryptopsy
seems wasteful for it to run recursive, so i opted to use a (let ((indent 0)) ...
13:17:05
cryptopsy
i am trying to call the function that returns the line in while line do but i am not sucessfull
13:20:34
cryptopsy
i think i might be in the wrong dir and the file it is supposed to open is not there
13:21:55
cryptopsy
i used to have a zone of dead pixels where i am keeping the sbcl window, but even after changing the screen my mind is still recognizing that zone as dead pixels and doesn't immediatly focus on the text in that corner
13:23:42
cryptopsy
i am a gentoo user, i have to set app-editors/vim python_single_target_python2_7 , this seems weird to me because it is set in make.conf
13:25:28
grewal
From the turorial, "You will also need a Python enabled Vim, and the same Python version installed that is Vim compiled against."
13:26:46
cryptopsy
perhaps because it is built with PYTHON_SINGLE_TARGETS python3_6 but slimv wants python2_7 on itself
13:28:21
grewal
vlime doesn't require python. But if you want paredit, you'll have to install that separately.
13:31:55
cryptopsy
it has been my dream this last decade to do that and i only have 3 days it until i start a career that won't permit me to have this hobby anymore
13:36:19
cryptopsy
what is the data-type of lines from (loop for line= (read-line stream nil) , i think it is a string?
13:44:35
gilberth
cryptopsy: What should the function do anyway? Counting the leading #\Space characters?
14:00:55
cryptopsy
i am aware there are many ways to do it, i am doing this as a learning exercise with primitives i understand
14:20:51
cryptopsy
i am attempting to write this function that takes a string but returns the value of a local variable
14:22:06
cryptopsy
would this be better to have beeen written as a recursive function considering the indents are almost always less than 20 ?
14:39:00
jurov
Can anyone please enlighten me why in my recursive function these two forms behave differently: (concatenate 'list '( [ ) (recurse x) '( ] ) ) and `( [ ,@(recurse x) ] )
14:41:40
Bike
cryptopsy: they may be garbage collected at any point at which they are no longer reachable.
14:43:43
cryptopsy
or are all the exercises virtually the same but in the lisp lang instead of scheme
14:44:21
cryptopsy
Bike: if the function ends, and is never used again, does it still have allocated that local var somewhere
14:49:06
sjl_
I think it's pretty uncommon for a garbage collected language to define when garbage collection of local variables occurs.
14:49:14
cryptopsy
so i am reading here: """ While Garbage Collection (GC) was one of the major innovations of Lisp, there have been many attempts to get rid of it in one way or another. """
14:50:36
sjl_
Tell Bike you have a local variable pointing at a 10gb array and you want to know when it's safe to open Slack again without the OOM killer killing your program or something, if you want to get an answer.
14:50:52
beach
cryptopsy: No, it is not interesting. You really don't want to know. The system will run the garbage collector when it sees fit.
14:51:10
Bike
Sometimes, for optimization reasons, it is important to care about what the garbage collector is doing.
14:51:28
beach
cryptopsy: Manual memory management is one of the main sources of bugs. We are lucky not to have to do that.
14:53:47
sjl_
Out of curiosity: does the spec even require garbage collection? Could a CL implementation just allocate until OOM and still satisfy the spec?
14:54:22
cryptopsy
does clasp compile to llvm ir directly or does it produce intermediate C or C++ ?
14:57:49
sjl_
https://groups.google.com/forum/message/raw?msg=comp.lang.ada/E9bNCvDQ12k/1tezW24ZxdAJ
15:02:38
grewal
cryptopsy: Some lisp implementations expose parts of the garbage collector. See, e.g. http://www.sbcl.org/manual/#Garbage-Collection
15:28:01
cryptopsy
beach: i saw a presentation a chemist made where he implemented his own lisp to solve molecular problems
15:35:32
verisimilitude
An implementation is supposed to signal STORAGE-CONDITION in cases such as memory exhuastion, sjl.
15:44:29
phoe
"In general, the entire question of how storage allocation is done is implementation-dependent, and so any operation might signal storage-condition at any time."
16:05:59
pjb
cryptopsy: Movitz https://common-lisp.net/project/movitz/movitz.html is a CL implementation that runs currently on Muerte, a x86 run-time environment that does not provide a garbage collector. (Movitz is designed to implement operating systems).
16:42:00
beach
housel: There are rumors that he wants to create a new version. And now there is also Mezzano.
16:54:15
beach
cryptopsy: Common Lisp is defined so that the READ function takes a sequence of characters and produces a Common Lisp data structure, which is then further processed by a compiler or an interpreter.
16:55:12
beach
Some people call it that. I don't. It's a Common Lisp data structure made up of lists and atoms.
16:55:42
beach
cryptopsy: Because, as I said, the language is defined to have a standard READ function.
16:56:22
cryptopsy
well it can read the terms and create the linked structure without translating to tokens and then build an AST from a tokenized string
16:58:39
beach
cryptopsy: I am not sure whether you are asking "Why does this particular implementation work like that?", or "Why was the language defined that way?".
17:21:14
verisimilitude
By having READ available by itself, you can easily, say, read in data that has previously been written, cryptopsy. With Common Lisp, you can delay defining a format for serialized data beyond READ and WRITE indefinitely.
17:30:17
phoe
sure, but there is a level when you are somewhat competent with its concepts and workings
17:30:39
dlowe
cryptopsy: okay. We have many, many people who pop in and want to know, abstractly, all about lisp without ever wanting to write a line of code
17:47:12
francogrex
(r% "glm" :formula "outcome ~ predictor" :family "poisson" :offset (r% "as.matrix" (r% "log" (r% "subset" sub :select "doses"))) :data sub)
17:51:39
cryptopsy
how do i disable the header when sbcl runs " This is SBCL 1.5.0, an implementation of ANSI Common Lisp. ...."
17:53:01
francogrex
and it's not really correct what you say dlowe because: (defparameter x (values))
17:54:02
francogrex
while (defparameter xx (r% "glm" :formula "outcome ~ predictor" :family "poisson" :offset (r% "as.matrix" (r% "log" (r% "subset" sub :select "doses"))) :data sub))
17:55:07
cryptopsy
how do i print lines with write when doing (loop for line = (read-line stream nil) while line do ( ...
17:58:36
dlowe
francogrex: the only thing I can think of is that r% is generating a condition that's being silently swallowed by the repl
17:59:50
sjl_
2019-01-11 12:44:23 <scymtym> if the implementation's REPL establishes an ABORT restart, the behavior could be caused by code like (multiple-value-list (handler-bind ((error #'abort)) (/ 2 0)))
18:02:57
francogrex
like the easy one that does not require changing the code? I guess it would be to
18:03:08
sjl_
Depends on what you're actually trying to do. Are you trying to detect when this happens?
18:08:14
dlowe
you can also (declaim (sb-ext:muffle-conditions style-warning)) at the top of the file
18:08:16
sjl_
Ideally you'd fix the code to not emit warnings. To hide the warnings, use the solution found in the first google result for "sbcl disable warnings"
18:09:09
dlowe
declarations about variables occur at the beginning of the body of the variable's scope
18:13:44
cryptopsy
should i use *standard-input* or *terminal-io* nil EOF for runnning my lisp program with a parameter which is the filename it is supposed to read
18:15:31
pjb
*standard-input* is for batch input. *query-io* is for interactive I/O. *terminal-io* is to communicate with the operator (and it cannot be rebound).
18:16:18
pjb
If you open a file you can rebind *standard-input*: (with-open-file (*standard-input* pathname) (read))