IRC channel logs

<apapsch>Hi, what is the recommended way to copy from port to port? I have a input port from (pipeline) result and want to write to a file. Currently I (get-bytevector-n!) in 8KB chunks and then (put-bytevector) to the output port, though it seems rather verbose

<leoprikler>that is indeed what you're supposed to do unless you somehow want line-based stuff

<leoprikler>see the Guix download logic as well ;)

<apapsch>alright, thanks :-) i think i'll go with get-bytevector-n without the !, seems a bit easier to use. guile docs don't mention that it returns an eof-object, while the r6rs doc says so. is it a docs issue or does the impl differ from r6rs?

<dsmith-work>apapsch: If you are willing to go linux-specific, sendfile and copy_file_range move data around without copying to userland.

<dsmith-work> https://man7.org/linux/man-pages/man2/sendfile.2.html

<dsmith-work> https://man7.org/linux/man-pages/man2/copy_file_range.2.html

<apapsch>dsmith-work: thanks, but that's going even deeper into the weeds :-) i'll make a mental note for the optimization on a rainy day

<dsmith-work>Guile does have sendfile. Often used in web servers. copy_file_range is new to me. Looks very interesting.

<apapsch>indeed interesting. though it requires a count parameter, which I can't know beforehand (the input port returns the xz output from the pipeline)

<wingo>miscompilation bug has me delving in the guts of "match" :/

<manumanumanu>haha. how much syntax rules can you take?

<wingo>omg

<wingo>(use-modules (srfi srfi-9) (ice-9 match))

<wingo>(define-record-type <pair>

<wingo> (make-pair head tail)

<wingo> pair?

<wingo> (head pair-head)

<wingo> (tail pair-tail))

<wingo>(define (test x)

<wingo> (match x

<wingo> (($ <pair> (x ...) (x ...))

<wingo> (pk x))))

<wingo>what do you think (test (make-pair '(1 2 3) '(1 2 3 4))) does

<wingo>aaaaahhh!!!

<manumanumanu>no guess.

<wingo>ok this one is simpler

<wingo>(use-modules (ice-9 match))

<wingo>(define (test x)

<wingo> (match x

<wingo> (((x ...) (x ...))

<wingo> (pk x))))

<manumanumanu>I think I know what it SHOULD do.

*dsmith-work really needs to lean match

<sneek>Yes please

<wingo>run as (test '((1 2 3) (1 2 3 4)))

<manumanumanu>oh yes

<wingo>am i misunderstanding something????

<wingo>i thought the two "x"'s in the pattern would unify

<manumanumanu>yup.

<manumanumanu>that shouldn't match?

<wingo>consider (match-lambda ((x . x) x) (_ #f)). returns false unless a pair and car and cdr are equal?

<wingo>but (match-lambda (((x ...) (x ...)) x) (_ #f)) -- should return false unless a 2-element list and first and second elements are the same list

<wingo>but not the case apparently??

<manumanumanu>wingo: (chibi match) does the same

<manumanumanu>Maybe I should continue my work to port Trivia from CL...

<wingo>does the same as guile you mean?

<manumanumanu>racket match fails: (match x ((list (list x ...) (list x ...)) x))

<manumanumanu>yes

<manumanumanu>and the code is the same.

<manumanumanu>I would guess

<wingo>right i think we use foof's match so it would make sense that chibi & guile's behavior has the same bug

<manumanumanu>That kind of syntax-rules is no fun to debug. I know. I wrote this: https://git.sr.ht/~bjoli/goof-loop/tree/master/item/goof-impl.scm#L120

<manumanumanu>ocaml does the correct thing as well.

<wingo>this causes a miscompilation, as you might well imagine :/

<manumanumanu>I thought you thought this was a result of a miscompilation, which is why I compared to chibi.

<wingo>seems to be an artifact of the optimizations for (p ...) -- match.upstream.scm:512

<wingo>because ((x . _) (x . _)) does correctly unify x

<manumanumanu>foof hangs in #scheme, btw

<wingo>manumanumanu: because the compiler uses match, it actually causes a miscompilation :)

<manumanumanu>That is better than "We have a miscompilation. it manifests itself in (ice-9 match)", though...

<wingo>marginally :P

IRC channel logs

2021-06-16.log