Fire is step THREE! | github.com/perl6/toolchain-bikeshed | Channel logs: irclog.perlgeek.de/perl6-toolchain/today | useful prior art: metacpan.org/pod/CPAN::Meta::Spec
Set by moderator on 5 September 2016.
ugexe it is probably worth noting that uploading dists to the PSIXDISTS username makes guessing the auth more difficult for consumers 00:31
for example: indexing is not exactly needed to do basic things like parse the upload file path into dist name, version, and author
however without the author representing the actual author it means you have to download all My::Module:ver<whatever>:auth<*> variants and check each meta data 00:33
ultimately it would not matter once there is indexing, but having all those parts in the path would act as enough information for current tooling to figure it out 00:34
maybe they could be uploaded under P/PS/PSIXDISTS/<original auth>/My-Module-<version>.tar.gz 00:36
although even that would have problems since some people use stuff like `My Name <abc@xyz.net>` as their auth 00:38
or would could upload them under their <sha1>.tar.gz! 00:39
if i were to put my own version of GTK::Simple into the ecosystem for example it would probably cause problems for the PSIXDISTS username plan 00:41
unfortunately i have no suggestions for a solution other than using the file path to contain that information, and as i pointed out that has its own points of failure 00:43
modules.perl6.org also fails in this regard (Foo is listed twice in ecosystem/META.list but only shown once) 00:47
github.com/ugexe/Perl6-Foo I've added this version of Foo to the ecosystem for testing such things, which duplicates the distribution name but uses a different auth 01:16
01:48 ilbot3 joined
moderator Fire is step THREE! | github.com/perl6/toolchain-bikeshed | Channel logs: irclog.perlgeek.de/perl6-toolchain/today | useful prior art: metacpan.org/pod/CPAN::Meta::Spec
llfourn sjn: I'll take a look. Probably some change in rakudo internals has broken it. It relies heavily on nqp. Cheers. 04:28
sjn: I removed some bit rot. It installs now :) 05:20
06:10 domidumont joined 06:15 domidumont joined 07:43 leont joined
sjn llfourn: thanks! ^^ 08:59
llfourn sjn: nw. Enjoy. 10:44
11:55 stmuk_ joined 14:18 lizmat joined
ugexe nine: the meta caching bits can be cleaned up a bit into using single assignment if you think it looks better gist.github.com/ugexe/017672ef3972...d08a854872 14:19
15:09 perlpilot_ joined
nine ugexe: that looks like it's based on an older version, as it still uses %extensions<Perl6> instead of @extensions? 15:15
ugexe yeah, thats from a few months ago. should just be able to s/%extensions<Perl6>/@extensions/ 15:18
mostly just pointing out how to structure it to avoid reassigning to $found 15:19
15:58 domidumont joined 16:18 tadzik joined 17:15 DrForr joined
stmuk_ how is zef handling installs from www.cpan.org/authors/id/P/PS/PSIXDISTS/Perl6/? is there a command line switch? 17:44
--cpan is actually metacpan? 17:49
ugexe --cpan is jdv79's metacpan fork with perl6 support 17:51
stmuk_ ah yeah I just was looking at port 5001 on hack
ugexe PSIXDISTS cant be properly consumed until its known how the module name, ver, and auth are stored in the file path 17:52
for instance: there are 3 variants of Distribution Foo: Foo:auth<github:FROGGS>:ver<1.0>, Foo:auth<github:FROGGS>:ver<1.2>, and Foo:auth<github:ugexe>:ver<1.2> 17:54
stmuk_ sure
mst I would presume that, currently, you're looking at 17:57
auth<PSIXDISTS>, then name and ver extracted from tarball name 17:58
ugexe auth isn't where its currently coming from, its the initial origin 17:59
so when taking multiple ecosystems into account (current perl6 ecosystem + cpan) you know when its really the same distribution 18:00
stmuk_ much as I hate them I'm tempted to suggest the SHA1 hash :/ ugly urls though
ugexe :auth<my@email.com> is probably a better default for people to use
the sha1 doesnt give you anything to index off of. the default My-Module-$ver.tar.gz lets you match on My::Module:ver<1.*> with just a file listing 18:10
stmuk_ wouldn't the index use the .meta files? eg. www.cpan.org/authors/id/P/PS/PSIXDI...0.001.meta 18:14
mst well, the real answer here is "I don't think we really know yet, that's why PSIXDISTS exists to start experimenting" 18:15
perl5's meta, btw, is trialing an x_authority meta key
ugexe if you were creating some index elsewhere you would. but if its just built into a client then they'd have to download the meta for every distribution just to search
mst maybe perl6 could do similarly and then what stmuk says
ugexe hmm the re-versioning makes it impossible to work with another ecosystem 18:17
stmuk_ not a great long term solution but maybe fine for some proof of concept
ugexe thats 600+ http requests to update your module meta data 18:18
stmuk_ or another random port on hack! I think both zef and panda use different ones anyway 18:20
mst well, yes, having a dumb file store means you need an index as well
hence why metacpan6 is looking at building one 18:21
ugexe yeah, but if it could be encoded into the file name then you dont really need an index at the basic level
mst true.
that's why I was thinking PSIXDISTS should be considered the authority by default 18:22
maybe we need a filename or pathname standard for tarballs that encodes it?
ugexe name/ver/auth represent an immutable distribution though, which should be the same no matter where it comes from. auth is just some additional identifying bits 18:23
[Coke] what would rakudo's author be? rakudobug@perl.org ?
mst ugexe: right, and the cpan approach to that has historically been to consider it AUTH/Name-1.23.tar.gz 18:24
ugexe my auth could be `github:ugexe` on cpan or `cpan:ugexe` on a github distribution. you just use whatever identifies the original
right, which doesn't map with how perl6 is designed to handle this. yeah its just a start, but its not really usable 18:25
mst sure
this is why we made a start under a dedicated PAUSE id that doesn't actually matter 18:26
because it was bound to turn out to suck the first N attempts
ugexe like which version of Foo:ver<1.2> will win? they will both map to the same file name, despite having different auth
stmuk_ Net::HTTP:ver(v1.0):api<>:auth<github:ugexe>.tgz (well not that but encoded to work on Windows)
underscores whatever
mst I mean, I was under the impression the auth<> thing was more so you could identify which team you were following if there was a fork or whatever 18:27
ugexe yes it can be used for that
stmuk_ . o O ( or is it "authority" or "author" :/ ) 18:30
ugexe :auth<> is not meant to tell you where you have to download from. its to search all the sources you can download from for it
i.e. you wouldnt have :auth<compunitrepo:ugexe> once you install a distribution 18:31
b2gills example.com:auth<ICANN:DynDns.com> 18:36
nine fears its more gonna be like LeftPad:auth<Facebook:John Doe27> 18:39
ugexe cpan:JRANDOM:JSON-Fast:1.23 <- maybe puny encode this format into a file name so it can be round tripped back to the original 18:40
mst ugexe: if you have an alternative proposal for PSIXDISTS filenames that you want us to try, we can always stick said files under 'exp1' or something 18:42
then if it doesn't work, we design the next experiment, and stick those in 'exp2/'
etc.
ugexe cpan:JRANDOM:JSON-Fast:1.23 <- This is a s22 distribution identity btw (not to be confused with a module identity of JSON::Fast:ver<1.23>:auth<cpan:JRANDOM>) 18:45
stmuk_ colons don't work in windows filenames 18:46
mst and < and > in a filename on *n?x is going to induce aneurysms
ugexe yeah, hence punycoding it or whatever first
nine File systems in general already have means to separate names into parts. They are called directory separators. So why not just AUTH/VER/SHORT-NAME.tgz? 18:50
mst that would, indeed, seem simpler to me. 18:51
ugexe that works but the current situation is putting them all under PSIXDISTS instead of the original auth from the meta
mst argh
you can still do that INSIDE there 18:52
you already told me PSIXDISTS isn't an acceptable auth anyway
PSIXDISTS/Perl6/github.ugexe/1.23/Foo-Bar.tar.gz
ugexe oh ok
nine Or maybe PSIXDISTS/Perl6/github.ugexe/Foo-Bar/1.23.tar.gz 18:53
mst and then all you have to do is, assuming you want to, double check that PSIXDISTS has somehow been granted 'authority' to upload for github.ugexe
oh, also, this still won't help you at all and you'll still need to read the .meta files
since Foo-Bar could contain Foo::Bar and Foo::Baz
ugexe i'll play devils advocate here and say its nice to have all that info in the file name itself so that you can have all your distributions in a single directory without having to recurse multiple times to see what all you have
mst and now you need to read the meta file for 'use Foo::Baz;'
so I'm not sure what we're gaining here anyway except extra moving parts 18:54
ugexe you dont have to search the provided modules neccesarily if you find a matching distribution
mst unless we say 'Foo-Bar.tar.gz' must only contain namespaces starting /^Foo::Bar/ 18:55
ugexe i could be searching for a specific distribution, Foo-Bar, that does not provide neccesarily a Foo::Bar. if I otherwise ask for either a distribution or module then yes it has to search the meta data. default behavior is just to search both which is sane. but if they specify the identity using the distribution format `blah install cpan:JRANDOM:Foo-Bar:1.2` it should only search for that distribution 19:00
stmuk_ or just append the usual SHA1 hash as a suffix to the tar.gz and the uploading process to PSIXDISTS creates a meta.meta file of the .meta files
ugexe if they ask for `blah install Foo::Bar` or `blah install Foo::Bar:auth<cpan:JRANDOM>:ver<1.2>` then it would need to search provided modules 19:01
wrap that long name with quotes naturally
the fact we've been using '::' in our distribution name separators kinda sucks because it implies such a module might be provided 19:16
mst at this point, PAUSE requires that a Foo-Bar dist to be indexed provides Foo::Bar 19:17
well, or at least it requires that the uploader has permissions for Foo::Bar
you could theoretically acquire said permissions separately
but ideally, if you did that, a small gnome would exit the back of your machine and hit you with a mallet
ugexe there are distributions without *any* provides at all. they are just for their bin scripts (i presume anyway) 19:29
github.com/ugexe/Perl6-ecosystems/...c.json#L63
nine Considering the advantages of precompilation, such distributions should be exceedingly rare. Scripts should really just be thin wrappers around modules 20:14
ugexe nine: i agree. maybe we can automate this github.com/ugexe/zef/blob/master/bin/zef 20:15
thats so nearly all its code is precompiled 20:16
but what if the wrapper just created something like `use <resources/bin/script>`
nine MAIN functions 20:17
mst this is why perl5's tradition now is to have App::nameofscript 20:18
ugexe you can precompile any perl6 code, you just need a perl6 entry point. so you precompile bin scripts (that are perl6 code at least) and then have the wrapper act as the sole perl6 entry point 20:19
it doesnt just have to be MAIN functions of course, thats just what zef uses that trick for 20:21
20:22 leont joined 21:19 leont_ joined
ugexe stmuk_: `perl6 -e "my $file = 'Foo:Bar.txt'; say $file.IO.spurt('hello'); say $file.IO.slurp"` this actually works on windows. It just wont show up in Explorer or cmd.exe /dir heh 21:35
:: doesn't work though 21:36
22:08 leont_ joined
stmuk_ oh thats interesting 23:36
msdn.microsoft.com/en-us/library/a...S.85).aspx 23:40
quite a long list of reserved chars

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!