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 7 July 2016.
08:19 domidumont joined 08:24 domidumont joined 08:53 autarch joined, ranguard joined, b2gills joined, FROGGS joined, hoelzro joined, retupmoca joined, moritz joined 11:36 edehont joined 12:03 domidumont joined 12:29 TimToady joined 14:56 ugexe joined
ugexe is processing -I and -M parameters in order instead of all -I followed by all -M a valid solution loading CUR from the command line? i.e. `-Ilib/CompUnit/Repository/Foo -MCompUnit::Repository::Foo -Ifoo#bar/baz` (so it would load the module/CUR before lib foo#bar/baz, which itself may contain another CUR used in another parameter further up) 14:56
mst so, perl5 does all -I first, then -M 14:57
and there are a bunch of advantages to this 14:58
surely perl6 has something lib.pm like?
so -Mlib=foo#bar/baz
nine ugexe: AFAIR -I are collected when setting up the repository chain. When we come across an unknown repo like CompUnit::Repository::Foo, we defer setting up that repo. Later we try to load the repo implementation from the repos we could resolve earlier. 14:59
I don't actually know how -M is implemented 15:01
ugexe i need to come up with a better example of what I was wrestling with, but heres a basic example 15:04
nickl@li685-90:~/perl6/Perl6-CompUnit--Repository--Tar$ perl6 -Ilib -MCompUnit::Repository::Tar -I"CompUnit::Repository::Tar#t/data/zef.tar.gz" -e 'use Zef;' 15:05
Could not find CompUnit::Repository::Tar
nine You shouldn't even need the -MCompUnit::Repository::Tar. Repository chain setup will already try to load a CompUnit::Repository::Tar after discovering that this name is not a known short-id
ugexe if the -I"CUR::Tar#t/data/zef.tar.gz" is moved into the -e as `use lib <CUR::Tar#t/data/zef.tar.gz>` then it works 15:06
mst right, so 15:09
-Mlib= should also work?
ugexe -M treats lib= as part of the actual module name. If you are asking theoretically then I'm not sure... I wasn't aware of that 15:11
nine: without the -M (or use CompUnit::Repository::Tar) I get `Could not find CompUnit::Repository::Tar` 15:12
nine We don't have support for import lists for -M yet 15:13
mst ah
ugexe: sorry. that's how perl5 handles it
I'm mostly used to perl6 trying to be a superset
ugexe github.com/rakudo/rakudo/blob/nom/...d.nqp#L607 # -M handling 15:14
nine ugexe: does this error come from resolve-unknown-repos?
ugexe nine: gist.github.com/ugexe/e93a68b28b76...c3377187d8 # under the `Could not find Zef::CLI` in the example lacking a `use CUR::Tar;` the first line of repos listed shows 3 items instead of 1 15:26
nine ugexe: and it should be in /home/nickl/perl6/Perl6-CompUnit--Repository--Tar/lib I guess? 15:28
15:29 domidumont joined
ugexe CUR::Tar is in that folder yes. Zef::CLI is in t/data/zef.tar.gz 15:29
nine I wonder if there is a difference between a one liner using -e and running a proper script?
Oh! "use lib" probably doesn't even set up $*REPO which would be where we resolve those Unknown repos. Can you try a BEGIN $*REPO; before the use lib? 15:30
mst ugexe: sorry if my thoughts are occasionally daft 15:31
I know a lot about toolchainy stuff, more than I ever wanted to know about the perl5 toolchain, and then sometimes I assume slightly more of it applies than actually does and make no sense 15:32
ugexe Same error with s/use CompUnit::Repository::Tar;/BEGIN $*REPO/ via github.com/ugexe/Perl6-CompUnit--R...t/03-use.t
mst: I'm happy to hear established solutions to similar problems to make things more intuitive for others 15:34
nine ugexe: ah in the test file it couldn't have made a difference as the use Test; before that does already set up $*REPO 15:36
ugexe but BEGIN $*REPO should occur before use Test no?
nine use is BEGIN time, too
ugexe but i thought `use XXX: BEGIN { use YYY; };` would load YYY first 15:37
nine No, use Foo; is (just like in Perl 5) something like BEGIN { need Foo; import Foo; } 15:38
In other words: the Grammar calls $*W.do_pragma_or_load_module() immediately upon finishing parsing that statement
ugexe hmm, over a year ago I *thought* I had used that to get around a bug in Test.pm but it must have somehow changed the import order or something 15:39
hell it was probably 2 years ago 15:41
mst ugexe: it's not perfect, but there's a bunch of uses for having -I first 15:42
notably options stacking due to scripts chaining stuff and etc.
I am *not*, btw, claiming that a priori the perl5 way is correct
I've had to deal with problems because of it doing 'all -I, then all -M'
ugexe but wouldnt you still get that effect, just you would have to put your -M after all your -I? 15:43
mst but the -Mlib= workaround is how I deal with it in perl5 because obviously it's not changeable there at this point
yes. but, again, if option are coming from multiple sources, it gets complicated
hang on, I need to recover some neurons and remember a concrete case where it helps rather than hindres
because of course when the default approach just works, you don't notice 15:44
ugexe i would imagine it would work more like `use XXX; use lib "lib/XXX-path";` where this would error, but `use lib "lib/XXX-path"; use XXX;` woudl succeed
mst ok, so, here's an example 15:47
PERL5OPT="-MFoo" 15:48
15:48 domidumont joined
mst to preload some sort of module 15:48
if you have a wrapper script that does 'exec perl -I$MYLIBS $MYPROG'
you want the -M processed after, because otherwise it might load some bits from the wrong @INC chain
which can be hilariously annoying later
ugexe: I'm right in thinking that rakudo can load more than one version of the same module, if two different modules ask for different versions of their dep, right? 15:51
ugexe true. but what if you want to control where in the repo chain Foo is loaded from? Say (as stupid as this example is) you have lib/Test.pm but you want -MTest to load from rakudos default repo-chain? The -Mlib= thing seems like it can fit such a bill I think 15:52
mst: yes, or at least its supposed to
mst ok, so, the biggest thing that goes wrong with the perl5 stuff isn't nearly so much an issue for you, I think 15:53
which means it might turn out that 'process in order for POLS' is a better trade-off
nine ugexe: your example is not stupid at all. It's actually exactly what I'm try to debug right now. There's a bootstrapping issue with using the Staging repo to install the Staging repo implementation. 15:55
mst ugexe: right, but, what I'm saying is, it might be ok to have '-MTest -Ilib' work the way you'd expect it to 15:57
ugexe yeah i understand. i was actually hoping to hear potential drawbacks because I wasn't able to think of any of the top of my head 16:00
mst the ones I can think of appear to basically be drawbacks that you don't suffer from
I should, however, probably try and take a survey of perl5-ers to see if anybody else can think of something 16:01
mst asks #p5p and #toolchain 16:08
ugexe: initial responses: paste.scsys.co.uk/526265 16:17
also 16:18
17:18 < Zefram> also, having all -I options processed first means that modules have a consistent view of @INC. that'd probably be useful for something
which, in *general*, is DWIM 16:19
I feel like for CURs themselves an escape hatch may be required or something, but Iunno
(I'm now at "I don't have an opinion here and I bet the problems with the other approach wouldn't become fully apparent unless we did it argh")
notable is that given PERL5LIB goes first (and probably needs to), 'lifting' -I options into PERL5LIB for whatever reason is way simpler if you don't have to worry about ordering so much 16:23
< kentnl> best I have is "easier to store in an options hash" 16:24
ugexe: if these seem convincing, 'make -Mlib= work so people can have finer control, and keep the current ordering' has worked fine for me in perl5 over the years 16:25
16:30 domidumont joined
mst I think if you see -I as REALLYREALLYBEGIN { use lib ... } 16:32
it all sort of makes sense
ugexe isn't zeframs example just a general repo chain ordering issue? i'm not wrapping my head around how loading a module at a specific point in the chain applies outside of loading any dependencies that module might have further down the chain 16:36
or is it along the lines of "-Mxxx would never be able to apply before PERL6LIB=..." (assuming PERL6LIB gets put at the head) 16:39
mst I went and asked other people because my brain was melting 16:42
I am not going to successfully answer that :)
(maybe tomorrow)
ugexe fwiw the order of `PERL6LIB="yyy/" perl6 -Izzz -e 'use lib "xxx";'` is xxx, zzz, yyy, <rest of default repos> 16:44
mst yes, also how perl5 would do it 16:52
17:26 mst joined, mst_ joined 17:46 domidumont joined

Powered by Cro and the Raku® Programming Language.

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

Thank you!