Module list should check modules META.info is correct about ecosystem

[jonathanstowe]

The META.list is basically maintained manually. So is only checked when the list is built into the aggregate of the META files by some cron job, however it doesn't currently check the source-url for validity.

I guess we could rewrite the url to http:// and do a head on it to check it's okay.

[ranguard]

@zoffixznet I currently store the sha of the master branch of each git repo, if that changes then I know it's a new version (as far as releasing to CPAN is concerned).

https://github.com/perl6modules/perl6-module-uploader publish_to_cpan.pl is the code and upload_tracker.json is the tracker file.

https://github.com/perl6modules/dists-uploaded-to-cpan is the actual releases as well.

[jonathanstowe]

I can look at the making a Perl 5 version if no-one else picks it up

[zoffixznet]

It already does. You can grep http://modules.perl6.org/update.log for [error]

I'm unsure what more can be done about it. GitHub network errors are frequent, so it nagging somewhere automatically whenever it sees an error is LTA.

[zoffixznet]

I stand corrected. It figures out the URL from the META.list META6.json URL, not from source-url field, so it doesn't check it.

[MARTIMM]

I am sorry for this bug, I'll repair it as soon as possible next morning.
Regards,
Marcel

On September 27, 2016 22:42:47 Leo Lapworth [email protected] wrote:

https://raw.githubusercontent.com/perl6/ecosystem/master/META.list

Includes:

https://raw.githubusercontent.com/MARTIMM/unicode-precis/master/META.info

Which currently looks like this

{
  "author": "Marcel Timmerman",
  "authors": [
    "Marcel Timmerman"
  ],
  "depends": [
  ],
  "description": "Implements PRECIS framework, rfc7564 and others. Obsoletes 
  stringprep and saslprep",
  "license": "The Artistic License 2.0",
  "name": "Unicode::PRECIS",
  "perl": "6.c",
  "provides": {
    "Unicode::PRECIS": "lib/Unicode/PRECIS.pm6",

    "Unicode::PRECIS::FreeForm": "lib/Unicode/PRECIS/FreeForm.pm6",
    "Unicode::PRECIS::FreeForm::OpaqueString": 
    "lib/Unicode/PRECIS/FreeForm/OpaqueString.pm6",

    "Unicode::PRECIS::Identifier": "lib/Unicode/PRECIS/Identifier.pm6",
    "Unicode::PRECIS::Identifier::UsernameCaseMapped": 
    "lib/Unicode/PRECIS/Identifier/UsernameCaseMapped.pm6",
    "Unicode::PRECIS::Identifier::UsernameCasePreserved": 
    "lib/Unicode/PRECIS/Identifier/UsernameCasePreserved.pm6",

    "Unicode::PRECIS::Tables": "lib/Unicode/PRECIS/Tables.pm6"
  },
  "source-url": "git://github.com/MARTIMM/unicode-stringprep.git",
  "version": "v0.3.0"
}

The _source-url_ is wrong and 404's - it would be nice for the system
maintaining the master list to verify this sort of thing automatically.

You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
#255

[MARTIMM]

Hi,
The link is repaired.
Btw, the check might also be done in Test::META module.
Greetings,
Marcel

P.s. Sorry for the mess i've got you into...

[ranguard]

@MARTIMM thanks for fixing - leaving ticket open as it wasn't specifically about your module's problem, so much as there should be a safe place to use as the master - otherwise everyone has to implement their own validation

[ranguard]

@drforr FYI https://raw.githubusercontent.com/drforr/perl6-Perl6-Parser/master/META.info - boom, missing commas

{
    "name"         : "Perl6-Parser",
    "version"      : "*",
    "description"  : "Perl 6 parsing",
    "test-depends" : [
        "Test"
    ],
    "provides"     : {
        "Perl6::Parser"        : "lib/Perl6/Parser.pm6"
        "Perl6::Parser::Factory" : "lib/Perl6/Parser/Factory.pm6"
        "Perl6::Parser::Validator" : "lib/Perl6/Parser/Validator.pm6"
    },
    "source-url"   : "git://github.com/drforr/perl6-Perl6-Parser.git"
}

[jonathanstowe]

This is why https://github.com/jonathanstowe/Test-META :)

[ranguard]

@jonathanstowe can you (ecosystem team) make that a pre-req going forward for any module that is going to be added to https://raw.githubusercontent.com/perl6/ecosystem/master/META.list ?

[zoffixznet]

can you (ecosystem team) make that a pre-req going forward for any module that is going to be added to

👎 on such puritanical requirements. Even with it added, there's no guarantee of lack of errors. Authors have to run those tests too whenever making changes. All such a requirement would accomplish is create lots of noise, lots of confused people, and lots of dead PRs to the ecosystem.

[zoffixznet]

And more to it.... I don't think this Issue even belongs in the Ecosystem domain. This is more to the tune of what Perl 5 Testers service does. Our version is http://smoke.perl6.org/report in whatever shape it currently exists.

[shadowcat-mst]

Ok, so, the thing here is: we're trying, with PSIXDISTS, to do trial uploads so e.g. zef can work against cpan dists for testing, and so the perl6 metacpan indexer can run against cpan, and various other infrastructure can be trialed.

Requiring valid metadata to introspect doesn't really seem "puritanical" to me, it just seems like the absolute bare mininum required to be able to start building out downstream things from the ecosystem.

Comparing it to cpantesters is a massive category error. cpantesters only tests dists that are at least vaguely valid, after all.

[zoffixznet]

Requiring valid metadata to introspect

Valid metadata doesn't require a module built by a third party that brings in 2 deps with it. Moreover, the module doesn't magically make that metadata valid. It requires module authors to execute the tests that use the module.

In your context, it would be the indexer's job to ensure metadata's validity and refuse to index a distro if it is invalid.

Forcing the users to depend on Meta-Checker-De-Jour is silly and doesn't even solve the actual problem.

[shadowcat-mst]

@zoffixznet You appear to have accidentally heard an argument I'm not making.

All I'm saying is that it would be nice to have an index where the metadata can be assumed valid and the code can be assumed to exist. Many downstream users could then consume said index.

[jonathanstowe]

I wouldn't be in favour of mandating the use of Test::META for all modules either, I would be in favour of encouraging a culture whereby people value using the greatest range of tests they have available which may include testing the META data, testing the documentation etc etc, but this is a discussion for elsewhere.

As it currently stands, the best place to ensure the validity of the META files is at the point they are aggregated into the projects.json from the META.list. This is less than perfect (because there isn't a mechanism to feed back a failure to the user directly,) at present, but we can work on that.

So no to mandating a particular set of tests, yes to better checking in the back end and possibly feeding failures back to the user earlier.

[shadowcat-mst]

That sounds great. @jonathanstowe key thing here is that it's basically blowing up the PSIXDISTS process, and while we could hack around it locally, it seems like it'd be a better idea to have a central "this should be valid" list instead.

[zoffixznet]

.oO( A GitHub bot that automatically checks the PRs to ecosystem, and merges them if the META is good, or comments on them if there are errors... )

[zoffixznet]

Oh, wait, that still has the issues of changes to modules...

.oO( hack modules.perl6.org dist builder to automatically submit Ecosystem issues and ping authors when issues detected ).

[zoffixznet]

@ranguard there was a brief discussion on IRC...

What's considered a "distro" for PSIXDISTS? That is, where would be the best place to put an automated META check if it's not in PSIXDISTS thing itself. How does it determine when a new release of a module appeared, etc?

The modules.perl6.org dist builder has an easy possibility to insert such a check as it already looks for whether a repo has got new commits. I'm just curious in which format PSIXDISTS stuff could find that useful.

[zoffixznet]

I've added a METAChecker postprocessor to modules.perl6.org builder that checks source-url URL points to something that gives a 200 code. What else is needed and how to make this info more useful?

Currently, it's visible for anyone curious on http://modules.perl6.org/update.log, but there's no automated checks for anything beyond that. Moreover, failures happen frequently, so GitHub may not return a 200 each and every time, so there needs to be some statistical (?) means to ascertain whether a distro's META is OK.

[zoffixznet]

Sounds like Test::META6 (Test::Perl6::META? Test::Perl6META?) Perl 5 module would be the perfect solution and it can be run by that script when preparing an upload, as well as the modules.perl6.org dist builder to indicate any errors in update.log.

[ranguard]

@jonathanstowe that would help me lots thanks, even if not the ideal long term solution...

It is tricky, I was going to propose 'valid dists' be maintained in a separate list and updated (via cron) in the same way I'm checking (based on sha) but then you need to update all links to include the sha... which ultimately is why I'm putting these on CPAN in the first place (so there are snapshot versions of the entire repo at a known point in time and that safeguards someone deleting their repo or doing other funky stuff with it).

Hopefully the p6 community can then work on the tools that encourage authors to upload their own releases to CPAN and I can turn off my script :)

[drforr]

Mea culpa as well; I'm in a habit of cp -r perl6-Random-Dist perl6-My-New-Dist ; rm -rf .git ; , submit, without really looking at the metadata closely. While having to use p5 to maintain a p6 repo is not ideal, maybe creating a set of small plugins for Dist::Zilla would be a good approach, as it's already got helpers for PAUSE upload?

[jonathanstowe]

I've actually got a script somewhere that I use to generate the META files for my modules, I could tosh it up a bit and add it to the META6 distribution, I think all that it needs is to be able to store a little bit of config (for author and the like.) I'll put it on the stack of things I'm planning to do (having a full-time job again sucks ....)

[ranguard]

Using https://raw.githubusercontent.com/perl6/ecosystem/master/META.list as my source of info...

• The following have metadata which points to sources which are not valid GH repos

p6-fcgi
p6-fcgi-psgi
p6-unix-privileges
p6-jdf

• https://raw.githubusercontent.com/gotoexit/Concurrent-BoundedChannel/master/META.info points to https://github.com/jnthn/p6-concurrent-iterator which is odd as that is already under jnthn/p6-concurrent-iterator

• https://raw.githubusercontent.com/drforr/perl6-Inline-Scheme-Guile/master/META.info points to https://github.com/niner/Inline-Scheme-Guile which doesn't exist

• For extra fun this one has no master branch

szabgab/Perl6-Maven

I'm going to make my code just skip these issues quietly from now on as my code is doing the right thing as far as I am concerned, but doesn't have a decent enough source to work from to ensure that all ecosystem modules are uploaded to CPAN.

[jonathanstowe]

The META.list is probably not the place to be getting the data from as this is totally raw and unvarnished, it gets processed into some other file which will already have checked that the META file is retrievable and somewhat sane, That file may or may not be http://ecosystem-api.p6c.org/projects.json but you would be better checking which file the modules site uses.

I would thoroughly recommend getting on #perl6 on IRC as it will enable a quicker exchange of information :)

[stmuk]

That uri is just the one panda uses. zef uses files under https://github.com/ugexe/Perl6-ecosystems Pragmatically that might be the best source at the moment. S

…

On 8 January 2017 at 15:46, Jonathan Stowe ***@***.***> wrote: The META.list is probably not the place to be getting the data from as this is totally raw and unvarnished, it gets processed into some other file which will already have checked that the META file is retrievable and somewhat sane, That file may or may not be http://ecosystem-api.p6c.org/ projects.json but you would be better checking which file the modules site uses. I would thoroughly recommend getting on #perl6 on IRC as it will enable a quicker exchange of information :) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#255 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEUuYywNEToyEHwe_pvdYB6K3hYGwMegks5rQQTAgaJpZM4KIIc5> .

-- 4096R/EA75174B Steve Mynott <[email protected]>

[jonathanstowe]

It would be better to use whatever the modules.perl6.org uses I would say :)

[ranguard]

19:16 < ranguard> RabidGravy: I understood modules.perl6.org to be built from
https://raw.githubusercontent.com/perl6/ecosystem/master/META.list
19:16 < ranguard> ^^ anyone know different?
19:16 < notviki> ranguard: That's accurate

[jonathanstowe]

There is a travis ci job that is run on every commit to the modules list that checks the META of the modules that are added now.

[ranguard]

Nice!

…

On 14 September 2017 at 14:07, Jonathan Stowe ***@***.***> wrote: There is a travis ci job that is run on every commit to the modules list that checks the META of the modules that are added now. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#255 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AADr53hl6nqBFFxHCPA6611K3SYHS32Wks5siSUhgaJpZM4KIIc5> .

[JJ]

Should this be closed?