[Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[swedebugia]

[-- Attachment #1: Type: text/plain, Size: 2715 bytes --]

Hi

I am a contributor to OSM and have seen how combining OSM and 
Wikidata/Wikipedia (WP) has been very useful.

I got the idea of adding Wikidata-entries to guix package objecs would 
be fruitful because:

It makes it possible to a more useful list of packages e.g. by showing 
links to WP entries for the program in the users local language. (E.g. 
by firing up a browser from emacs or the shell, or by populating a (per 
channel) html package list (with screenshots, local WP-links, etc.) and 
firing up a throw-away web-server instance serving this with e.g. /guix 
package --list-available-packages-html)/

It would also perhaps be of benefit to WP-contributors because we could 
easily make statistics for how many of the packages in guix a 
Wikidata-entry and/or WP-entry exists. Thus perhaps leading to creating 
of more articles for notable packages or improving WP-articles with 
outdated release information.

*Implementation:*

It could be implemented by adding the fields to package-objects.

The rationale for adding screenshot-url to the recipe is that this 
parsing of wikidata->en-WP->url-for-first-image for every package in our 
list is quite expensive. Better to do it once and perhaps update all the 
screenshot-urls once a year or so.

The rationale for adding WP (list of Wikipeidas with an article in the 
wikidata entry, e.g. ("en" "sv" "es") to the recipe is that this parsing 
of wikidata->WP for every package in our list is quite expensive. Better 
to do it once and perhaps update  once a year or so.

Also to help us to associate new and existing packages with 
wikidata-entries we could devise a guile-programmed way to associate 
wikidata-entries to existing package objects and perhaps use this to 
populate new package-recipes created with /guix import/

Guix would then be the first package manager to both be completely free 
of proprietary software and to leverage knowledge from Wikidata and WP.

What do you think?

/Cheers/
Swedebugia

PS: we could further improve our recipes by adding fields like "release 
date" either via guix import from upstream or by populating from WP. 
This would make it easy for WP-contributors to track when new releases 
happen and perhaps with a script automatically update WP-articles based 
on our recipes when we have newer information.

PPS: Perhaps over time it will even be feasible for WP to use our 
synopsis/descriptions somehow. This would enable us to integrate 
descriptions and translations for programs. E.g. a WP-contributor sees 
that an article in spanish for program x does not yet exist in es.WP but 
a translated synopsis and description already exists in Guix.

[-- Attachment #2: Type: text/html, Size: 3265 bytes --]

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Pjotr Prins]

On Thu, Nov 01, 2018 at 10:44:19AM +0100, swedebugia wrote:
>    Also to help us to associate new and existing packages with
>    wikidata-entries we could devise a guile-programmed way to associate
>    wikidata-entries to existing package objects and perhaps use this to
>    populate new package-recipes created with guix import
> 
>    Guix would then be the first package manager to both be completely free
>    of proprietary software and to leverage knowledge from Wikidata and WP.
> 
>    What do you think?

Absolutely the way forward. Totally excited you want to run with this!

Wikidata is linked data and dry and by using it we can share between
distros and software building projects (conda, easybuild etc.). It
scales because the software maintainers themselves will be encouraged
to update project information - such as a reference to a mailing list
- which is the only way to really keep up-to-date in a scalable way.
Wikipedia will use that information too. Wikidata is DRY.

People don't realise it, but in science to find the right tools for
the job is often a challenge. Wikidata will help us create sections
for tools that address certain tasks. For example variant calling in
sequencing data. I and others here have a direct stake in solving this
problem. This meta-information does not belong in Guix, so we need a
place to handle it. When you have a proof-of-concept we can even
consider writing a paper about it.

Will you join our Guix event at FOSDEM? This would be an interesting
working group. I know also a few people outside Guix who will be
interested.

Pj.

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Amirouche Boubekki]

Hello,

Like Pjotr I think it's a very good idea and the way forward.

Find below my comments with some modulation.

Le jeu. 1 nov. 2018 à 10:39, swedebugia <swedebugia@riseup.net> a écrit :
>
> Hi
>
> I am a contributor to OSM and have seen how combining OSM and Wikidata/Wikipedia (WP) has been very useful.
>
> I got the idea of adding Wikidata-entries to guix package objecs would be fruitful because:

The idea is to add a wikidata identifier for guix packages.

For those that are not familiar with wikidata here is a little summary
of my own.

wikidata is wikimedia project that put together structured data about the world.
wikidata is itself a wiki like wikipedia that anybody can improve it. The goal
of the project is to have a machine readable form of knowledge. One of the use
case for that, is to easily keep wikipedia (and other wik) up-to-date regarding
metadata. Simply said, one could generate, so called, info boxes on wikipedia
from wikidata.

See https://www.wikidata.org/wiki/Q937466 for GNU mailman wikidata entity.

> It makes it possible to a more useful list of packages e.g. by showing links to WP entries for the program in the users local language.
> (E.g. by firing up a browser from emacs or the shell, or by populating a (per channel) html package list (with screenshots, local WP-links, etc.)
> and firing up a throw-away web-server instance serving this with e.g. guix package --list-available-packages-html)

The benefits for guix project:

Immediate benefit:

- It will be easier to translate description and synopsis
- Improve guix packages discover-ability via wikidata SPARQL endpoint
(e.g. give me all guix packages that deal with biology)
- Grab screenshot and other media or metadata about a given package

Other benefits:

- If upstream and other distro adopt wikidata as the Single Source Of
Truth, it will help with packaging and keeping guix up-to-date
- Everything is connected!

> It would also perhaps be of benefit to WP-contributors because we could easily make statistics for how many of the packages
> in guix a Wikidata-entry and/or WP-entry exists. Thus perhaps leading to creating of more articles for notable packages or improving
> WP-articles with outdated release information.

This will be of great benefit for wikidata.

>
> Implementation:
>
> It could be implemented by adding the fields to package-objects.

nitpick, those are records in guile scheme.

> The rationale for adding screenshot-url to the recipe is that this parsing of wikidata->en-WP->url-for-first-image
> for every package in our list is quite expensive. Better to do it once and perhaps update all the screenshot-urls
> once a year or so.

I think the screenshot-url field will not be very helpful that can be
fetched based on wikidata identifier.

>
> The rationale for adding WP (list of Wikipeidas with an article in the wikidata entry, e.g. ("en" "sv" "es")
> to the recipe is that this parsing of wikidata->WP for every package in our list is quite expensive. Better
> to do it once and perhaps update  once a year or so.

Based on the wikidata entry, you can use SPARQL to retrieve the
wikipedia page in various language
and use wikipedia commons links to fetch screenshot.

Simply said, I think we should not add more fields than necessary to
build the package and push to wikidata
the information guix might need for other purposes than distribution and build.

The benefit of this approach is that package definition is not
overloaded with fields and non-code contributors
can still contribute to guix by submitting a screenshot to wikipedia
commons and editing wikidata.

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Tobias Geerinckx-Rice]

Hullo swedebugia,

swedebugia wrote:
> I am a contributor to OSM and have seen how combining OSM and
> Wikidata/Wikipedia (WP) has been very useful.

I'm not too familiar with Wikidata but, like Pjotr and Amirouche, 
think it's a worthwile idea to explore and discuss!

> Guix would then be the first package manager to both be 
> completely
> free of proprietary software and to leverage knowledge from 
> Wikidata
> and WP.

Something to seriously consider here are the FSDG. Where do we 
draw the line between ‘external content’ (as we presumably treat 
home pages, which currently contain everything from proprietary 
software recommendations to #TeamWhite race war progaganda) and 
Guix when such WD/WP data is presented inside our UI?

The FSD[0] comes to mind. Coincidentally, a new mail was just 
posted to the directory-discuss list about Wikidata. It should be 
enlightening to read the responses, if any.

I also assume that we'll be keeping our current set of 
metadata. The requirement for a full-blown Internet connection 
just to explore (not install) packages would be a regression.

Kind regards,

T G-R

[0]: https://directory.fsf.org/wiki/Free_Software_Directory:About
[1]: 
https://lists.gnu.org/archive/html/directory-discuss/2018-11/msg00000.html

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[swedebugia]

[-- Attachment #1: Type: text/plain, Size: 3323 bytes --]

Hi :)

On 2018-11-01 11:21, Pjotr Prins wrote:
> On Thu, Nov 01, 2018 at 10:44:19AM +0100, swedebugia wrote:
>>     Also to help us to associate new and existing packages with
>>     wikidata-entries we could devise a guile-programmed way to associate
>>     wikidata-entries to existing package objects and perhaps use this to
>>     populate new package-recipes created with guix import
>>
>>     Guix would then be the first package manager to both be completely free
>>     of proprietary software and to leverage knowledge from Wikidata and WP.
>>
>>     What do you think?
> Absolutely the way forward. Totally excited you want to run with this!
>
> Wikidata is linked data and dry and by using it we can share between
> distros and software building projects (conda, easybuild etc.). It
> scales because the software maintainers themselves will be encouraged
> to update project information - such as a reference to a mailing list
> - which is the only way to really keep up-to-date in a scalable way.
> Wikipedia will use that information too. Wikidata is DRY.

First I did not understand DRY. Found 
https://en.wikipedia.org/wiki/Don't_repeat_yourself

Agreed, it would be nice if changes to Wikidata would be driven by the 
authors of the programs and shared with leaf nodes (package managers, etc.)

> People don't realise it, but in science to find the right tools for
> the job is often a challenge. Wikidata will help us create sections
> for tools that address certain tasks. For example variant calling in
> sequencing data. I and others here have a direct stake in solving this
> problem. This meta-information does not belong in Guix, so we need a
> place to handle it.
I agree. Though it would be good to find a way to keep queries to 
wikidata to a minimum. Caching maybe.
> When you have a proof-of-concept we can even
> consider writing a paper about it.
Eh, unfortunately my guile and guix fu is not really anything to brag 
about, yet ;-). I am reading up on guile and trying to understand the 
code in guix.

Right now my skill level is at

  * finding spelling errors and unclear text in the manual
  * contribute new simple packages (about to package recoll
    https://en.wikipedia.org/wiki/Recoll and splint)
  * adding better error messages to guix
  * sharing new ideas


This wikidata endeavor would likely take some time for me to accomplish 
with a good mentor.

First up is deciding whether the core procedures interacting with 
wikidata should be in guix or as a separate module. I suggest separate 
module.

Then writing client procedures to interface with the SPARQL API in 
wikidata. This has already been done in python 3 (beta) see 
https://github.com/dahlia/wikidata gplv3+

We could piggyback on this client (essentially making guix dependent on 
python :/) or better yet contribute to one of the existing guile sql 
libraries:

  * https://sourceforge.net/projects/guile-simplesql/files/latest/download
    (unmaintained since 2014 it seems)
  * https://github.com/opencog/guile-dbi (active fork)

The last one looks most promising but I did not look at the code yet.

> Will you join our Guix event at FOSDEM? This would be an interesting
> working group.
Thanks for the invitation. I will think about it.

I look forward to hone my guile and guix skills :)

Cheers
Swedebugia


[-- Attachment #2: Type: text/html, Size: 4932 bytes --]

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[swedebugia]

[-- Attachment #1: Type: text/plain, Size: 1079 bytes --]

Hi Amirouche


On 2018-11-01 14:37, Amirouche Boubekki wrote:
>
>> Implementation:
>>
>> It could be implemented by adding the fields to package-objects.
> nitpick, those are records in guile scheme.
Did you mean to correct my use of "object" here?
Your are right about that.

Scheme Syntax: *define-record-type* /type
(constructor fieldname …)
predicate
(fieldname accessor [modifier]) …
/
I finally begin to understand all these words an Scheme-ways of doing 
things./
/
Have I understood correctly that we have at least 1 nested record types 
in guix? E.g. the package record contains an origin record and a lot of 
other fields.
>> The rationale for adding screenshot-url to the recipe is that this parsing of wikidata->en-WP->url-for-first-image
>> for every package in our list is quite expensive. Better to do it once and perhaps update all the screenshot-urls
>> once a year or so.
> I think the screenshot-url field will not be very helpful that can be
> fetched based on wikidata identifier.
Ok, I understand your point.

/Swedebugia

[-- Attachment #2: Type: text/html, Size: 1955 bytes --]

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Pjotr Prins]

On Thu, Nov 01, 2018 at 11:33:48PM +0100, swedebugia wrote:
> Wikidata is linked data and dry and by using it we can share between
> distros and software building projects (conda, easybuild etc.). It
> scales because the software maintainers themselves will be encouraged
> to update project information - such as a reference to a mailing list
> - which is the only way to really keep up-to-date in a scalable way.
> Wikipedia will use that information too. Wikidata is DRY.
> 
>    First I did not understand DRY. Found
>    [1]https://en.wikipedia.org/wiki/Don't_repeat_yourself
>    Agreed, it would be nice if changes to Wikidata would be driven by the
>    authors of the programs and shared with leaf nodes (package managers,
>    etc.)

Yeah, sorry for the jargon.

>    I agree. Though it would be good to find a way to keep queries to
>    wikidata to a minimum. Caching maybe.

It lends itself naturally to caching. We can use fetching links for
the website - that would be a good start - and later see if we can
enrich package descriptions in Guix itself. In both cases the user
should decide whether they want to use internet access/use a cache
instead.

> When you have a proof-of-concept we can even
> consider writing a paper about it.
> 
>    Eh, unfortunately my guile and guix fu is not really anything to brag
>    about, yet ;-). I am reading up on guile and trying to understand the
>    code in guix.
>    Right now my skill level is at
>      * finding spelling errors and unclear text in the manual
>      * contribute new simple packages (about to package recoll
>        [2]https://en.wikipedia.org/wiki/Recoll and splint)
>      * adding better error messages to guix
>      * sharing new ideas
> 
>    This wikidata endeavor would likely take some time for me to accomplish
>    with a good mentor.

No problem! I think it is actually a very good learning project. We
can help. Start small is my advice.

>    First up is deciding whether the core procedures interacting with
>    wikidata should be in guix or as a separate module. I suggest separate
>    module.

Agree. I think it can be a tool that is separate from Guix itself.
Just start with a simple query and store that either as an
S-expression or as JSON. I think (eventually) we ought to do both so
other languages may use output too. Have a look at the tooling that
generates the website.

>    Then writing client procedures to interface with the SPARQL API in
>    wikidata. This has already been done in python 3 (beta) see
>    [3]https://github.com/dahlia/wikidata gplv3+
> 
>    We could piggyback on this client (essentially making guix dependent on
>    python :/) or better yet contribute to one of the existing guile sql
>    libraries:
>      * [4]https://sourceforge.net/projects/guile-simplesql/files/latest/do
>        wnload (unmaintained since 2014 it seems)
>      * [5]https://github.com/opencog/guile-dbi (active fork)
> 
>    The last one looks most promising but I did not look at the code yet.

Personally I would use the Python stuff first and then slowly replace
that with Guile. That way you get to results fast and we can improve over
time. I personally take no issue with mixing stuff. And because it is
a separate tool it is your choice anyway. I think also, initially, we
should build a separate website that can display all this information.
That way you have full freedom on implementation and experiments.

> Will you join our Guix event at FOSDEM? This would be an interesting
> working group.
> 
>    Thanks for the invitation. I will think about it.
>    I look forward to hone my guile and guix skills :)

Please come. FOSDEM is awesome.

Pj.

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[swedebugia]

[-- Attachment #1: Type: text/plain, Size: 2297 bytes --]



On 2018-11-02 08:24, Pjotr Prins wrote:
> On Thu, Nov 01, 2018 at 11:33:48PM +0100, swedebugia wrote:  >> When you have a proof-of-concept we can even consider writing a >> 
paper about it.
That would probably be fun :) I did not write a paper for a long time 
and never in the field of computing.

>> This wikidata endeavor would likely take some time for me to  >> accomplish with a good mentor. > > No problem! I think it is 
actually a very good learning project. We > can help. Start small is my 
advice.
Thank you! I feel motivated.
>  >> First up is deciding whether the core procedures interacting with >> 
wikidata should be in guix or as a separate module. I suggest >> 
separate module. > > Agree. I think it can be a tool that is separate 
from Guix itself. > Just start with a simple query and store that either 
as an > S-expression or as JSON. I think (eventually) we ought to do 
both so > other languages may use output too. Have a look at the tooling 
that > generates the website.
Ok. Is there a json guile module?
Will take a close look at the python module.

>  >> Then writing client procedures to interface with the SPARQL API in 
 >> wikidata. This has already been done in python 3 (beta) see >> 
[3]https://github.com/dahlia/wikidata gplv3+ >> >> We could piggyback on 
this client (essentially making guix >> dependent on python :/) or 
better yet contribute to one of the >> existing guile sql libraries: > > 
Personally I would use the Python stuff first and then slowly > replace 
that with Guile. That way you get to results fast and we can > improve 
over time. I personally take no issue with mixing stuff. And > because 
it is a separate tool it is your choice anyway. I think also, > 
initially, we should build a separate website that can display all > 
this information. That way you have full freedom on implementation > and 
experiments.
How would I go about mixing python and guile?
Export the list of package records from guile -> JSON and import in a 
python script?
Can I call a python-script from guile and receive input from it?

So something like:
iterate over record fields
calling a python script to fetch data from wikidata
acting on the data
feeding it to the console/web template code
fire up the webserver serving the html

Cheers
Swedebugia


[-- Attachment #2: Type: text/html, Size: 3262 bytes --]

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Pjotr Prins]

On Fri, Nov 02, 2018 at 12:37:15PM +0100, swedebugia wrote:
>    Ok. Is there a json guile module?

Yup. Dave wrote one, I believe. There is one shipped with Guix. It is
probably this one https://directory.fsf.org/wiki/Guile-json#Details

>    How would I go about mixing python and guile?  Export the list of
>    package records from guile -> JSON and import in a python script?
>    Can I call a python-script from guile and receive input from it?
>    So something like: iterate over record fields calling a python
>    script to fetch data from wikidata acting on the data feeding it
>    to the console/web template code fire up the webserver serving
>    the html

That could work, for sure.

Pj.

Re: [Feature idea] Adding wikidata, wikipedia & screenshot-url fields to package-recipes

[Catonano]

[-- Attachment #1: Type: text/plain, Size: 810 bytes --]

Il giorno ven 2 nov 2018 alle ore 00:20 swedebugia <swedebugia@riseup.net>
ha scritto:

> Hi Amirouche
>
> On 2018-11-01 14:37, Amirouche Boubekki wrote:
>
>
> Implementation:
>
> It could be implemented by adding the fields to package-objects.
>
> nitpick, those are records in guile scheme.
>
> Did you mean to correct my use of "object" here?
> Your are right about that.
>
> Scheme Syntax: *define-record-type*
>
>
>
> *type (constructor fieldname …) predicate (fieldname accessor [modifier])
> … *
> I finally begin to understand all these words an Scheme-ways of doing
> things.
>
> Have I understood correctly that we have at least 1 nested record types in
> guix? E.g. the package record contains an origin record and a lot of other
> fields.
>

yes, that's correct

[-- Attachment #2: Type: text/html, Size: 1478 bytes --]