Some linkified urls are including the app server name

[JulienParis]

In a dataset's text, an url written in a markdown format is transformed into a <a> tag but its href attribute contains the app server name within the url...

You can check an example of this behaviour here => test Bleach dataset

Detailed Description

if you write a text as :

link 03 : [rte-france-03.com](rte-france-03.com)

you get on the client side something like :

<p>link 03 : <a href="https://www.data.gouv.frrte-france-03.com">rte-france-03.com</a></p>

... which is wrong and not desired behaviour I guess : it shoud not contain www.data.gouv.fr before rte-france-03.com in the href attribute

Possible Implementation

check PR #2486

[abulte]

What does the markdown says about that? I wouldn't be shocked if http(s):// was mandatory in a markdown external link.

[abulte]

Cf https://pad.incubateur.net/WK9dDgvlSZSP--G059UVMA?both for a test on another markdown parser (similar behaviour as data.gouv.fr today).

[abulte]

What's your suggested behaviour BTW?

[JulienParis]

the wrong netloc value is injected/recreated at this moment =>

udata/udata/frontend/markdown.py

Line 46 in 01534c6

attrs[(None, 'href')] = '{scheme}://{netloc}{path}'.format(

[JulienParis]

What's your suggested behaviour BTW?

link 03 : [rte-france-03.com](rte-france-03.com)

should produce

<p>link 03 : <a href="https://rte-france-03.com">rte-france-03.com</a></p>

and also a rel="nofollow" attribute, but that's described in another issue

[abulte]

Cf https://gist.github.com/abulte/9b4ffe706f4667a13db3265bec14538e

So I think the real issue is that we don't add a / between data.gouv.fr and the (internal) URL when we need one. Other than that it's OK. #2497 is not an issue either.

[JulienParis]

What should it produce then ?

[abulte]

To be clear, [rte-france-03.com](rte-france-03.com) should yield a link to https://www.data.gouv.fr/rte-france-03.com

[JulienParis]

but following that logic [www.rte-france-06.com](www.rte-france-06.com) should also yield href="https://www.data.gouv.fr/www.rte-france-06.com"... which seems very weird to me (but better for SEO :)

[abulte]

It's not a question of SEO, it's that we shouldn't try to guess if a link is internal or external. If the user wants an external link, she uses http(s)://, if not we consider it as an internal link and enforce our site name in front of it. By doing that we ensure that nofollow is used when needed.

[abulte]

This is what GitHub does in the end. They keep it as a relative link but we can't because we need the nofollow.

[JulienParis]

nope, to https://www.data.gouv.fr/www.rte-france-06.com

sorry I rectified my comment above, but ok I get it better now...

[JulienParis]

Github behaviour just to compare...

- email text : me@lalala.com
- email text : me@lalala-composed.com
- email md : [me@lalala-composed.com](mailto:me@lalala-composed.com)
- plain text : lalala.com
- plain text + http : http://lalala.com
- alone : [lalala.com](lalala.com)
- www : [lalala.com](www.lalala.com)
- http : [lalala.com](http://lalala.com)
- https : [lalala.com](https://lalala.com)

• email text : me@lalala.com => to mailto
• email text : me@lalala-composed.com => to mailto
• email md : me@lalala-composed.com => to mailto
• plain text : lalala.com => to plain text
• plain text + http : http://lalala.com => external to GH
• alone : lalala.com => internal on GH
• www : lalala.com => internal on GH
• http : lalala.com => external to GH
• https : lalala.com => external to GH

So yes Bleach behaves another way and induce an external url, that is a root cause of our mutual miscomprehension on this matter :)

[abulte]

So yes Bleach behaves another way and induce an external url, that is a root cause of our mutual miscomprehension on this matter :)

It's not Bleach, it's our nofollow_callback function. And it's because we enforce a nofollow, like the function name says.

[JulienParis]

I did an update on a live test => https://www.data.gouv.fr/fr/datasets/bleach-test-002/

test **BLEACH** 002

----
### emails bleaching

#### as markdown 
email-01-md : [email-01-md@email.com](mailto:email-01-md@email.com)
email-02-md : [email-02-md@email-multi-composed.com](mailto:email-02-md@email-multi-composed.com)

#### as plain text
email-01-txt : email-01-txt@email.com
email-02-txt : email-02-txt@email-multi-composed.com

---- 
### links

#### as markdown 
link-md-01 : [link-md-01.com](link-md-01.com)
link-md-02 : [http://link-md-02.com](http://link-md-02.com)
link-md-03 : [https://link-md-03.com](https://link-md-03.com)

#### as plain text
link-txt-01 : link-txt-01.com
link-txt-02 : http://link-txt-02.com
link-txt-03 : www.link-txt-03.com
link-txt-04 : http://link-txt-04.com

--- 
### internal / client explicitly writes an internal link
internal-md-01 : [local/route/internal-01](local/route/internal-md-01)
internal-md-02 : [/local/route/internal-md-02](/local/route/internal-02)
internal-md-03 : [/ ... internal-md-03](/)

[JulienParis]

what confused me was link-txt-01 and link-txt-02 are not currently behaving as you mentionned above but link-md-01 is (approximatly modulo the /)...

[JulienParis]

Anyway I'll work on a solution giving the exact same behaviour than GH + adding nofollow if link is external + neutralizing all emails...

seems easier to describe that way... is that ok for u @abulte @quaxsze ?