#20199 Reverted BaseUrl::isRelative() behaviour to previous 2.0.49 ve…

[edegaudenzi]

…rsion due to yii\web\UrlManager::createAbsoluteUrl() malfunction depending on this.

Q	A
Is bugfix?	✔️
New feature?	❌
Breaks BC?	❌
Fixed issues	#20199

[samdark]

@edegaudenzi please check failing tests.

[rob006]

This change is incorrect, in reverts fix for #17191 (that is why tests fails).

Also, see #20199 (comment)

[edegaudenzi]

Ok @samdark, I will at a certain point today. I've just given a brief look at them and seems a couple of tests are starting from the wrong assumption links like acme.com/test?tnt-link=https://tnt.com/ are valid URIs; accordingly to rfc3986 (T. Berners-Lee) they are not: their 'legal' counterpart would be instead the Percent-Encoding version of it: acme.com/test?tnt-link=https%3A%2F%2Ftnt.com%2F
But I'll let you know as soon as I can give a deeper look into it.

[rob006]

I'm not sure why do you think that acme.com/test?tnt-link=https://tnt.com/ is not valid. This RFC explicitly mentions that you can skip encoding for readability:

https://stackoverflow.com/a/68151221/5812455

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 64.99%. Comparing base (ec46ede) to head (9f74eba).
Report is 24 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff              @@
##             master   #20203      +/-   ##
============================================
- Coverage     64.99%   64.99%   -0.01%     
- Complexity    11391    11393       +2     
============================================
  Files           430      430              
  Lines         36923    36925       +2     
============================================
  Hits          23998    23998              
- Misses        12925    12927       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[edegaudenzi]

As promised, on Friday I've dug deeper and found out the problem was in fact the a non rfc3986 compliant url in the tests. I've changed that one, and for sake of clarity I've also added an additional case aimed to test a more relaxed version of rfc3986, but still legal. Everything's is committed now in the pr.

[rob006]

@edegaudenzi You may argue that acme.com/test?tnt-link=https://tnt.com/ is not a valid URL, but it is correctly handled by many tools (PHP included) and it is definitely not an absolute URL. Your PR introduces regression that treats acme.com/test?tnt-link=https://tnt.com/ as absolute URL (and thus breaking ensureScheme()) and you altered tests to hide this regression.