should fix #34840 i guess... i'll test later

@alikon yes it should fix it

I have tested this item ✅ successfully on ef15ffe

I have tested this item ✅ successfully on ef15ffe

RTC

Sorry this isn't correct (in my opinion) - I accept some of this is somewhat more subjective than other PR's though. There's kinda two parts here so I'm going to address each separately

First of all the getRouter method. Happy to add that to the application so we can use \Joomla\CMS\Router\Route::link. However in terms of the default router even though there is no site router - I think it's wrong to load the site router explicitly as the default. I'd rather it errors out by default if it's called with the wrong data rather than be wrongly relied on.

In terms of the setting of the HTTP_HOST I'm very sure that overloading server side variables is wrong - plus we still end up with potentially bad behaviour here when live_site isn't set (I'm still yet to set that on any of my sites) - and we just end up with that being null and still getting junk URLs out on each of those instances. This needs a proper fix (I'm not 100% sure how) where I guess the application has the ability to inject the base domain into the Uri::getInstance method which would override the HTTP_HOST. Possibly by taking advantage of the WebApplication's https://github.com/joomla-framework/application/blob/2.0-dev/src/AbstractWebApplication.php#L1001-L1004 variables and doing some sort of similar config in the console application so we have consistency over both applications. Open to ideas on the best way forward there - but the idea being we need consistency over both web and console and to treat them as equals rather than bodging around the console.

First of all the getRouter method. Happy to add that to the application so we can use \Joomla\CMS\Router\Route::link. However in terms of the default router even though there is no site router - I think it's wrong to load the site router explicitly as the default. I'd rather it errors out by default if it's called with the wrong data rather than be wrongly relied on.

I changed the code, so when no name is set that that an exception is thrown. But I left the code in place when a cli router is requested that it uses the site router as this is the case when Route::_() is called.

I'm very sure that overloading server side variables is wrong

There is no other way to get the web site url than from the live site in console applications. But I removed it and now every extension dev has to set it in the console command by itself. Anyway, for me it is important to have routing work in console applications.

I changed the code, so when no name is set that that an exception is thrown. But I left the code in place when a cli router is requested that it uses the site router as this is the case when Route::_() is called.

But the entire point of Route::_ has always been it's been application specific. That's why we introduced Router::link to start with in 3.7 so we could do cross-application routing.

Ok, removed the fallback. All I did was to remove some new code since RTC, So @wilsonge can you consider it now ready to merge?

Thanks!

I think for 4.1 (as technically it's a low impact b/c break) we might also want to consider putting getRouter into the \Joomla\CMS\Application\CMSApplicationInterface so typehints work nicer - but this is definitely all we can do for the 4.0-dev branch.

Definitely, the whole getRouter stuff needs some love. @bembelimen can you ping me when 4.0-dev is merged into 4.1? So I will do a followup pr.

@wilsonge made a 4.1 pr for the interface changes in #36631. But #36630 should be merged first.