[BrowserKit] fixed BC Break for HTTP_HOST header

[brizzz]

Q	A
Branch?	2.7, 2.8, 3.x, 4.x
Bug fix?	yes
New feature?	no
BC breaks?	no
Deprecations?	no
Tests pass?	yes/no
Fixed tickets	#22933
License	MIT
Doc PR	n/a

The situation well described in the original issue. I will add only:

• Fixed server values in BrowserKit #10549 - makes server parameters to take precedence over URI.
• [BrowserKit] Corrected HTTP_HOST logic #16265 - partially revererts Fixed server values in BrowserKit #10549. Makes server parameters do not affect URI. But this is only true for Client::request(). It is still possible to set host for URI by Client::setServerParameters() when URI is realative (see examples below).

I propose a compromise solution: add to HTTP_HOST header power to override URI when it is relative.

Proposed solution:

• if the request URI is relative, then use the HTTP_HOST header passed to Client::request() to generate an absolute URI
• if the request URI is absolute, then ignore the HTTP_HOST header (as it now works)
• do the same with HTTPS server parameter

Profit:

• fix BC Break
• the documentation will be correct
  • http://symfony.com/doc/2.8/routing/hostname_pattern.html#testing-your-controllers
  • https://symfony.com/doc/2.8/testing.html#testing-configuration

Before:

$client->setServerParameters(['HTTP_HOST' => 'example.com']);
$client->request('GET', '/');
$this->assertEquals('http://example.com/', $client->getRequest()->getUri());

$client->request('GET', '/', [], [], ['HTTP_HOST' => 'example.com']);
$this->assertEquals('http://localhost/', $client->getRequest()->getUri());

Fixed (see last line):

$client->setServerParameters(['HTTP_HOST' => 'example.com']);
$client->request('GET', '/');
$this->assertEquals('http://example.com/', $client->getRequest()->getUri());

$client->request('GET', '/', [], [], ['HTTP_HOST' => 'example.com']);
$this->assertEquals('http://example.com/', $client->getRequest()->getUri());

[MathieuMindid]

Thanks for the fix.
As mentioned in #26257 the problems also exist in last version (3.x and 4.x)

[brizzz]

Yes, this fix (if its correct) should be merged to all maintained versions. I have edited description to clarify.

[nicolas-grekas]

I finally reviewed this PR, it makes sense to me!

[brizzz]

this is great news!

[fabpot]

Thank you @brizzz.

[Naktibalda]

It broke my tests, so it is a BC-break.
I was posting Host header to localhost to see if some header functionality is working.

[c33s]

@nicolas-grekas does this PR really make sense? this PR broke flexible testing of domain specific routes complete or is there a way to do this after the PR? see #32791