Cyclic dependency: perl-HTML-Parser <-> perl-HTTP-Message

[r-ricci]

perl-HTTP-Message is a check dependency of perl-HTML-Parser,
perl-HTML-Parser is a runtime and build dependency of perl-HTTP-Message.

When updating both packages (https://github.com/4ricci/void-packages/pull/31), ./xbps-src -Q pkg perl-HTTP-Message returns the error:

=> ERROR: Build-time cyclic dependency perl-HTML-Parser(host), perl-HTTP-Message(host), perl-HTML-Parser(host) detected.

./xbps-src -Q pkg perl-HTML-Parser returns the error:

=> ERROR: Build-time cyclic dependency perl-HTTP-Message(host), perl-HTML-Parser(host), perl-HTTP-Message(host) detected.

[ericonr]

I think the correct path forward would be disabling tests for perl-HTML-Parser (or see if they can run without perl-HTTP-Message. Otherwise this means bootstrap is broken when tests are enabled, and we want to make bootstrap with tests possible.

[Chocimier]

On the contrary, I think cycles due to checkdepends are better than not testing package at all. In case like here, one can first build package(s) locally without tests, than rebuild with tests enabled.

[dkwo]

I also had to deal with the same issue here #32229

[q66]

i think not testing is better because if tests are not cyclic, they can be properly automated (we could even have the builders run them eventually) while if you allow cycles they can't (and tests on stuff like this are mostly meaningless either way, i see no reason to let a few rare cases break it)

[q66]

also i think checkdepends are in general a poor idea because it means that your build environment differs if you're running tests vs if you're not

while ideally the build environment should only ever be affected by build options

[dkwo]

Then, shall we treat checkdepends just as (host)makedepends, regardless of whether we are permorming tests or not?