TEST_DEPENDS in all-depends-list et al.?

[lists at tcm.yi.org]

On Mon, Jan 18, 2016 at 01:55:35PM +1100, Kubilay Kocak wrote:
> On 18/01/2016 7:11 AM, lists at tcm.yi.org wrote:
> > ports/head from around 2015-12-18 had about 15(?) dependencies in
> > www/py-requests for example. Now it wants to build over 90!
> > 
> > How can I turn this off? Is this a bug?
> 
> By definition all-depends means (and should mean) literally all
> dependencies (*_DEPENDS), so in that regard it makes sense.

I'm not sure what exactly changed between 2015-12-18 and now.
Specifically, www/py-requests' Makefile last changed in October,
but from December to today its all-depends-list grew from
~10 to 90.

Mk/Scripts/depends-list.sh also last changed outside the time
frame in question.


> Having said that, given the special nature of the test stage (orthogonal
> and independent to build/run dependencies), it sounds reasonable to ask
> whether they should be included in the target (all-depends-list) that
> has otherwise always only meant "what things does this port need to 'work'".
> 
> My personal opinion is that all-depends-list is fine, and a less
> encompassing <something-else>-depends-list that only shows actual
> end-user functional dependencies is needed.

My problem isn't exactly with all-depends-list, but rather with
showconfig-recursive and package-recursive, whose dependency list grew
as well, since they use the same dependency script. I don't want
to build, package or configure TEST_DEPENDS ports in the course of
normal package building.


> This question is also relevant for the case of OPTION'al dependencies
> (they're not included in all-depends by default unless they're
> inOPTIONS_DEFAULT too), and would *also* apply if ports/pkg's had (or
> ever will have) any notion of 'suggested/recommended' but otherwise
> non-compulsory dependencies like other packaging systems.

I'm perfectly fine with not dealing with ports that I have
excluded or that are excluded by default. This is what I would
expect.

For the time being, I have changed this line in Mk/bsd.port.mk

_UNIFIED_DEPENDS=${PKG_DEPENDS} ${EXTRACT_DEPENDS} ${PATCH_DEPENDS} ${FETCH_DEPENDS} ${BUILD_DEPENDS} ${LIB_DEPENDS} ${RUN_DEPENDS} ${TEST_DEPENDS}

by removing the ${TEST_DEPENDS} variable. Now I get the same
behaviour as before, but clearly, this is a horrible kludge.

Mk/bsd.port.mk also had no relevant changes that I can see
between 2015-12-18 and now.

So I'm still not sure if this is a bug or not. Is the
package-recursive target really meant to pull in all
TEST_DEPENDS ports? I was under the impression that
tests are exclusively run via 'make test' in the ports tree,
making it completely irrelevant to binary packages.

Thanks and best regards.