On Fri, Oct 14, 2016 at 8:33 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Thu, Oct 13, 2016 at 9:23 PM, Peter Eisentraut
>> <peter.eisentraut@2ndquadrant.com> wrote:
>>> Well, the purpose of make all or make world is to build the things that
>>> are to be installed by make install or make install-world, which is the
>>> stuff that users want to use.
>
>> Well, I think what Tom, Andres, Michael, and I are saying is precisely
>> that we should review this approach and revise it so that 'make world'
>> builds everything. Or else add 'make universe' to really build
>> everything, but personally I think that's an unnecessary complication.
>
> Not sure. There's something to be said for the equivalence Peter
> proposes above. What you actually wanted, as I understood it, was
> that "make world" plus "make check-world" should test absolutely
> everything. I don't have a problem with the idea that some bits
> of test scaffolding don't get built until you do "make check-world".
> The real problem here is that "make check-world" missed some tests,
> which Peter agrees is a bug.
No, the problem is that worker_spi has no tests, so 'make check-world'
never tries to build it at all. Something in the buildfarm does cause
it to get built, though.
> Andres complained about insufficient parallelism in the "check"
> target, but that seems to me to be something to address separately;
> redefining the set of actions to be taken is not the way to fix that.
Sure, that's another issue.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company