On Mon, Mar 17, 2014 at 5:32 PM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
> On 03/17/2014 07:49 AM, Michael Paquier wrote:
>>
>> On Fri, Mar 14, 2014 at 8:02 PM, Heikki Linnakangas
>> <hlinnakangas@vmware.com> wrote:
>>>
>>> Hmm. The regression test Makefile really ought to pick up the same
>>> settings
>>> we use for the main makefile, so that if you do "./configure
>>> --with-unixodbc=...", the regression tests are automatically built
>>> against
>>> the same library.
>>>
>>> I'm not sure what's the best way to achieve that. The regression Makefile
>>> is
>>> currently completely separate from the automake system. We could add
>>> "SUBDIRS=test" into Makefile.am, and then you could do "make
>>> installcheck"
>>> from the top directory to run the regressions, and you could use the
>>> LDFLAGS
>>> and other variables set by automake.
>>>
>>> A problem with that is that the regression suite Makefile depends on
>>> PostgreSQL's pg_config to find the pg_regress program. I guess the proper
>>> solution would be to add an optional configure flag to provide a path to
>>> pg_config. If pg_config is not found, then you couldn't run the
>>> regression
>>> tests, but you could still build the driver without the PostgreSQL header
>>> files etc.
>>
>>
>> Even with that, don't we need first to move the automake process
>> currently done on ./Makefile.am to another file? Like let's say
>> Makefile.global.am, which contains all the variables set by automake.
>> Then we create a new file ./Makefile at the root folder that includes
>> Makefile.global. Something similar should be done with test/Makefile
>> (inclusion of ./Makefile.global) and it would be able to use the
>> values of CFLAGS, CDFLAGS and even PG_CONFIG that configure has set.
>
>
> Hmm, yes, something like that would be good. We can't easily restructure the
> main Makefile like that, though, because it's generated by automake. But we
> can still create a new Makefile.global file, with the same CFLAGS/LDFLAGS
> that the generated Makefile contains.
As the code tree of odbc gets more complex, perhaps it might be worth
doing it... Your patch seems to be enough for now though.
> Actually, we could just move test/Makefile to test/Makefile.in, and let
> configure set the LDFLAGS/CFLAGS variables directly in test/Makefile.
>
> I pushed a patch to do that. It doesn't do anything automatic about
> PG_CONFIG; it would still be nice to add a configure flag for that.
configure already sets PG_CONFIG, so we could do like in the patch
attached and add a flag in test/Makefile.in similarly to CFLAGS and
LDFLAGS.
> PS. Have you set up a Windows build environment? I don't think anyone's put
> any effort in running the regression tests on Windows yet. If we could make
> that work somehow, that would be nice.
No, I don't have that, but it is on my TODO list. I don't know
precisely when I will be able to come to that, and I am still in the
mood of giving a better test coverage first, like tests for catalog
functions and deprecated functions.
--
Michael