[postgis-devel] Enabling wagyu by default

Raúl Marín Rodríguez rmrodriguez at carto.com
Fri May 31 05:31:10 PDT 2019


Hi,

> This means that if libfoo is optional, then the user can pass
> --enable-foo to use it (and fail the build if not present) or
> --disable-foo (to ignore it even if present).  That requires a bit more
> thought, but I think the reduction in un-understood side effects is
> worth it.

> Here, I think you are talking about changing a default (which is not
> objectionable), and there won't be any kind of different outcome based
> on what is installed.

Yes, that's how it is setup right now (`--with-wagyu` to enable) and
the proposal
is to change the default.


> Presumably this does not change the required version of C++, and would
> not cause postgis to fail to build anyplace it builds now.  Or I am
> guessing that is a hypothesis and it is of course not really known.

This introduces a new requirement as you'll need C++11 to build with
this option.
Beforehand, you only needed a C99 compiler.

My current guess is that any platform that was able to build postgis
should be able
to build also with wagyu as a C++11 compiler was already a requirement
for other dependencies (GEOS, GDAL...), but as it is introducing the
direct dependency
I'd like to have test it in the while during the alpha period.

> Can you open some tickets on the GEOS trac for these? I'd like to get the GEOS implementation improved, but as far as I know there are no open bugs. (Not claiming that the bugs don't exist, only that I can't fix something I don't know about.)

Although I was blaming GEOS, the issues were on the assumptions that the
postgis code was making and can't be grouped into 2:
- Issues related to clipping (with clipbyRect) without making sure the
polygon is valid:
This is a known GEOS limitation, but calling makeValid beforehand was expensive.
- Issues related to MakeValid and double to int  transformation. An
invalid polygon with
integer coordinates would become a valid one with doubles, then
changing the doubles
to ints would make it invalid, and so on.

So, they aren't GEOS issues but MVT special cases. In any case all have been
already fixed (at the cost of performance).

Regards,


--
Raúl Marín Rodríguez
carto.com


More information about the postgis-devel mailing list