From: Landry Breuil <landry@openbsd.org>
Subject: Re: [new] databases/arrow 18.0.0
To: ports@openbsd.org
Date: Fri, 1 Nov 2024 19:17:31 +0100

Le Fri, Nov 01, 2024 at 10:10:35AM +0000, Stuart Henderson a écrit :
> On 2024/11/01 10:56, Landry Breuil wrote:
> > hi,
> > 
> > following thrift, here's the port for the c++ part of arrow:
> > https://github.com/apache/arrow/blob/main/cpp/README.md
> > it provides the parquet library for https://parquet.apache.org/.
> > 
> > some open questions:
> > - i've put the port in databases because for me its sort-of a database
> >   format: "The universal columnar format and multi-language toolbox for
> > fast data interchange and in-memory analytics"
> > 
> > but it can go into devel or textproc, i'm not settled on it. devel is
> > already a bit crowded...
> 
> databases sounds good
> 
> > - the toplevel in https://github.com/apache/arrow/ has zero build goo,
> >   so from the same distfile one has to build by subdir (eg setting
> > WRKDIST=${WRKDIR}/${DISTNAME}/cpp), hence the pkgname being arrow-cpp
> > since i'm only interested in the c++ part.
> 
> shouldn't that be WRKSRC=${WRKDIST}/cpp?

yes, i'm always confused by the various variations..

> > should i name the port databases/arrow-cpp ? databases/arrow/cpp in
> > preparation for potential other ports for various bindings ?
> 
> databases/arrow/cpp sounds a good plan to me. common parts can be
> factored in Makefile.inc later when we find out what the common parts
> are :)

here's a new version that:
- enables json support via textproc/rapidjson
  (CXXFLAGS=-I/usr/local/include was the missing key so that cmake finds
rapidjson headers)
- enables building tools & tests, lots of tests run fine:
89% tests passed, 9 tests failed out of 81
(and i just realized some of the parquet test failures are only because
i forgot to set PARQUET_TEST_DATA in the env)

only the threading test needs to be manually killed, otherwise it
infinitely gobbles cpu, and i haven't investigated it yet.

the process test is patched out because it includes a boost header that
tries to include a non-existent wordexp.h, which will be fixed in the
next boost update (per https://github.com/boostorg/process/issues/391)

i @comment'ed out all the testing files from PLIST so that there was
no LIB_DEPENDS on gtest/boost, as the tests are run from objdir fine.

Landry