Re: zstd headers?

[Martin Husemann <martin%duskware.de@localhost>]

On Sun, Apr 26, 2026 at 07:42:50PM -0400, Thor Lancelot Simon wrote:
> On Sun, Apr 26, 2026 at 09:18:39PM +1000, Simon Burge wrote:
> > 
> > I asked about this a while ago [1] and got a response off-list that we
> > don't want to expose the base-system zstd because that could conflict
> > with pkgsrc zstd, especially if pkgsrc has a newer version.
> 
> This seems very wrong.  Is core on board with hamstringing the base NetBSD
> operating system to ease dependency management for pkgsrc?

I am not sure I would use that term.

Generally it is not about ease of dependency management and not restricted
to pkgsrc, but about random 3rd party programs compiled on NetBSD running
into strange runtime or (if lucky) configure time issues.

Historically we had lots trouble when trying to reuse e.g. sqlite from
base in pkgsrc. Sqlite has dozens of compile time options enabling/
disabling support for features. Some of them could easily be added in
the base system, others require external libraries we do not want to
include (or that have less loved licenses). So (AFIR) we ended up
not using base sqlite anywhere in pkgsrc (unless configure stuff goes
wrong/plays dirty tricks).

For zstd things look better, the only relevant external library (that we do
not yet have in base) would be lz4 and it is 2-clause-bsd.
And zstd seems to be relatively slow moving, with a roughly yearly release
cycle.

So besides hiding it fully as Simon mentioned, the other option would be to
add liblz4 and set HAVE_LZ4 in the zstd build, plus publishing the headers.

Martin