Robert Elz <kre%munnari.OZ.AU@localhost> writes: > Date: Fri, 22 Sep 2017 20:07:25 -0500 > From: Jason Bacon <bacon4000%gmail.com@localhost> > Message-ID: <8b21c031-4f7c-6f25-7bee-4b3959fdf1cf%gmail.com@localhost> > > | So there's nothing like the FreeBSD ports PACKAGE_BUILDING variable, > | which is automatically set during a binary package build (by poudriere, > | I believe)? > > As I understand things, in pkgsrc, all builds are binary package builds. > When you (or any user) builds a pkg from source to install it, pkgsrc > makes the binary package, and then installs that. > > The only real differences from a pbulk build is how many packages are being > built at about the same time, whether they actually end up installed in > a place where normal users of the system will see them, and what happens > to the binary package files that are constructed along the way. > > Doing things this way has lots of advantages, not least ensuring that > some who (without changing the build options) builds from source gets > the exact same thing installed, in the exact same way, as someone who > downloads and installs a binary package. That is, if you want different > you have to explicitly configure it. Agreed. I think as that as a design goal a manual build (without setting options) and a bulk build should produce the same bits (exactly, in the glorious reproducible build future). I am unclear on Jason wants to get a different result for the standard pbulk build that leads to distributed packages. I think the big issue here is when it's ok for a package to be built in such a way that it's either tuned for on only works right on the same CPU that it's built on, rather than being targetted for the standard ABI and CPU of the system. Basically, I think a random user doing "make package" should get a package that is not locally bound.
Description: PGP signature