Re: pbulk best practice for fixing packages that fail to build

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2015-04-13 at 18:06 BST, J. Lewis Muir wrote:

> The closest I could think of was to create a sandbox, enter it, extract
> the bootstrap tarball, become the pbulk user, and configure the PATH
> environment variable so that the extracted pkg_info, etc. binaries will
> be found.

I have a script which does all this for me automatically.  Once you
have all the necessary infrastructure set up it's pretty
straight-forward and basically just runs the steps you've enumerated.

> But when I run "bmake package", it tries to rebuild all
> dependencies rather than using the existing packages that have already
> been built.

As part of my script it sets DEPENDS_TARGET=bin-install in the
environment, and mk.conf is already configured with the PACKAGES
location, so that's taken care of.

> Or do you generally treat pbulk as a black box and try to fix package
> build failures *outside* of the sandbox?  I can't think of a good way
> to do this without having to rebuild all of the package's dependencies.

One of the advantages of pbulk is that it tests within a clean
environment, and this often exposes packages which do not list their
correct dependencies, and happen to work outside of a sandbox simply
because those dependencies happen to be installed.

So no, I never build packages outside of a sandbox.

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com