Re: Build failure tooling improvements

[David Holland <dholland-pkgtech%netbsd.org@localhost>]

On Fri, Mar 21, 2025 at 11:48:07AM -0400, Greg Troxel wrote:
 > It sounds like you want
 > 
 >   The "smoke-test" target is conceptually similar to test, but is
 >   intended to take very little CPU time and real time to run.
 >   Generally, it should run in less than half of the time it takes to
 >   build the package.  Ideally, it will verify that the program starts
 >   and does something right, but it will not attempt to run test suites.
 > 
 > but:
 > 
 >   - it's a lot of work to write these
 >   - I don't think upstreams are going to want to add these
 >   - it won't work with cross builds
 >   - it will be testing the binaries in the build tree, using the
 >     package's mechanisms to find the built (and not installed shlibs)

Providing glue to run it in a VM would fix the last two problems.

(In order to have it be not super expensive, you'd want it to start
from a preexisting VM image, install just the package and the deps and
the test script, and run it...)

Still a lot of work, but it might be worth building out that glue.

It's not exactly uncommon for packages to build and not run, and we
tend not to notice until someone reports it.

-- 
David A. Holland
dholland%netbsd.org@localhost