On 04/09/07 at 02:00 +0200, Pierre Habouzit wrote: > Packages that have a test-suite should always call it. Most > autotool-ed packages have a "check" target to run and that's it. It > would help tracking down regressions from the build logs, and would > avoid breakages like recently the grep one. > > Most of the sensitives packages in debian already run the testsuite > (libc, compilers, git, whatever...) and I can tell as a libc packager > that this already has been proven invaluable. I'm not sure that's a good idea. Some classes of packages that often provide test suites (e.g perl modules) usually enable them, and they are the cause of many build failures, because they often contain timing problems, or use some page on the internet during the test. Ian Jackson wrote autopkgtest as a framework to test packages. Typically, it should be used to enable such testsuites, and run them on a regular basis. Another advantage of this is that it would allow to find regressions in packages after they have been built, and not only at build time. So, a better release goal would probably be to enable testsuites for autopkgtest when the package already provides a testsuite. This is also less likely to break packages, since it wouldn't affect the build. But for this RG to be useful, you'd need to have regular runs of autopkgtest over the archive. Ian told at debconf that he was planning to setup something to do that. Ian, can we get an update on that? Would it be for Ubuntu, Debian, or both? -- | Lucas Nussbaum | lucas@lucas-nussbaum.net http://www.lucas-nussbaum.net/ | | jabber: lucas@nussbaum.fr GPG: 1024D/023B3F4F |
Attachment:
signature.asc
Description: Digital signature