Re: [Bug-gnulib] module maintainers

[Bruno Haible]

Jeff Bailey writes:

> I still have it on my private todo list to begin writing a small
> testsuite for gnulib

Excellent! What shall it test? That things compile and provide the
declared function? Or a real extensive testsuite with functionality
tests?

> If a testcase and a patch were applied, if someone disagreed with
> the change, they could at least hack until the testsuites all passed
> again.

It shouldn't be the guy who writes a testsuite who corrects the lib/*
code until the testsuite passes. This is based on my experience:

  - When someone sends me a test and a patch for gettext, or when I
    send a test and a patch for glibc, the test is OK with 90%
    probability, but the patch is OK with only 50% probability.

  - People who write testsuites sometimes write wrong tests, and
    need to be convinced by the maintainers of the implementing code
    that their test is broken. (Speaking from experience with the
    two ANSI Common Lisp test suites.)

Therefore it's best if the testsuite guy, when making new tests, only
_proposes_ a patch, but leaves the decision (1. whether his test is
testing the correct behaviour, 2. whether his patch is OK) to the
maintainer.

Bruno