Re: warnings in unit tests

[Bruno Haible <bruno@clisp.org>]

Bernhard Voelker wrote:
> One little aspect of the tests code is that people might look (also) there to
> learn how to use a certain gnulib module, and then copy/paste the code from
> there into their projects.

Yes, a secondary value of the unit tests is to show how the APIs can be
correctly used. (For example, the unit tests of the 'list' and 'hamt' modules.)

> That's at least true for the good test cases, surely not for those provoking errors.
> This means it may be worthwhile to have at least the good test cases in a warning-
> free shape (which I think it most often already is).

Yes, the parts of the unit tests that exercise the normal use of an API should
normally compile without warnings with '-Wall'.

The parts that exercise corner cases (NULL pointer accesses, invalid arguments,
endless loops, endless recursions, etc.), on the other hand, can produce warnings,
from compilers and from static analysis tools.

That's one reason why we cannot tolerate '-Werror' on gnulib tests.

The other reason is that every package maintainer has their preferred set of
warnings — that's what the 'manywarnings' module is made for —, but it does
not make sense for package maintainers to enforce the absence of certain
warnings on code that 1) they don't maintain, 2) does not end up in the
binaries produced (installed) by their package.

Bruno