[fetchmail-devel] Code quality - signedness

[Matthias A. <ma...@dt...>]

Dear fellow fetchmail interested,

I have recently compiled fetchmail with a decent set of warnings enabled
(i. e. -pedantic -W -Wextra and some more on gcc 3.3 or 3.4 or icc 8.0)
and I am _flooded_ by complaints about signedness mismatch. It is a bit
unfortunate this has been handled so laxly in the past, and this will
need to be fixed before a next release we can seriously recommend. It is
a task that will take a longer time because we'll effectively need to
align the types throughout the file with respect to signedness. Getting
the interfaces and data structures cleaned up would be the first step
and would require more eyeballs than just mine, and the second step
would then to cast the interface changes into .h files. After that, the
implementation cleanup would begin as a third step. It is a long-winded
effort, and tools I've come across in the past that can help us are:

- GCC with lots of warnings enabled, the more, the better.

Particularly useful to find ctype.h madness: GCC on Solaris 8, where in
32- or 64-bit mode (I don't recall as I usually test both) the ctype.h
is[]* functions are implemented as arrays, and GCC will complain if a
vanilla char is used as argument in is*() which leads to bogus results
on 8-bit data.

- ICC (Intel C++) 8.0. It's free for personal use, but a registration
  with Intel is required.

- Splint. It is probably a tad too early even with -weak to start with
  Splint.

Opinions solicited.

-- 
Matthias Andree

Encrypted mail welcome: my GnuPG key ID is 0x052E7D95 (PGP/MIME preferred)