On Sun, 2007-05-06 at 23:56 +0200, Peter J. Holzer wrote:
> > Just check once during 'make test'. If it were a GNU package and not
> > perl, the check would be done during ./configure.
>
> The environment isn't necessarily the same during "make test" and
> runtime. An extreme case is an RPM or Debian package: The make test is
Debian has a policy manual. I bet you will find that LANG is covered
(see my remarks about google earlier in the thread). I expect most
deriviatives like Ubuntu do not deliberately go about trying to hurt
themselves.
> run on the packagers machine, and then the package is installed on lots
> of different machines with different environments.
That might be more of a problem with rpm because I would expect there to
be more variation between platforms.
>
> So, if this is tested at all (and not just written into the
Well ... that is another problem ;-).
> documentation), it should be tested at least once before it is used. But
> as I wrote in the next paragraph, it doesn't have to be tested each time
> rfc822_date is called - the result can be cached.
Ugh. I bet Wietse just uses strftime.
$ cd tmp
$ apt-get source postfix
$ find -name '*.c' -exec grep 'strftime' '{}' \; | wc
7 46 454
There. I win :-)
The quickest way to check whether this is a real issue is to see what
apr does ...
--
--gh
