[guus@debian.org: Bug#290306: dbus-1-dev: silly typedefs]

[Havoc Pennington]

On Thu, 2005-01-13 at 21:19 +0100, Guus Sliepen wrote:
> On Thu, Jan 13, 2005 at 02:31:37PM -0500, Havoc Pennington wrote:
> 
> > > It also introduces its own types, while there are perfectly well defined
> > > and standardised types that do exactly what dbus needs in stdint.h
> > > (int16_t, uint16_t, etc.) and stdbool.h (bool). At best DBus's
> > > proprietary typedefs will confuse a programmer, and in the worst case it
> > > will break on other architectures.
> > 
> > These aren't portable enough yet. Programmers can assume that uint16_t
> > is the same as dbus_uint16_t though.
> 
> What do you mean, they aren't portable enough? They are MADE to be
> portable! You can check for the availability of these types with
> autoconf. If you compile on a system that doesn't define them (I think
> you have to look very hard for those), then you can always typedef them
> yourself as a last resort. I have more trust in the standard header
> files getting the typedefs right on all architectures than anyone else.
> 

The "emulate glibc if it's not there" approach isn't used much in
desktop software, we usually use the "portable wrappers" approach as
with Qt and GLib. I don't think that debate is very on-topic for this
list, but it extends well beyond typedefs. See dbus-sysdeps.h, or all of
GLib/Qt

Even outside of software, see vsftpd for example and their secure coding
guidelines which include wrapping all system headers/functions as a
guideline.

Havoc