Re: autorescan?

[Keith Marshall]

On Wednesday 02 July 2008 06:42:50 Ralf Wildenhues wrote:
> * Tim Post wrote on Wed, Jul 02, 2008 at 07:39:01AM CEST:
> > I've been including most checks that autoscan suggests. My
> > reasoning is that someone down the line might need those checks
> > and may not be able to autoreconf in order to include them.
> > Probably bad practice ...
>
> Not necessarily.  But likely it leads to more configure tests than
> are really needed.

In some cases, it may actually be harmful.  For example, I'm 
autoconfiscating a package as a first step towards getting it to 
build for MS-Windows, but I'm actually doing my build with a mingw32 
cross, hosted on GNU/Linux.  autoscan suggests that I should include
AC_FUNC_MALLOC and AC_FUNC_REALLOC; if I blindly follow that advice, 
then I have to provide implementations for rpl_malloc() and for 
rpl_realloc().  However, the application I'm working on isn't 
actually vulnerable to the issue those macros test, so there really 
is no reason to go to this length.

Ok, you may say, "why not just include the tests anyway; what harm can 
they do?"  Well, they actually break the build when cross-compiling, 
because neither is cross-compile capable, and will force the build to 
rely on the replacement functions, even when the targetted host has 
malloc() and realloc() implementations which would pass the configure 
tests, if configured natively.

The moral of this is that blindly including everything autoscan 
suggests may not be wise; you should give careful consideration to 
whether the suggested tests are really necessary.

Regards,
Keith.