[sane-devel] Internal getopt opinions

[Chris Bagwell]

Hi all,

I'm trying to improve the way I modified the getopt.c to always be 
compiled... It has a problem, as pointed out by  Mattias, that it 
requires some ordering issues with config.h.  I'm seeking opinions 
before proceeding just to be safe.

In case people didn't catch it, I've modified lib/getopt*.c to *always* 
compile now on any platform and have it do some #define magic to prefix 
it with sanei_ (as not to conflict with platform versions).  The #define 
magic has issues with platforms that have getopt defined in stdio.h 
though and requires moving stdio.h before #include "config.h"... There 
are some things in config.h that make having any #include's before 
"config.h" a bad idea.  So I have a circular logic bug I need to break.

Fixes boil down to 2 options:

1) Go back to conditionally compiling getopt*.c but to do this right, I 
need to also rename include/getopt.h.  Right now, there is no way to 
include platform's <getopt.h> because of -L../include and there are 
conflicting prototype issues in todays <getopt.h> compared to what can 
be linked in from platform.  Pros:  This approach aligns with lalloca.h, 
lassert.h, and _stdint.h.  Cons: a) To keep CVS history, someone would 
need access to backend of CVS server. b) it wasn't correctly 
conditionally compiling before for Solaris/Darwin and I'll need to 
create some new configure.in logic.

2) Continue to always compile getopt*.c files and update the header 
file.  Instead of using #define magic to add sanei_ prefixes, manually 
change the prototypes to include it.  Thats around 5 variables and 5 
functions in header to add sanei_ to and 5 #define's added to each 
source file.  Pros: No configure support to maintain for various 
platforms.  Cons: Its deviating alot from standard getopt* files and 
will be harder to port any updates.

Chris