Date: Sun, 27 Feb 2000 17:53:38 -0500
From: Harlan Stenn <[EMAIL PROTECTED]>
The biggest reason NTP is going for correct prototypes is that the NTP code
is kinda widely used, and we make sure the code will compile on a number of
C compilers.
I'm tolerably familiar with the problems. My UUCP code compiles on
everything I know of from SVR2 on up.
We also like to crank up as many warnings as we can stand, and do our best
to make the code "correct".
I do that during development, but not in shipping code. In fact, my
standard compilation options during development are
-O2 -pedantic -ansi -Wall -Wshadow -Wpointer-arith -Wcast-align -Wwrite-strings
-Wconversion -Wmissing-prototypes -Wnested-externs -Werror -g
But that only works on systems with proper header files. On other
systems I'm willing to accept a few declarations without complete
prototypes.
There are a number of "longian" problems out there, and we have to take some
pains to make sure that network packets are the right length.
I guess I'm not sure what a longian problem is. I get network packets
right by building them in arrays of unsigned char. Otherwise you have
to worry about precise type sizes and alignments, and I don't think
it's worth it. (Well, it's worth it when you not trying for
portability and can assume gcc; then you can use gcc extensions to
control them yourself.)
Ian
