Re: [gpsd-dev] ARM port problem in rtcm3.2 handling

[Eric S. Raymond]

Hal Murray <address@hidden>:
> address@hidden said:
> >> > I had been thinking that since fiexed width type (uint88_t, etc.)
> >> > are in C99 we should use those instead.  But I see in C99 7.1.1.1 3
> >> > that fixed width types are optional.  I don't recall that ever
> >> > causing us trouble. 
> >> So what do I do if I want a signed 8 bit int?  How would you have fixed 
> this
> >> if the ARM was wrong?
> 
> I just poked around and learned that there are "signed char"s.  Somehow, I 
> don't remember ever  running into them before.  That answers my question 
> above.

You're not being forgetful or oblivious.  It's pretty freakin' rare
for anyone to explicitly declare a signed char, so much so that I'm
not surprised you've never seen it done before.

> > They're optional in C99 but not in POSIX-2001 and descendants.  It's been at
> > least a decade since anyone shipped a C or C++ compiler without them. We're
> > considering them save for NTPsec; GPSD can, too. 
> 
> My take from this is that "char" without a modifier is evil.
> Unfortunately, it's used all over the place, starting with string
> literals.

Yeah, because it's not a problem there.  It only becomes an issue when the
char value is assigned to a short or int and might get sign-extended.

Good practice, which NTPsec aims to follow, is to (a) always and only
use char when the storage is semantically a char and never assigned to
or from a numeric scalar variable, (b) always and only use int8_t or
uint8_t when you want an 8-bit quantity to be interpreted as numeric,
and (c) never use "signed char" or "unsigned char".

(The 2 "signed char" exceptions are unit test framework code that's
not ours; there might be some remnant "unsigned char"s in ancient
header declarations, but I think I fixed them all.)

The GPSD code hasn't had quite as thorough a POSIX makeover, but the few
signed chars that are there have been static checked by both Coverity
and splint.

> Are there any static code checking tools that check for it?  If so,
>  what do they actually check for?

Possible sign-extension bugs. This is easy to do with elementary
assignment-graph coloring - both Coverity and Splint do it.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>