Re: [png-mng-implement] Fwd: [libpng] security improvements related to integral behavior (#9)

[John B. <joh...@gm...>]

On Mon, Aug 24, 2015 at 7:19 PM, Robert Seacord <rcs...@gm...> wrote:

>  I would imagine porting to C99 would be advisable.  Let's see, it is 2015
> - 1990 that makes 25 years.
>

Your point?  1.6 *requires *a C90 compiler, and has existed for some
time.  *C99
*dates from 1999/2000 (ISO C99, ANSI C00), so it's now almost 16 years old.

My rule of thumb, which has proven consistently right over the last 20
years, is that it takes 10 years for a software engineering standard to
obtain widespread acceptance and 15-20 years for the acceptance to be
complete (there will still be outlying cases though.)  Adoption of language
standards takes longest and adoption *as requirements* by portable widely
used libraries is at the very end of the timescale for, I hope, obvious
reasons.

>
> The C Committee is very good about posting copies of the standard. For
> example, C11 can be found here:
>
> http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf
>

That's a draft, from April 2012.  Can you post a link to the ISO *C99*
standard; the full text not a draft?  Last time I looked the ISO C
committee hadn't addressed this problem, but maybe I don't know where to
look.

I said:

> At some point libpng will probably move to require C99 and I had always
assumed that when
>it did stdint.h would get used, however after this discussion I'm inclined
to believe that the
>result would not be portable code with completely defined behavior and
that we are better
>off sticking with what we already do in pngconf.h for the 32-bit types.

You said:

> I'm not sure I agree.
>

Take a look at this machine:

http://docs.cray.com/books/004-2179-001/html-004-2179-001/rvc5mrwh.html

I don't see any way of writing *png_save_uint_32 *with *uint32_t *arguments,
however *png_uint_32 *arguments will work just fine.

John Bowler