🚀 go-pugleaf

In <1992Oct30.174207.22464@texhrc.uucp>, pyeatt@Texaco.com (Larry D. Pyeatt) wrote:
> In article <mwm.2gwf@contessa.palo-alto.ca.us>, mwm@contessa.palo-alto.ca.us (Mike Meyer) writes:
[Discussing OOB array references, ala:]
> 
> oops, I got to thinking in Pascal mode and didn't quite make the switch.
>
> #include <stdlib.h>
> #include <stdio.h>
> typedef float mytype[26];
> main()
> { mytype mine;
>   printf("%f\n",mine['a']);
> }
> 
> |> You need to get a better compiler. Mine responds to the above with the
> |> message:
> |> 
> |> test.c 7 Warning 83: reference beyond object size
> |> 
> |> 	<mike
> 
> What compiler are you using?

SAS/C, any version. The earlier ones tended to incorrectly flag
calculations that resulted in OOB addresses, even though they didn't
dereference them.

> I just tried it on 6 different C compilers.  None of them gave any sort of
> warning.  C does not do range checking.  If your C compiler complains, then
> it is not a standard C compiler. 

Nope, it _is_ standard C. It just isn't standard Unix C behavior.  The
ANSI standard tags references to invalid addresses as having
"undefined behavior". This means poor compilers can ignore them, and
let you take your chances with your computers run-time system. Good
ones are allowed to issue compile-time warnings.

> Not only does it compile, it also runs
> without reporting a run time error.

Again, the ANSI standard allows compilers to catch such things at run
time, and "[behave in] a documented manner characteristic of the
environment." For Unix, this would be dumping core or giving wrong
answers.

With my compiler, I can get anything from the wrong answers to a
run-time requester reporting the problem and offering to take dump the
process image before terminating the process.

> No language of the Pascal family would
> let such an obvious error slip by unless the programmer specifically requested
> it.

I've played with Pascal implementation that were poor enoungh to do
so. Your average Pascal implementation wouldn't do that, though. And
yes, your average C implementation would.

> See if you get a warning on this piece of code:
> 
> #include <stdlib.h>
> #include <stdio.h>
> typedef float mytype[26];
> main()
> { mytype mine;
>   char tm = `a`;
>   printf("%f\n",mine[tm]);
> }

Well, I get two errors for the "`"'s (they should be "'"'s). After
fixing that, I get a warning that mine is uninitialized.  That's an
error your average Pascal compiler probably won't catch (at least,
they wouldn't when I was using them).

These are all questions of "quality of implementation". You need to
get a better compiler.

	<mike