Re: gplot.txt

[Daniel J Sebald]

On 03/07/2016 01:31 PM, Rik wrote:
> Ben, Dan, Lachlan,
>
> I checked in this cset
> http://hg.savannah.gnu.org/hgweb/octave/rev/9f7282acc30f to resolve the
> segfaults with speye on Mac platforms.
>
> I made the minimum fix necessary: removing the use of the OCTAVE_NORETURN
> attribute in a single instance and adding a useless return statement to the
> previously decorated function in order to suppress the compilation
> warning.  If further problems develop then we should take the other tack
> and explicitly distinguish between GCC and clang in the macros that test
> for GCC.

I'm fine with that.  As I said in a previous post with a different 
thread title, the whole non-return clarification seems very special use. 
 I could imagine it used for, say, some special hardware-specific 
function one writes in assembly code to shutdown an embedded system. 
But using non-return for general compilation of C++ commands doesn't 
seem standard-portable.

The structure of Octave is to assume processing proceeds as expected, 
it's just that along the way errors happen and the code needs to abort 
the processing.  Need to devise a scheme for doing that, whether that be 
a series of return flags that indicate error/no-error (laborious), what 
I wrote about redefining new/delete (behind the scenes and has some nice 
attributes), global variables (eh, avoid global variables), signals 
(non-linux doesn't have them), in any case something simple and 
convenient.  The no-return classification is difficult to keep track of, 
especially if it gets several calls deep (Does a function that possibly 
calls an OCTAVE_NORETURN also get the classification OCTAVE_NORETURN?).

Dan