Re: [boost] Exception specifications in smart_ptr.hpp

From: Lois Goldthwaite (loisg_at_[hidden])
Date: 2000-07-24 07:05:47

Well, I ought to admit up front that I only contributed to this thread because it's fun to
watch Dave Abrahams jump up and down when anyone mentions empty throw specs.  :-)  [1]

The compiler was IBM's Visual Age C++. According to the tests we ran (using help from one of
its developers), if  "throw()" was specified VAC++ totally eliminated the try and catch
overhead which was otherwise generated. This is not to suggest that other compilers can't do
it also, nor that VAC++ might always generate optimized code. The tests were too limited to
be definitive (that's why I hesitated to name names in my original post), but they did rebut
the assumption that empty throw specs _always_ pessimize the generated code.

Lois

[1] BTW, my own hot button is reference-counted strings -- a sure route to performance
quagmire and poor robustness.

Mark Borgerding wrote:

> Valentin Bonnard wrote:
>
> > Mark Borgerding wrote:
> >
> > > Since at least one compilers *is* apparently capable of using a throw() to their
> > > benefit
> >
> > What kind of benefits ? Time ? Space ? Both ?
> >
> > ``at least one compilers *is* apparently'' is pretty vague.
> >
>
> I was responding in part to a post (last Sunday -- I had to dig)  made by Lois
> Goldthwaite, who wrote:
>
>         Some of our tests in Hawaii showed that at least one compiler was able to use the
>
>         "throw()" specification to generate more optimized code, not less.
>
> Any questions regarding which compiler and which optimizations should be directed to
> Lois.
>
> ------------------------------------------------------------------------
> Learn Wireless Development Now!
> http://click.egroups.com/1/6356/4/_/9351/_/964214451/
> ------------------------------------------------------------------------