Re: Question about iterator management in sw/source/core/fields/cellfml.cxx

On 03/02/12 14:01, Stephan Bergmann wrote:
> On 02/02/2012 09:08 PM, julien2412 wrote:
> > Would this patch better ? (I kept the for loop)
>
> Unfortunately that still has a problem.  After "rBoxes.erase(toErase)", 
> "it" (which is the same as "toErase") is invalidated, so incrementing it 
> (up in the for(...;...;...) part) has undefined behavior.
>
> The "standard idiom" is
>
>    for (iterator i = m.begin(); i != m.end();) {
>      if (doErase) {
>        m.erase(i++);
>      } else {
>        ++i;
>      }
>    }

but doesn't that have the same problem? "i" is incremented only after
the erase is complete, when "i" is already invalid.

shouldn't this be something like:

    for (iterator i = m.begin(); i != m.end(); ) {
        if (doErase) {
           iterator const j = i++;
           m.erase(j);
        } else {
           ++i;
        }
    }

> > I read that to erase a position on a iterator invalidate this iterator from
> > position to the end
> > (http://www.cplusplus.com/reference/stl/vector/erase/)
> > I don't know if to create this temporary variable "toErase" change something
> > or it's all the same.
>
> The text you cite is about vector, for which *all* iterators into a 
> vector are invalidated upon an erase.

i think i read this "invalidate from position to end" in the SGI
documentation as well; does the actual standard give implementations
more freedom here?