Re: Correct byte compiler error/warning positions. The solution!

[Alan Mackenzie]

Hello, Andrea.

On Wed, Dec 01, 2021 at 16:18:48 +0000, Andrea Corallo wrote:
> Andrea Corallo <akrl@sdf.org> writes:

> > Hi Alan,

> > I think would be interesting to have a comparative run also using
> > <https://elpa.gnu.org/packages/elisp-benchmarks.html>

> >   Andrea

> Adding on this,

> given we should very carefully evaluate the performance impact (one
> single benchmark is certainly not sufficient), I think we should stay
> away for solutions adding performance cost to the run-time.

Any help evaluating the performance of the scratch/correct-warning-pos
branch would certainly be appreciated.

> There's no question we'll have to pay a cost for this, but the other
> solutions on the table (hashing conses or fat conses) are impacting only
> the compile time and therefore IMO should definitely be preferred.

One great advantage of the current approach is that it's implemented
working code.  Well, almost (it doesn't yet work with native-compilation
enabled).

The disadvantage in performance, as measured so far by me, is so small
(less than 1%) as to be negligible.

The bug reports about warning positions being wrong have remained
unresolved for many years.  It is surely now time to fix them with
scratch/correct-warning-pos when it becomes stable.

If there come to be better ways of addressing the problem in the future,
then one of those ways could supersede the current proposal when it
becomes available.

> Regards

>   Andrea

-- 
Alan Mackenzie (Nuremberg, Germany).