Re: -Wabi is misdesigned (answers the wrong question)

[Michael Matz <matz at suse dot de>]

Hi,

On Wed, 22 Oct 2003, Joe Buck wrote:

> I think that -Wabi as currently implemented is useless.

I 100% agree, given the 300+ warnings per compilation unit in KDE.

> The picture I had in mind when we first talked about it was this:
> if -Wabi is on, we would compute the layout of a class using the
> 3.2 algorithm and the ABI-compliant (modulo unknown bugs) algorithm,
> and object iff there is a difference.

So, if that's too complicated to implement, we should simply make -Wabi a
nop (for compat) and in the future abstain from implementing such warning
(IIRC Mark already was against it in the first hand because it basically
produced too many false positives.)

OTOH I was told (I don't know if it's correct, and don't have the energy
to verify it) that for instance the layout for implicit virtual
destructors will _definitely_ change with the corrected ABI, and hence
with 3.4.  An implicit vdtor is not a corner case in the context of KDE,
so this will definitely make KDE non backward-compatible.  (But as I said,
as long as the standard lib is not compatible that's theoretic babble
anyway).

> However, -Wabi as implemented doesn't do that: it objects to any
> usage that potentially *might* tweak an algorithm difference.


Ciao,
Michael.