bug#46670: 28.0.50; [feature/native-comp] possible miscompilation affecting lsp-mode

[Andrea Corallo]

Andrea Corallo via "Bug reports for GNU Emacs, the Swiss army knife of
text editors" <bug-gnu-emacs@gnu.org> writes:

[...]

> Hi Pip thanks for the patch,
>
> the approach of adding a cstr directly in the assume works for this case
> but is not generic as referencing there an mvar.
>
> The reason is that a later run of fw-prop after add-cstrs might be able
> to prove more precisely what the mvar is if the code was morphed in the
> meanwhile by some other pass.  This in contrast with adding a cstr that
> being "written into the stone" will stay as such no matter what.
>
> Admittedly ATM the only pass rewriting some code after assumes are
> placed and before the last fw-prop is run is 'tco' so this might be a
> real case only for functions affected by this, but in the future we may
> (and most likely want to) have more passes in that position of the
> compiler pipeline.
>
> So yeah I still prefer to general approach of keeping an mvar live till
> there and referencing it in the assume so that any future propagation
> within the SSA lattice can update this.
>
> Yesterday evening I did some work in that direction, doesn't look too
> invasive or complex.  I'll finish it this week as soon as I've some more
> time to put into.
>
> Thanks
>
>   Andrea

Right I've pushed bddd7a2d13 implementing the discussed solution as I'm
convinced is more general and future proof.

The patch is passing the tests and bootstrapping clean, also is adding a
test that should cover this specific bug.

Mauricio could you verify it actually solves the lsp-mode issue?

Thanks

  Andrea