[PATCH] Fix debug info for COMMON and EQUIVALENCEd variables (regression from 3.4.x)

[Richard Henderson]

On Tue, Oct 11, 2005 at 11:25:15AM -0400, Jakub Jelinek wrote:
> But for SAVEd EQUIVALENCEs which are function local static unions
> the debug info for the variables should be only emitted if the
> EQUIVALENCE has been/is going to be emitted and not if the union is
> optimized out.

Eh, I'm not sure about that.  I think it would be technically better
if the variable were present in the debug info but with no location,
so that the debugger says that it's optimized out.

I think this is what happens with other local variables.

> The other would be to
> 1) teach middle-end that it should never emit DECL_HAS_VALUE_EXPR_P
>    vars, just emit debug info for them

This would be consistent with expand_one_var in cfgexpand.c.



r~