Re: Your patch to PR 30700

["=?big5?b?RG91ZyBLd2FuICjD9q62vHcp?=" <dougkwan at google dot com>]

Alternatively, I would like to propose replacing the assert into a
simple "return *tp".  Without the complete call-graph, we cannot tell
if a DECL will be still unused eventually.  So a reasonable thing to
do is not emit DWARF2 info early.

I'm running dejagnu right now.

-Doug

2007/7/26, Diego Novillo <dnovillo@google.com>:
> Jan,
>
> Your patch http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=123922 is
> causing an ICE when compiling this snippet with -g
>
> namespace NS {
>   int x = 0;
>   int &ref = x;
> }
>
> using NS::ref;
>
> The problem is that we are emitting debug info for 'x' during parsing,
> so cgraph_global_info_ready is still false.  Since 'x' has already been
> processed in this case, we will have a varpool entry for it.
>
> Since it's too early, the 'needed' indicator may be too pessimistic, but
> it didn't seem to me that we have much of a choice.  So, I thought that
> if the varpool entry for the decl is already finalized, we can use it.
>
> The other option would be to try and push this analysis much later after
> parsing, but that seems fairly intrusive.  I'm currently testing the
> attached patch.  Honza, does this make sense at all?
>
> Thanks.

Attachment: diffs-31899.txt
Description: Text document