sibcalls and emit_eh_context

[Richard Earnshaw]

I've just been looking at replacing the existing hack in the ARM backend 
to detect tail-calls with the new sibcall functionality.  I've now run 
into a problem building tinfo.cc.

The problem is that emit_eh_context is generating a call_placeholder insn 
(to get the eh_context), but it is doing this after 
optimize_sibling_and_tail_recursive_calls() has replaced all the other 
placeholders with their appropriate calls.  The result is a 
call_placeholder insn that isn't fixed up, so eventually we get an abort 
when it isn't recognized.

Has anyone else come across this?  If not, why not, since it looks like a 
fairly generic problem?

R.