Re: [C/C++] obscure ICEs with __complex__ on Solaris

[Gabriel Dos_Reis <Gabriel dot Dos_Reis at sophia dot inria dot fr>]

Mumit Khan <khan@xraylith.wisc.EDU> writes:

| Gabriel Dos_Reis <Gabriel.Dos_Reis@sophia.inria.fr> writes:
| > 
| > Well, these took me two days to track :-(
| 
| If I had seen yours earlier, mine would've taken a lot shorter ;-) See
| below.

I was working on this since yesterday.  I should have posted earlier.
(I must tank Jeff for his helpful advice)

| Hi Gaby, 
| 
| Ah, I just submitted a bug report that dies in the same place. Please see
| my post to see the gdb backtrace and cause of the crash.
| 
| The culprit seems to be 
|   tree part_type = TREE_TYPE (TREE_TYPE (decl));
| in function.c:put_var_into_stack().

Exactly!  From my investigations, it appears that TREE_TYPE(decl)
reprensents a RECORD_TYPE (instead of COMPLEX_TYPE) causing 
TREE_TYPE (TREE_TYPE (decl)) to be NULL.  I tried to figure out how
that is so but failed to understand the whole logic :-(

-- Gaby