Re: [PATCH] PR c++/30659 segfault w/ invalid input

[Mark Mitchell <mark at codesourcery dot com>]

Seongbae Park wrote:
> On 06 Feb 2007 07:50:08 -0800, Ian Lance Taylor <iant@google.com> wrote:
>> "Seongbae Park" <seongbae.park@gmail.com> writes:
>>
>> > This patch fixes the compiler segmentation fault on an invalid input.
>> > Bootstrapped and regtested on i686-linux.
>> > As I have never looked at c+ frontend before,
>> > there's a high chance I'm not doing the right thing.
>> > OK for mainline ?
>> >
>> > 2007-02-05  Seongbae Park <seongbae.park@gmail.com>
>> >
>> >         PR c++/30659
>> >         * search.c (lookup_member): Return error_mark_node on NULL
>> basetype.
>>
>> It would be appropriate to add the test case to the testsuite as part
>> of this patch.
> 
> The new patch with the testsuite is attached.
> 
>> (I can't approve this patch, but looking at it I wonder if should be
>> catching a NULL basetype earlier.)
> 
> I wondered about that as well. The NULL xbasetype comes from
> DECL_CONTEXT of a decl, and I couldn't quite figure out
> whether this is allowed or not and I didn't try to trace back
> where it is coming from - and thought it's easier to ask :)
> I'm willing to hunt down deeper if that's the right thing to do.

I think we should look deeper; it seems wrong to ever call lookup_member
with a NULL xbasetype; we should have noticed the problem at some
earlier point.  What is the DECL whose DECL_CONTEXT is NULL, and why do
we care? :-)

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713