Re: [PATCH] gdb: restore nullptr check in compunit_symtab::find_call_site

[Tom Tromey <tom@tromey.com>]

>>>>> "Simon" == Simon Marchi <simark@simark.ca> writes:

>> Yeah, that just doesn't seem significantly better to me.  It is more
>> like a spelling change.

Simon> I don't think it's just a spelling change, because a destructor would
Simon> invoke the destructor of fields.  So I think that if we started invoking
Simon> ~compunit_symtab() manually (instead of compunit_symtab::finalize()),
Simon> we could start using non-trivially-destructible stuff directly as fields
Simon> of compunit_symtab.  In other words, m_call_site_htab wouldn't have to
Simon> be a pointer anymore.

I thought you meant something like 'symtab->m_mumble.~unique_ptr()', as
opposed to 'delete symtab->m_mumble'.

So I see what you mean, but this also seems less good to me.
Currently the obstack allocator prohibits non-trivial destructors, and
this is a good thing, because it means it is harder to make a mistake
here -- adding a field that requires destruction will cause a
compile-time error.

I'd be reluctant to lift that restriction.

Another way forward is a slab allocator that is type-specific so it
knows how to run the destructors.  Inheriting from an interface class
and keeping objects on a list would also work for this purpose.

I somewhat doubt we'll need this kind of thing but it's fine by me if we
do.

Tom