Re: C++ PATCH: Do not allocate conversions in GC-memory

[Mark Mitchell <mark at codesourcery dot com>]

> memory in conversion_obstack gets zeroed (or something) in each place
> where it's freed?  Otherwise there's no check that a pointer does not
> escape.
>  
There are checks that all of the memory in the obstack is freed.

I could add zeroing, but there's no need for that.  It's evident from 
construction that none of the memory can escape.

> How does this reduce the number of calls to ggc_set_mark?
Because there is less GCable memory, fewer collections, and fewer ways 
of reaching the same GCable objects.

>  Is it
> possible that you're now storing pointers to GCed memory in non-GCed
> memory, or is this entirely due to fewer ggc_collect runs?
>
>  
We're definitely storing some memory to GCed memory in non-GCed memory.

That's a good thing; it's no different than storing a pointer to GCed 
memory on the stack.  This memory is transient, and there can be no 
calls to the collector while these routines are active unless a new 
context is pushed.  The only GC-able memory pointed to from these data 
structures is memory that was extant before the data structures were active.

Similarly, no PCH file can be created while these routines are active.

> Did you run a bootstrap with gcac checking?  
Yes.

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com