[PATCH]: Commence binfo extraction

[Andrew Haley <aph at redhat dot com>]

Nathan Sidwell writes:
 > Hi,
 > this patch separates binfos from TREE_VECs.  As a stepping stone to turning
 > a binfo into not-a-tree I've turned it into a new tree node.  I've separated
 > the uses of tree_type.binfo into those that are as binfos, and those that
 > are other.  Note the following points,
 > *) The only two languages that appear to be using binfos are C++ and Java.
 > *) There is a tiny memory reduction with this patch, as the tree_binfo struct
 > is slightly smaller than the equivalent tree_vec struct.
 > *) I've introduced a new global variable 'binfo_lang_slots' to record
 > the number of language dependent slots.  This is a *temporary* situation.
 > *) The java changes are extremely mechanical here
 > *) The objc change was more complicated, but mechanical.  Objc was using the
 > binfo slot for not-a-binfo, but still calling the binfo machinery to process
 > it. This confused me.
 > *) I grepped the various debugging info outputters for binfo use, but only
 > dwarf2out appeared to be using them.  It's possible I missed something.
 > 
 > bootstraped and tested will all languages on i686-pc-linux-gnu, ok?

This looks OK, but we're trying to track down a recent regression in
Java.  Would it be a problem to delay for a day or two?

Thanks,
Andrew.