[Ada] Improve support for casing on types with controlled parts

gcc/ada/

	* sem_case.adb (Check_Bindings): Provide a second strategy for
	implementing bindings and choose which strategy to use for a
	given binding. The previous approach was to introduce a new
	object and assign the bound value to the object.  The new
	approach is to introduce a renaming of a dereference of an
	access value that references the appropriate subcomponent, so no
	copies are made.  The original strategy is still used if the
	type of the object is elementary.  When the renaming approach is
	used, the initialization of the access value is not generated
	until expansion. Until this missing initialization is added, the
	tree looks like a known-at-compile-time dereference of a null
	access value: Temp : Some_Access_Type; Obj : Designated_Type
	renames Temp.all; This leads to problems, so a bogus initial
	value is provided here and then later deleted during expansion.
	(Check_Composite_Case_Selector): Disallow a case selector
	expression that requires finalization. Note that it is ok if the
	selector's type requires finalization, as long as the expression
	itself doesn't have any "newly constructed" parts.
	* exp_ch5.adb (Pattern_Match): Detect the case where analysis of
	a general (i.e., composite selector type) case statement chose
	to implement a binding as a renaming rather than by making a
	copy. In that case, generate the assignments to initialize the
	access-valued object whose designated value is later renamed
	(and remove the bogus initial value for that object that was
	added during analysis).
	* sem_util.ads, sem_util.adb: Add new function
	Is_Newly_Constructed corresponding to RM 4.4 term.

0 files changed, 0 insertions, 0 deletions