[Ada] Casing on composite values

Pierre-Marie de Rodat derodat@adacore.com
Thu Jun 17 14:33:11 GMT 2021


Initial support for casing on composite values if extensions are allowed
(e.g., if -gnatX is specified). The implementation is far from complete.

Tested on x86_64-pc-linux-gnu, committed on trunk

gcc/ada/

	* exp_ch5.adb
	(Expand_N_Case_Statement.Expand_General_Case_Statement): New
	subprogram.
	(Expand_N_Case_Statement): If extensions are allowed and the
	case selector is not of a discrete type, then call
	Expand_General_Case_Statement to generate expansion instead of
	flagging the non-discrete selector as an error.
	* sem_case.ads (Is_Case_Choice_Pattern): New Boolean-valued
	function for testing whether a given expression occurs as part
	of a case choice pattern.
	* sem_case.adb (Composite_Case_Ops): New package providing
	support routines for the new form of case statements. This
	includes a nested package, Composite_Case_Ops.Value_Sets, which
	encapsulates the "representative values" implementation of
	composite value sets.
	(Check_Choices.Check_Case_Pattern_Choices): New procedure for
	semantic checking of non-discrete case choices. This includes
	the checks pertaining to coverage and overlapping.
	(Check_Choices.Check_Composite_Case_Selector): New procedure for
	semantic checking of non-discrete case selectors.
	(Check_Choices): If extensions are allowed then a non-discrete
	selector type no longer implies that an error must have been
	flagged earlier.  Instead of simply returning, call
	Check_Composite_Case_Selector and Check_Case_Pattern_Choices.
	(Is_Case_Choice_Pattern): Body of new function declared in
	sem_case.ads .
	* sem_ch5.adb (Analyze_Case_Statement): If extensions are
	allowed, then we can't use RM 5.4's "The selecting_expression is
	expected to be of any discrete type" name resolution rule.
	Handle the case where the type of the selecting expression is
	not discrete, as well as the new ambiguous-name-resolution error
	cases made possible by this change.
	* sem_res.adb (Resolve_Entity_Name): It is ok to treat the name
	of a type or subtype as an expression if it is part of a case
	choice pattern, as in "(Field1 => Positive, Field2 => <>)".
	* exp_aggr.adb (Expand_Record_Aggregate): Do not expand case
	choice aggregates.
	* gen_il-fields.ads: Define two new node attributes,
	Binding_Chars and Multidefined_Bindings.
	* gen_il-gen-gen_nodes.adb: The new Multidefined_Bindings
	attribute is Boolean-valued and may be set on
	N_Case_Statement_Alternative nodes. The new Binding_Chars
	attribute is Name_Id-valued and may be set on
	N_Component_Association nodes.
	* par-ch4.adb (P_Record_Or_Array_Component_Association): When
	parsing a component association, check for both new syntax forms
	used to specify a bound value in a case-choice aggregate.  In
	the case of a box value, an identifier may occur within the box,
	as in "Foo => <Abc>" instead of "Foo => <>". In the more general
	case, an expression (or a box) may be followed by "is
	<identifier>", as in
	"Foo => Bar is Abc" instead of just "Foo => Bar".
	* sem_aggr.adb (Resolve_Record_Aggregate): Do not transform box
	component values in a case-choice aggregate.
	* sinfo.ads: Provide comments for the new attributes added in
	gen_il-fields.ads.
	* doc/gnat_rm/implementation_defined_pragmas.rst: Describe this
	new feature in documentation for pragma Extensions_Allowed.
	* gnat_rm.texi: Regenerate.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch.diff.gz
Type: application/gzip
Size: 22126 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/gcc-patches/attachments/20210617/862adb54/attachment-0001.gz>


More information about the Gcc-patches mailing list