* ada-exp.y: Considerable reorganization to move functionality

from ada-lex.l to here, where it is logically more appropriate.
	The original reason, however, was to prevent premature name
	lookups for selector names in record aggregates.
	(BLOCKNAME, TYPENAME, OBJECT_RENAMING): Remove; lexer now returns
	NAME for all of these.
	(VAR): New artificial token to clarify precedence rules.
	(OTHERS): New lexeme.
	(empty_stoken): New symbol.
	(%union): Remove ssym, voidval.
	(%type): Remove <voidval> type declarations.
	(syntax definitions): Add aggregates.
	Remove distinction between NAME, TYPENAME, BLOCKNAME, OBJECT_RENAMING.
	Rename some non-terminals to be closer to reference manual usage.
	Tighten up expression syntax to disallow certain non-Ada
	constructions such as X and then Y or else Z.
	(ada_parse): Remove initialization of left_block_context.
	(write_var_from_name): Remove.
	(write_var_or_type): New function, containing previous code from
	defunct write_var_from_name and name_lookup.
	(block_lookup): New function, moved from ada-lex.l
	(select_possible_type_sym): New function, factored out of
	name_lookup, which used to be in ada-lex.l.
	(find_primitive_type): Ditto.
	(chop_selector): Ditto.
	(write_ambiguous_var): New function, factored out of defunct
	write_var_from_name.
	(write_selectors): New function.
	(write_name_assoc): New function.
	(write_exp_op_with_string): New function.

	* ada-lex.l (processId): Change interface to return stoken.
	(tempbuf, resize_tempbuf, tempbuf_size, tempbuf_len): Remove.
	(block_lookup, name_lookup): Remove.  Functionality moved to
	ada-exp.y.
	(state IN_STRING): Remove.
	(rules): Handle string escapes in processString.
	Add 'others' token.
	Return all NAMEs, BLOCKNAMEs, OBJECT_RENAMINGs, TYPENAMEs in
	yylval.sval (as simple strings).
	All name look-ups now handled in ada-exp.y.
	Introduce "::" (COLONCOLON) token and return as separate token.
	(processId): Change return convention.  Comment.
	Leave leading "'" in place.
	(processString): New function.
	(find_dot_all): Add note to comment.
	Fix problem that allowed match only at the end.

	* ada-lang.c: Introduce aggregates.
	(find_struct_field): Add new parameter to count fields skipped, and
	allow other output parameters to be NULL.
	(value_tag_from_contents_and_address, ada_value_struct_elt): Use
	new find_struct_field.
	(ada_index_struct_field, assign_aggregate, ada_is_array_type)
	(num_visible_fields, ada_index_struct_field_1, ada_index_struct_field)
	(num_component_specs, assign_component, assign_aggregate):
	(aggregate_assign_from_choices,aggregate_assign_positional)
	(aggregate_assign_others,add_component_interval):
	New functions.
	(ada_evaluate_subexp): Declare.
	Add aggregate-related operators.
	(ada_forward_operator_length): Declare.
	(resolve_subexp): Add cases for new aggregate operators and OP_NAME.
	Consolidate Ada operators, using ada_forward_operator_length.
	(ada_search_struct_field): Search in forward order.
	(ADA_OPERATORS): Add new aggregate operators.
	(ada_operator_length, ada_op_name, ada_forward_operator_length)
	(ada_dump_subexp_body, ada_print_subexp): Handle new aggregate
	operators and OP_NAME.
	(ada_type_of_array): Use longest_to_int.
	(value_assign_to_component): New function.
	(ada_forward_operator_length, ada_op_name, ada_dump_subexp_body):
	Add OP_NAME case.
	(ada_forward_operator_length, ada_dump_subexp_body):
	Add OP_STRING case.

	* ada-lang.h (enum ada_operator): Add OP_AGGREGATE, OP_OTHERS,
	OP_CHOICES, OP_DISCRETE_RANGE, OP_POSITIONAL.

gdb/ada-lang.h

@@ -115,6 +115,54 @@ enum ada_operator
        type TYPE (typically a subrange). */
     UNOP_IN_RANGE,
 
+    /* An aggregate.   A single immediate operand, N>0, gives
+       the number of component specifications that follow.  The
+       immediate operand is followed by a second OP_AGGREGATE.  
+       Next come N component specifications.  A component
+       specification is either an OP_OTHERS (others=>...), an
+       OP_CHOICES (for named associations), or other expression (for
+       positional aggregates only).  Aggregates currently
+       occur only as the right sides of assignments. */
+    OP_AGGREGATE,
+
+    /* An others clause.  Followed by a single expression. */
+    OP_OTHERS,
+
+    /* An aggregate component association.  A single immediate operand, N, 
+       gives the number of choices that follow.  This is followed by a second
+       OP_CHOICES operator.  Next come N operands, each of which is an
+       expression, an OP_DISCRETE_RANGE, or an OP_NAME---the latter 
+       for a simple name that must be a record component name and does 
+       not correspond to a single existing symbol.  After the N choice 
+       indicators comes an expression giving the value.
+
+       In an aggregate such as (X => E1, ...), where X is a simple
+       name, X could syntactically be either a component_selector_name 
+       or an expression used as a discrete_choice, depending on the
+       aggregate's type context.  Since this is not known at parsing
+       time, we don't attempt to disambiguate X if it has multiple
+       definitions, but instead supply an OP_NAME.  If X has a single
+       definition, we represent it with an OP_VAR_VALUE, even though
+       it may turn out to be within a record aggregate.  Aggregate 
+       evaluation can use either OP_NAMEs or OP_VAR_VALUEs to get a
+       record field name, and can evaluate OP_VAR_VALUE normally to
+       get its value as an expression.  Unfortunately, we lose out in
+       cases where X has multiple meanings and is part of an array
+       aggregate.  I hope these are not common enough to annoy users,
+       who can work around the problem in any case by putting
+       parentheses around X. */
+    OP_CHOICES,
+
+    /* A positional aggregate component association.  The operator is 
+       followed by a single integer indicating the position in the 
+       aggregate (0-based), followed by a second OP_POSITIONAL.  Next 
+       follows a single expression giving the component value.  */
+    OP_POSITIONAL,
+
+    /* A range of values.  Followed by two expressions giving the
+       upper and lower bounds of the range. */
+    OP_DISCRETE_RANGE,       
+
     /* End marker */
     OP_ADA_LAST
   };