mercury/compiler/mlds.m

%-----------------------------------------------------------------------------%
% Copyright (C) 1999-2004 The University of Melbourne.
% This file may only be copied under the terms of the GNU General
% Public License - see the file COPYING in the Mercury distribution.
%-----------------------------------------------------------------------------%

% MLDS - The Medium-Level Data Structure.
% Main author: fjh.

% This module defines the MLDS data structure itself.
% The MLDS is an intermediate data structure used in compilation;
% we compile Mercury source -> parse tree -> HLDS -> MLDS -> target (e.g. C).
% See notes/compiler_design.html for more information about the MLDS & LLDS.
%
% The MLDS is intended to be suitable for generating target code in
% languages such as Java, Java bytecode, high-level C, C++, or C--, etc.
% This is as opposed to the LLDS, which is better suited for generating
% assembler or the very low-level C or GNU C (lower level than C--) that
% the original Mercury compiler backend generates.  In the LLDS, we are
% doing all our own register allocation and stack manipulation, but with
% MLDS, those tasks are the responsibility of the target.
% The one really important simplification that we do make relative to the
% HLDS is that MLDS does not have any support for non-determinism, so the
% HLDS->MLDS compiler must compile non-deterministic code into code that
% uses continuation passing.

% The MLDS data structure is quite full-featured, including for example
% support for arbitrary nesting, multiple return values, and tagged pointers.
% However, many
% of the intended target languages don't support all of those features. 
% Therefore the HLDS->MLDS compiler must ensure that the final MLDS code that
% it eventually generates does not use features which the target does not
% support.  This will (presumably) be accomplished by having handle_options.m
% set various flags according to the selected target, and having the
% HLDS->MLDS compiler take account of those flags and either generate simpler
% MLDS code in the first place or run some extra simplification passes over
% the MLDS code before invoking the MLDS->target compiler.
% 

%-----------------------------------------------------------------------------%
%
% Mapping Mercury names to MLDS names
%

% 1. Modules
%
% Mercury module names map directly to MLDS package names, except that
% modules in the Mercury standard library map get a `mercury' prefix,
% e.g. `mercury.builtin', `mercury.io', `mercury.std_util', etc.
% [Rationale: omitting the `mercury' prefix would lead to namespace
% pollution in the generated target language code.]

% 2. Procedure names
%
% HLDS procedure names map directly to MLDS function names.
% MLDS function names are structured terms that include sufficient
% information (arity, pred_or_func indicator, mode number, etc.)
% to avoid any ambiguities.
%
% [Rationale: the reason for keeping structured names rather than flattened
% names at the MLDS level is that we don't want to do name mangling at
% the HLDS -> MLDS stage, since we don't know what restrictions the target
% language will impose.  For example, some target languages (e.g. C++, Java)
% will support overloading, while others (e.g. C) will not.]

% 3. Procedure signatures
%
% MLDS function signatures are determined by the HLDS procedure's
% argument types, modes, and determinism.
% Procedures arguments with type `io__state' or `store(_)' are not passed.
% Procedure arguments with top_unused modes are not passed.
% Procedures arguments with top_in modes are passed as input.
% Procedures arguments with top_out modes are normally passed by reference.
% However, several alternative approaches are also supported (see below).
%
% Procedures with determinism model_det need no special handling.
% Procedures with determinism model_semi must return a boolean.
% Procedures with determinism model_non get passed a continuation;
% if the procedure succeeds, it must call the continuation, and if
% it fails, it must return.
%
% With the `--copy-out' option, arguments with top_out modes will be returned
% by value.  This requires the target language to support multiple return
% values.  The MLDS->target code generator can of course handle that by mapping
% functions with multiple return values into functions that return a struct,
% array, or tuple.
% With the `--nondet-copy-out' option, arguments for nondet procedures with
% top_out modes will be passed as arguments to the continuation.
%
% [Rationale: this mapping is designed to be as simple as possible,
% and to map as naturally to the target language as possible.
% The complication with the `--copy-out' and `--nondet-copy-out'
% options is needed to cater to different target languages.
% Some target languages (e.g. C) fully support pass-by-reference,
% and then it is most consistent to use it everywhere.
% Some (e.g. IL) support pass-by-reference, but don't allow references
% to be stored in the environment structs needed for continuations,
% and so can't use pass-by-reference for nondet procedures.
% Others (e.g. Java) don't support pass-by-reference at all.]

% 4. Variables
%
% MLDS variable names are determined by the HLDS variable name and
% (in some cases, to avoid ambiguity) variable number.  The MLDS
% variable name is a structured term that keeps the original variable
% name separate from the distinguishing variable number. 
% It is up to each individual backend to mangle the variable name
% and number to avoid ambiguity where necessary.
% All references to MLDS variables must however be fully qualified
% with the name of the enclosing entity that defines them.
%
% [Rationale: the reason for keeping structured names rather than
% mangled names at the MLDS level is that in some cases the mangling is
% undesirable, as the original HLDS variable names are what is required
% (for instance, when interfacing with foreign code which includes
% references to the original HLDS variable names).]

% 5. Global data
%
% MLDS names for global data are structured; they hold some
% information about the kind of global data (see the mlds__data_name type).
%
% It's not clear whether this is actually useful.
% And we're not completely consistent about applying this approach.
% Many of the passes which create global data items do not actually
% generate structured names, but instead just generate var/1 names
% where the variable name already contains some mangling to ensure uniqueness.
% Examples of this include "string_table" and "next_slots_table" (for
% string switches), "float_*" (boxed floating point constants),
% and "obj_*" (reserved objects, for representing constants in
% discriminated union types).

% 6. Types
%
% If there is an MLDS type corresponding to a Mercury type, then
% the Mercury type name maps directly to the MLDS type name,
% suitably module-qualified of course. 
% The MLDS type name includes the type arity (arity overloading is allowed).
% However, if a low-level data representation scheme is used,
% then some Mercury types may not have corresponding MLDS type names
% (that is, the corresponding MLDS type may be just `MR_Word' or its
% equivalent).

% 7.  Data constructors.
%
% With --high-level-data, Mercury data constructors normally get mapped to
% MLDS types nested within the MLDS type for the Mercury data type to which
% the constructors belong.  There are some exceptions; see ml_type_gen.m
% for full details.  The MLDS type name includes the constructor arity.
% [Rationale: Mercury allows data constructors to be overloaded on
% their result type within a single module, and also allows them
% to be overloaded on arity within a single type.  Nesting resolves
% the first problem, and keeping the arity in the MLDS type name
% resolves the second.]
% XXX There is a problem in the Java back-end with Mercury types for which
% the constructor name is the same as the type name.
%
% XXX Also need to think about equivalence types and no_tag types.

% 8.  Insts and modes
%
% Inst and mode definitions do not get translated into MLDS.
%
% [Inst and mode definitions do however affect the signatures of the
% MLDS functions used to implement each procedure.]

% 9. Type classes.
% 
% Currently type classes are handled early and at a fairly low level.
% It's not yet clear how easy it will be to convert this to MLDS.
% It may depend to some degree on the target language.
% But if it is possible, then when it is done the target language
% generator will be able to ignore the issue of type classes.
%
% For language interoperability, however, it might be nice if the
% translation were done at higher level.
%
% Mercury type classes should map directly to MLDS interfaces.
%
% Mercury instance definitions should map to classes which implement the
% corresponding interface.  Note that if there is an instance declaration
% `:- instance foo(bar)', then the MLDS type for `bar' will *not* implement 
% the MLDS interface for `foo' -- instead, there will be a new MLDS type
% for `instance foo(bar)' which implements that interface.

%-----------------------------------------------------------------------------%
%
% Mapping MLDS names to the target language
%

% Ultimately the exact choice of how MLDS names are mapped to
% the target language is up to the target language generator.
% So the remarks here are just guidelines, not strict rules.

% 1. Names.
%
% If the target language has standard conventions about certain categories
% of names beginning with uppercase or lowercase letters, then the target
% language generator should generate names that respect those conventions.
%
% An MLDS name may contain arbitrary characters.
% If the target language has restrictions on what names can be used
% in identifiers, then it is the responsibility of the target language
% generator to mangle MLDS names accordingly to ensure that they abide
% by those restrictions.

% 2. Packages.
%
% MLDS packages should be mapped directly to target packages, if possible.
% If the target doesn't have a notion of packages, then they should be
% mapped to names of the form "foo.bar.baz" or if dots are not allowed
% then to "foo__bar__baz".

% 3. Overloading.
%
% If the target does not support overloading, then any Mercury names which
% are overloaded within a single Mercury module must be qualified to avoid
% ambiguity.  However, Mercury names which are not overloaded should not
% be qualified.  If a name is overloaded but occurs only once in the module's
% interface then the version in the interface should not be qualified.
% The arity-zero version of type should not be arity-qualified
% unless this would cause ambiguity with an unqualified name generated by
% the previous rule.  Likewise, the arity-zero version of function
% (i.e. a constant) should not be function-qualified or arity-qualified
% unless this would cause ambiguity with an unqualified name generated
% the aforementioned rule.
% The first mode of a predicate should not be mode-qualified.
%
% [Rationale: name mangling should be avoided where possible, because
% this makes it easier for the user to interoperate with other languages
% and to use tools which do not properly demangle Mercury names.]

% 4. Procedures.
%
% If a procedure name needs to be qualified, the qualification should be
% done by appending "_f" for functions or "_p" for predicates,
% optionally followed by the arity,
% optionally followed by an underscore and then the mode number,
% optionally followed by "_i" and then the MLDS function sequence number
% (for internal MLDS functions, used e.g. to implement backtracking).
%
% [Rationale: any qualifiers should go at the end of a name, so that
% command-line completion works even for mangled names (and hopefully
% even shows all the possible alternatives...).]
%
% To avoid ambiguity as to whether a name is qualified or not,,
% any procedures whose unqualified name matches the pattern for qualified
% names, i.e. the regular expression `.*_[fp][0-9]*(_[0-9]+)?(_i[0-9]+)?',
% should always be qualified (even if not overloaded).

% 5. Types.
%
% If a type name needs to be qualified, the qualification should be
% done by appending an underscore followed by the arity.
%
% To avoid ambiguity as to whether a name is qualified or not,
% any types whose unqualified name matches the pattern for qualified
% names, i.e. the regular expression `.*_[0-9]+',
% should always be qualified (even if not overloaded).

%-----------------------------------------------------------------------------%
%
% Notes on garbage collection and liveness.
% 

% "Liveness-accurate GC" is GC in which the collector does not trace local
% variables which are definitely not live according to a straight-forward
% static analysis of definite-deadness/possible-liveness.  Liveness-accurate
% GC is desirable, in general, since tracing through variables which are
% no longer live may lead to excessive memory retention for some programs.
% However these programs are relatively rare, so liveness-accurate GC
% is not always worth the extra complication.
%
% The MLDS is therefore designed to optionally support liveness-accurate
% GC, if the target language supports it.  If liveness-accurate GC is
% supported and enabled, then it is the responsibility of the target
% language implementation to do whatever is needed to avoid having the GC
% trace variables which have gone out of scope.
%
% That means that to support liveness-accurate the HLDS->MLDS code
% generator just needs to cover the cases where a straight-forward liveness
% calculation on the generated MLDS does not match up with the desired
% result of a straight-forward liveness calculation on the HLDS.  That is,
% the HLDS->MLDS code generator must generate code to clobber variables
% which are no longer live according to a straight-forward liveness
% calculation on the HLDS but which have not gone out of scope in
% the generated MLDS.  For example, with our current HLDS->MLDS code
% generation scheme, this is the case for variables in the `else' of a
% nondet if-then-else once the `then' has been entered.
% (XXX Currently ml_code_gen.m does _not_ clobber those variables, though.)

% The rationale for leaving most of the responsibility for liveness-accurate
% GC up to the MLDS back-end is as follows: at very least we need the
% target language implementation's _cooperation_, since if the MLDS code
% generator inserts statements to clobber variables that are no longer live,
% then an uncooperative target language implementation could just optimize
% them away, since they are assignments to dead variables.  Given this need
% for the MLDS target back-end's cooperation, it makes sense to assign as
% much of the responsibily for this task as is possible to the MLDS target
% back-end, to keep the front-end simple and to keep the responsibility
% for this task in one place.
%
% But in cases such as nondet if-then-else, where HLDS-liveness does not
% match MLDS-liveness, we can't just leave it to the MLDS target back-end,
% because that would require assuming an unreasonably smart liveness
% calculation in the MLDS target back-end, so in such cases we do need
% to handle it in the HLDS->MLDS code generator.

%-----------------------------------------------------------------------------%

:- module ml_backend__mlds.

:- interface.

:- import_module backend_libs__builtin_ops.
:- import_module backend_libs__code_model.
:- import_module backend_libs__foreign.
:- import_module backend_libs__rtti.
:- import_module check_hlds__type_util.
:- import_module hlds__hlds_module.
:- import_module hlds__hlds_pred.
:- import_module libs__globals.
:- import_module parse_tree__prog_data.

:- import_module bool, list, std_util, map.

%-----------------------------------------------------------------------------%

:- type mercury_module_name == prog_data__module_name.

%
% The type `mlds' is the actual MLDS.
% XXX we ought to make this type abstract
%
:- type mlds
	---> mlds(
			% The Mercury module name.
		name		:: mercury_module_name,

			% Code defined in some other language, e.g.  for
			% `pragma c_header_code', etc.
		foreign_code	:: map(foreign_language, mlds__foreign_code),

			% The MLDS code itself
			% Packages/classes to import
		toplevel_imports :: mlds__imports,

			% Definitions of code and data
		defns		:: mlds__defns
	).

:- func mlds__get_module_name(mlds) = mercury_module_name.

:- type mlds__imports == list(mlds__import).

	% For the C backend, we generate a `.mh' file containing the
	% prototypes for C functions generated for `:- pragma export'
	% declarations, and a `.mih' file containing the prototypes
	% for the C functions generated by the compiler which should
	% not be visible to the user.
	%
	% For languages which don't support multiple interfaces, it's
	% probably OK to put everything in the compiler-visible interface.
:- type mercury_mlds_import_type
	--->	user_visible_interface
	;	compiler_visible_interface.

% Currently an import just gives the name of the package to be imported.
% This might perhaps need to be expanded to cater to different kinds of
% imports, e.g. imports with wild-cards ("import java.lang.*").
:- type mlds__import
	--->	mercury_import(
			mercury_mlds_import_type :: mercury_mlds_import_type,
			import_name		:: mlds_module_name
		)

	;	foreign_import(
			foreign_import_name	:: foreign_import_name
		).

:- type foreign_import_name
	--->	il_assembly_name(mlds_module_name).

% An mlds_module_name specifies the name of an mlds package or class.
:- type mlds_module_name.

% An mlds__package_name specifies the name of an mlds package.
:- type mlds__package_name == mlds_module_name.

% Given the name of a Mercury module, return the name of the corresponding
% MLDS package in which this module is defined.
:- func mercury_module_name_to_mlds(mercury_module_name) = mlds_module_name.

% Given the name of a Mercury module, and a package name, return the
% name of the corresponding MLDS module name in which this module is
% defined.
% In this case, the package is specified as the first parameter (c.f.
% mercury_module_name_to_mlds above).
:- func mercury_module_and_package_name_to_mlds(mercury_module_name,
		mercury_module_name) = mlds_module_name.

% Given the name of a Mercury module return the fully qualified module
% name.  ie For the name System.Object which is defined in the source
% package mscorlib it will return System.Object.
:- func mlds_module_name_to_sym_name(mlds__package_name) = sym_name.

% Give the name of a Mercury module, return the name of the corresponding 
% MLDS package.
:- func mlds_module_name_to_package_name(mlds_module_name) = sym_name.

% Is the current module a member of the std library,
% and if so which module is it?
:- pred is_std_lib_module(mlds_module_name::in, string::out) is semidet.

% Given an MLDS module name (e.g. `foo.bar'), append another class qualifier
% (e.g. for a class `baz'), and return the result (e.g. `foo.bar.baz').
% The `arity' argument specifies the arity of the class.
:- func mlds__append_class_qualifier(mlds_module_name, mlds__class_name,
		arity) = mlds_module_name.

% Append a wrapper class qualifier to the module name and leave the
% package name unchanged.
:- func mlds__append_wrapper_class(mlds_module_name) = mlds_module_name.

% Append an arbitrary qualifier to the module name and leave the package
% name unchanged.
:- func mlds__append_name(mlds_module_name, string) = mlds_module_name.

% When targetting languages such as IL, C#, and Java, which don't
% support global methods or global variables, we need to wrap all the
% generated global functions and global data inside a wrapper class.
% This function returns the name to use for the wrapper class.
:- func wrapper_class_name = string.

:- type mlds__defns == list(mlds__defn).
:- type mlds__defn
	---> mlds__defn(
		mlds__entity_name,	% the name of the entity being declared
		mlds__context,		% the source location
		mlds__decl_flags,	% these contain the following:
			% mlds__access,		% public/private/protected
			% mlds__member_type,	% static/per_instance
			% mlds__virtuality,	% virtual/non_virtual
			% mlds__finality,	% final/overridable (funcs only)
			% mlds__constness,	% const/modifiable  (data only)
			% mlds__is_abstract,	% abstract/concrete
			% etc.
		mlds__entity_defn	% the definition of the entity
	).

% An mlds name may contain arbitrary characters.
% If the target language has restrictions on what names can be used
% in identifiers, then it is the responsibility of the target language
% generator to mangle these names accordingly.
:- type mlds__fully_qualified_name(T)
	---> 	qual(mlds_module_name, T).
:- type mlds__qualified_entity_name
	==	mlds__fully_qualified_name(mlds__entity_name).

:- type mlds__entity_name
	--->	type(mlds__class_name, arity)	% Name, arity.
	;	data(mlds__data_name)
	;	function(
			mlds__pred_label,	% Identifies the source code
						% predicate or function.

			proc_id,		% Mode number.

				% A sequence number used to distinguish 
				% different MLDS functions when compiling a
				% single HLDS predicate into multiple MLDS
				% functions (e.g. to handle backtracking).
			maybe(mlds__func_sequence_num),

				% This should generally not be needed much,
				% since all the necessary information should
				% be in the mlds__pred_label and/or in the
				% mlds__entity_defn.  However, the target
				% generator may want to refer to the HLDS
				% for additional information.
			pred_id			% Specifies the HLDS pred_id.
		)
	;	export(string)	% A pragma export name.
	.

:- type mlds__func_sequence_num == int.

	% This specifies information about some entity being defined
	% The entity may be any of the following:
	%	constant or variable
	%	function
	%	class, including
	%		package (class with only static (one_copy) members)
	%		interface (abstract class, no data members)
	%		struct (value class)
	%		enum
:- type mlds__entity_defn
		% constants or variables
	--->	mlds__data(
			mlds__type,
			mlds__initializer,
				% If accurate GC is enabled, we associate
				% with each variable the code needed to
				% trace that variable when doing GC.
			mlds__maybe_gc_trace_code
		)
		% functions
	;	mlds__function(
			maybe(pred_proc_id),	% identifies the original
						% Mercury procedure, if any
			mlds__func_params,	% the arguments & return types
			mlds__function_body,	% the function body
			list(mlds__attribute)	% attributes
		)
		% packages, classes, interfaces, structs, enums
	;	mlds__class(
			mlds__class_defn
		).

	% If accurate GC is enabled, we associate with each variable
	% (including function parameters) the code needed to trace that
	% variable when doing GC.
	% `no' here indicates that no GC tracing code is needed,
	% e.g. because accurate GC isn't enabled, or because the
	% variable can never contain pointers to objects on the heap.
:- type mlds__maybe_gc_trace_code == maybe(mlds__statement).

	% It is possible for the function to be defined externally
	% (i.e. the original Mercury procedure was declared `:- external').
	% (If you want to generate an abstract body consider adding another
	% alternative here).
:- type mlds__function_body 
	--->	defined_here(mlds__statement)
	;	external.

	% Note that `one_copy' variables *must* have an initializer
	% (the GCC back-end relies on this).
	% XXX Currently we only record the type for structs.
	%     We should do the same for objects and arrays.
:- type mlds__initializer
	--->	init_obj(mlds__rval)
	;	init_struct(mlds__type, list(mlds__initializer))
	;	init_array(list(mlds__initializer))
	;	no_initializer
	.

:- type mlds__func_params
	---> mlds__func_params(
		mlds__arguments,	% names and types of arguments (inputs)
		mlds__return_types	% types of return values (outputs)
	).
:- type mlds__arguments == list(mlds__argument).
:- type mlds__argument
	---> mlds__argument(
		mlds__entity_name,		% argument name
		mlds__type,			% argument type
		mlds__maybe_gc_trace_code	% GC tracing code for this
						% argument, if needed
	).
:- type mlds__arg_types == list(mlds__type).
:- type mlds__return_types == list(mlds__type).

:- func mlds__get_arg_types(mlds__arguments) = list(mlds__type).

	% An mlds__func_signature is like an mlds__func_params
	% except that it only includes the function's type, not
	% the parameter names.
:- type mlds__func_signature
	---> mlds__func_signature(
		mlds__arg_types,	% argument types
		mlds__return_types	% return types
	).

:- func mlds__get_func_signature(mlds__func_params) = mlds__func_signature.

:- type mlds__class_kind
	--->	mlds__class		% A generic class:
					% can inherit other classes and
					% interfaces
					% (but most targets will only support
					% single inheritence, so usually there
					% will be at most one class).
	;	mlds__package		% A class with only static members
					% (can only inherit other packages).
					% Unlike other kinds of classes,
					% packages should not be used as types.
	;	mlds__interface		% A class with no variable data members
					% (can only inherit other interfaces)
	;	mlds__struct		% A value class
					% (can only inherit other structs).
	;	mlds__enum		% A class with one integer member and
					% a bunch of static consts
					% (cannot inherit anything).
	.


:- type mlds__class_name == string.
:- type mlds__class == mlds__fully_qualified_name(mlds__class_name).

	% Note that standard C doesn't support empty structs,
	% so when targetting C, it is the MLDS code generator's
	% responsibility to ensure that each generated MLDS class
	% has at least one base class or non-static data member.
:- type mlds__class_defn
	---> mlds__class_defn(
		kind	::	mlds__class_kind,
		imports	::	mlds__imports,	% imports these classes (or
						% modules, packages, ...)
		inherits ::	list(mlds__class_id),
						% inherits these base classes
		implements ::	list(mlds__interface_id),
						% implements these interfaces
		ctors	::	mlds__defns,	% has these constructors
		members ::	mlds__defns	% contains these members
	).

	% Note: the definition of the `mlds__type' type is subject to change.
	% In particular, we might add new alternatives here, so try to avoid
	% switching on this type.
:- type mlds__type
	--->	% Mercury data types
		mercury_type(
			prog_data__type,	% the exact Mercury type
			type_category,		% what kind of type it is:
						% enum, float, etc.
			exported_type		% a representation of the type
						% which can be used to
						% determine the foreign
						% language representation of
						% the type.
		)

	 	% The Mercury array type is treated specially, some backends
		% will treat it like any other mercury_type, whereas other may
		% use a special representation for it.
		% Arrays are type constructors in some backends, and so it is
		% easier to represent it here as a special type constructor.
		% (if we used the mercury_type representation above, we would
		% only classify the topmost level of the type, whereas we
		% really want to classify the element type for arrays, so
		% we can generate int[] for array(int)).
		% Note that mlds__mercury_array_type/1 is used for representing
		% Mercury arrays, whereas mlds__array_type/1 (see below)
		% is used for representing the target language's native arrays.
	;	mlds__mercury_array_type(mlds__type)

		% The type for the continuation functions used
		% to handle nondeterminism
	;	mlds__cont_type(mlds__return_types)

		% mlds__commit_type is used for storing information about a
		% commit. This is an abstract type; the exact definition
		% will depend on the back-end.  The only operations on this ADT
		% are `try_commit' and `do_commit'.  This type holds
		% information about the `try_commit' stack frame that is
		% needed to unwind the stack when a `do_commit' is executed.
		%
		% For the C back-end, if we're implementing do_commit/try_commit
		% using setjmp/longmp, then mlds__commit_type will be jmp_buf.
		% If we're implementing them using GNU C nested functions, then
		% it will be `__label__'; in this case, the local variable
		% of this "type" is actually a label, and doing a goto to that
		% label will unwind the stack.
		%
		% If the back-end implements commits using the target
		% language's, try/catch-style exception handling, as in
		% Java/C++/etc., then the target language implementation's
		% exception handling support will keep track of the information
		% needed to unwind the stack, and so variables of this type
		% don't need to be declared at all.
		%
		% See also the comments in ml_code_gen.m which show how commits
		% can be implemented for different target languages.
	;	mlds__commit_type

		% MLDS native builtin types.
		% These are the builtin types of the MLDS target language,
		% whatever that may be.
		% Currently we don't actually use many of these.
	;	mlds__native_bool_type
	;	mlds__native_int_type
	;	mlds__native_float_type
	;	mlds__native_char_type

		% This is a type of the target language.
	;	mlds__foreign_type(
			foreign_language_type
		)

		% MLDS types defined using mlds__class_defn
	;	mlds__class_type(
			mlds__class,		% name
			arity,
			mlds__class_kind
		)

		% MLDS array types.
		% These are single-dimensional, and can be indexed
		% using the `field' lval with an `offset' field_id;
		% indices start at zero.
		%
		% Note that mlds__array_type/1 is used for representing
		% the target language's native arrays, whereas
		% mlds__mercury_array_type/1 (see above) is used for
		% representing Mercury arrays.
		%
		% Currently MLDS array types are used for
		% (a) static constants that would otherwise be allocated
		%     with a `new_object', if we're using the low-level
		%     data representation (--no-high-level-data)
		% (b) for static constants of certain Mercury types which are
		%     always represented using the low-level data
		%     representation, regardless of --high-level-data,
		%     in particular closures and type_infos.
		% (c) for any other arrays generated internally by the
		%     MLDS code generator, e.g. the arrays used for
		%     string switches.
	;	mlds__array_type(mlds__type)

		% Pointer types.
		% Currently these are used for handling output arguments.
	;	mlds__ptr_type(mlds__type)

		% Function types.
		% For the C back-end, these are mapped to function
		% pointer types, since C doesn't have first-class
		% function types.
	;	mlds__func_type(mlds__func_params)

		% A generic type (e.g. `Word') that can hold any Mercury value.
		% This is used for implementing polymorphism.
	;	mlds__generic_type

		% A generic pointer type (e.g. `void *' in C)
		% that can be used to point to the environment
		% (set of local variables) of the containing function.
		% This is used for handling nondeterminism,
		% if the target language doesn't support
		% nested functions, and also for handling
		% closures for higher-order code.
	;	mlds__generic_env_ptr_type

	;	mlds__type_info_type

	;	mlds__pseudo_type_info_type

	;	mlds__rtti_type(rtti_id_maybe_element)

		% A type used by the ML code generator for references 
		% to variables that have yet to be declared.  This occurs 
		% once in ml_code_util.m where references to env_ptr's are 
		% generated but the declaration of these env_ptr's does not 
		% occur until the ml_elim_nested pass.
	;	mlds__unknown_type.

:- type mercury_type == prog_data__type.

:- func mercury_type_to_mlds_type(module_info, mercury_type) = mlds__type.

% Hmm... this is tentative.
:- type mlds__class_id == mlds__type.
:- type mlds__interface_id == mlds__type.

%-----------------------------------------------------------------------------%
%
% Declaration flags
%

:- type mlds__decl_flags.

:- type access
	%
	% used for class members (this includes globals,
	% which are actually members of the top-level package)
	%
	--->	public		% accessible to anyone
	;	protected	% only accessible to the class and to
				% derived classes
	;	private		% only accessible to the class
	;	default		% Java "default" access or .NET assembly
				% access: accessible to anything defined
				% in the same package.
	%
	% Used for entities defined in a block/2 statement,
	% i.e. local variables and nested functions.
	%
	;	local		% only accessible within the block
				% in which the entity (variable or
				% nested function) is defined
	.

:- type per_instance
	--->	one_copy	% i.e. "static" storage duration
				% (but not necessarily static linkage)
				% or static member function
				% Note that `one_copy' variables
				% *must* have an initializer
				% (the GCC back-end relies on this.)
	;	per_instance.	% i.e. "auto" local variable in function,
				% or non-static member of class.

:- type virtuality
	--->	non_virtual
	;	virtual.

:- type finality
	--->	overridable
	;	final.

:- type constness
	--->	modifiable
	;	const.

:- type abstractness
	--->	concrete
	;	abstract.

:- func access(mlds__decl_flags) = access.
:- func per_instance(mlds__decl_flags) = per_instance.
:- func virtuality(mlds__decl_flags) = virtuality.
:- func finality(mlds__decl_flags) = finality.
:- func constness(mlds__decl_flags) = constness.
:- func abstractness(mlds__decl_flags) = abstractness.

:- func set_access(mlds__decl_flags, access) = mlds__decl_flags.
:- func set_per_instance(mlds__decl_flags, per_instance) = mlds__decl_flags.
:- func set_virtuality(mlds__decl_flags, virtuality) = mlds__decl_flags.
:- func set_finality(mlds__decl_flags, finality) = mlds__decl_flags.
:- func set_constness(mlds__decl_flags, constness) = mlds__decl_flags.
:- func set_abstractness(mlds__decl_flags, abstractness) = mlds__decl_flags.

:- func init_decl_flags(access, per_instance, virtuality, finality,
		constness, abstractness) = mlds__decl_flags.

%-----------------------------------------------------------------------------%
%
% Foreign code interfacing
%

	%
	% Foreign code required for the foreign language interface.
	% When compiling to a language other than the foreign language,
	% this part still needs to be generated as C (or whatever) code
	% and compiled with a C (or whatever) compiler.
	%
:- type mlds__foreign_code
	---> mlds__foreign_code(
		foreign_decl_info,
		foreign_import_module_info,
		list(user_foreign_code),
		list(mlds__pragma_export)
	).

	%
	% Information required to generate code for each
	% `pragma export'.
	%
:- type mlds__pragma_export
	---> ml_pragma_export(
		string,			% Exported name
		mlds__qualified_entity_name, % MLDS name for exported entity
		mlds__func_params,	% MLDS function parameters
		mlds__context
	).

%-----------------------------------------------------------------------------%
%
% Attributes
%

:- type mlds__attribute
	---> custom(
		mlds__type
	).

%-----------------------------------------------------------------------------%
%
% Contexts (i.e. source code locations)
%

	% mlds__context is probably == prog_context,
	% but might also contain goal_path or other information.
:- type mlds__context.

:- func mlds__make_context(prog_context) = mlds__context.

:- func mlds__get_prog_context(mlds__context) = prog_context.

%-----------------------------------------------------------------------------%
%
% Statements
%

:- type mlds__statements == list(mlds__statement).

:- type mlds__statement
	--->	mlds__statement(
			mlds__stmt,
			mlds__context
		).

:- type mlds__stmt
	--->

	%
	% sequence
	%
		block(mlds__defns, list(mlds__statement))	

	%
	% iteration
	%
	;	while(mlds__rval, mlds__statement, bool)
			% the `bool' is true iff the loop is guaranteed
			% to iterate at least once -- in that case,
			% the compiler can generate a `do...while' loop
			% rather than a `while...' loop.

	%
	% selection (see also computed_goto)
	%

	;	if_then_else(mlds__rval, mlds__statement,
			maybe(mlds__statement))

		% This representation for switches is very general:
		% it allows switching on any type, and cases can match
		% on ranges as well as on values.
		% Many target languages only allow switches on ints or enums.
		% Some (e.g. C#) also allow switches on strings.
		% Most target languages only allow matching on values;
		% only some (e.g. GNU C) allow matching on ranges.
		% The MLDS code generator should only generate switches
		% that the target will support.
		%
		% Note that unlike C, MLDS cases do NOT fall through; if you
		% want to achieve that effect, you need to use an explicit goto.
	;	switch(
			% The value to switch on
			mlds__type,
			mlds__rval,

			% The range of possible values which the
			% value might take (if known)
			mlds__switch_range,

			% The different cases
			mlds__switch_cases,

			% What to do if none of the cases match
			mlds__switch_default
		)

	%
	% transfer of control
	%

	;	label(mlds__label)
			% Defines a label that can be used as the
			% target of calls, gotos, etc.

	;	goto(mlds__goto_target)
			% goto(Target)
			% Branch to the specified address.

	;	computed_goto(mlds__rval, list(mlds__label))
			% Evaluate rval, which should be an integer,
			% and jump to the (rval+1)th label in the list.
			% e.g. computed_goto(2, [A, B, C, D])
			% will branch to label C.


	%
	% function call/return
	%

	;	call(
			mlds__func_signature,	% signature of the func
			mlds__rval,		% the function to call
			maybe(mlds__rval),	% for method calls, this field
						% specifies the `this' object
			list(mlds__rval),	% ordinary function arguments
			list(mlds__lval),	% places to store the
						% function return value(s)
			call_kind		% indicates whether this
						% call is a tail call
		)

	;	return(list(mlds__rval))	% Some targets will not support
						% returning more than one value
						
	%
	% commits (a specialized form of exception handling)
	%

		% try_commit(Ref, GoalToTry, CommitHandlerGoal):
		%	Execute GoalToTry.  If GoalToTry exits via a
		%	`commit(Ref)' instruction, then execute
		%	CommitHandlerGoal.
		%
		% do_commit(Ref):
		%	Unwind the stack to the corresponding `try_commit'
		%	statement for Ref, and branch to the CommitHandlerGoal
		%	that was specified in that try_commit instruction.
		%
		% For both try_commit and do_commit instructions,
		% Ref should be the name of a local variable of type
		% mlds__commit_type.  (This variable can be used by
		% the back-end's implementation of do_commit and
		% try_commit to store information needed to unwind
		% the stack.)  There should be exactly
		% one try_commit instruction for each Ref.
		% do_commit(Ref) instructions should only be used
		% in goals called from the GoalToTry goal in the
		% try_commit instruction with the same Ref.
		%
		% The C and GCC back-ends require each try_commit
		% to be put in its own nested function, to avoid
		% problems with setjmp() and local vars not declared
		% volatile.  They also require each do_commit
		% to be put in its own function -- this is needed when
		% using __builtin_setjmp()/__builtin_longjmp() to
		% ensure that the call to __builtin_longjmp() is
		% not in the same function as the call to
		% __builtin_setjmp().
		%	
	;	try_commit(mlds__lval, mlds__statement, mlds__statement)
	;	do_commit(mlds__rval)

	%
	% exception handling
	%
/*********
XXX Full exception handling support is not yet implemented.

	% We use C++-style exceptions.
	% For C, the back-end can simulate them using setjmp/longjmp.
	%
	% XXX This is tentative -- the current definition may be
	% a bit too specific to C++-style exceptions.
	% It might not be a good choice for different target languages.

		% throw the specified exception
	;	throw(mlds__type, mlds__rval)

		% rethrow the current exception
		% (only valid inside an exception handler)
	;	rethrow

		% Execute the specified statement, and if it throws an exception,
		% and the exception matches any of the exception handlers,
		% then execute the first matching exception handler.
	;	try_catch(
			mlds__statement,
			list(mlds__exception_handler)
		)
**********/

	%
	% atomic statements
	%

	;	atomic(mlds__atomic_statement)
	
	.

%-----------------------------------------------------------------------------%
%
% Extra info for switches
%
	% The range of possible values which the
	% switch variable might take (if known)
:- type mlds__switch_range
	--->	range_unknown
	;	range(range_min::int, range_max::int).
			% From range_min to range_max, inclusive.

	% Each switch case consists of the conditions to match against,
	% and the statement to execute if the match succeeds.
	% Unlike C, cases do NOT fall through; if you want to achieve that
	% effect, you need to use an explicit goto.
:- type mlds__switch_cases == list(mlds__switch_case).
:- type mlds__switch_case == pair(mlds__case_match_conds, mlds__statement).

	% case_match_conds should be a _non-empty_ list of conditions;
	% if _any_ of the conditions match, this case will be selected.
:- type mlds__case_match_conds == list(mlds__case_match_cond).

	% A case_match_cond specifies when a switch case will be selected
:- type mlds__case_match_cond
	--->	match_value(mlds__rval)		% match_value(Val) matches if
						% the switch value is equal to
						% the specified Val

	;	match_range(mlds__rval, mlds__rval).  % match_range(Min, Max)
						% matches if the switch value
						% is between Min and Max,
						% inclusive.
						% Note that this should only be
						% used if the target supports
						% it; currently the C back-end
						% supports this only if you're
						% using the GNU C compiler.

	% The switch_default specifies what to do if none of the switch
	% conditions match.
:- type mlds__switch_default
	--->	default_is_unreachable		% The switch is exhaustive,
						% so the default case should
						% never be reached.

	;	default_do_nothing		% The default action is to
						% just fall through to the
						% statement after the switch.

	;	default_case(mlds__statement).	% The default is to execute
						% the specified statement.

%-----------------------------------------------------------------------------%
%
% Extra info for labels
%

:- type mlds__label == string.

:- type mlds__goto_target
	--->	label(mlds__label) % Branch to the specified label.
	;	break		% Branch to just after the end of the
				% immediately enclosing loop or switch,
				% just like a C/C++/Java `break' statement.
				% Not supported by all target languages.
	;	continue.	% Branch to the end of the loop body for
				% the immediately enclosing loop,
				% just like a C/C++/Java/C# `continue'
				% statement.
				% Not supported by all target languages.

%-----------------------------------------------------------------------------%
%
% Extra info for calls
%

	% The `call_kind' type indicates whether a call is a tail call
	% and whether the call is know to never return.
	%
	% Marking a call as a tail_call is intended as a hint to
	% the target language that this call can be implemented
	% by removing the caller's stack frame and jumping to the
	% destination, rather than a normal call.
	% However, the target code generator need not obey this hint;
	% it is permitted for the target code generator to ignore the
	% hint and generate code which does not remove the caller's
	% stack frame and/or which falls through to the following
	% statement.
	% 
:- type call_kind
	--->	no_return_call	% a call that never returns
				% (this is a special case of a tail call)
	;	tail_call	% a tail call
	;	ordinary_call	% just an ordinary call
	.

%-----------------------------------------------------------------------------%
%
% Extra info for exception handling
%

	% XXX This is tentative -- the current definition may be
	% a bit too specific to C++-style exceptions.
	% It might not be a good choice for different target languages.
:- type mlds__exception_handler
	--->	handler(
			maybe(mlds__type),
				% if `yes(T)', specifies the type of exceptions to catch
				% if `no', it means catch all exceptions

			maybe(string)
				% if `yes(Name)', gives the variable name to use for
				%	the exception value
				% if `no', then exception value will not be used
		).

%-----------------------------------------------------------------------------%

	%
	% atomic statements
	%
:- type mlds__atomic_statement

	--->	comment(string)
			% Insert a comment into the output code.

	;	assign(mlds__lval, mlds__rval)
			% assign(Location, Value):
			% Assign the value specified by rval to the location
			% specified by lval.

	%
	% heap management
	%

	;	delete_object(mlds__lval)
			% Compile time garbage collect (ie explicitly
			% deallocate) the memory used by the lval.

			% new_object(Target, Tag, Type,
			%	Size, CtorName, Args, ArgTypes):
			% Allocate a memory block of the given size,
			% initialize it with a new object of the given
			% type by calling the constructor with the specified
			% arguments, and put its address in the given lval,
			% possibly after tagging the address with a given tag.
			% (Some targets might not support tags.)
	;	new_object(
			mlds__lval,	% The target to assign the new object's
					% address to.
			maybe(mlds__tag),
					% A (primary) tag to tag the address
					% with before assigning the result to
					% the target.
			bool,		% Indicates whether or not there is
					% a secondary tag.  If so, it will
					% be stored as the first argument
					% in the argument list (see below).
			mlds__type,	% The type of the object being
					% allocated.
			maybe(mlds__rval),
					% The amount of memory that needs to
					% be allocated for the new object,
					% measured in words (NOT bytes!).
			maybe(ctor_name),
					% The name of the constructor to
					% invoke.
			list(mlds__rval),
					% The arguments to the constructor.
			list(mlds__type)
					% The types of the arguments to the
					% constructor. 
					%
					% Note that for --low-level-data, we box
					% all fields of objects created with
					% new_object, i.e. they are reprsented
					% with type mlds__generic_type.
					% We also do that for some fields
					% even for --high-level-data
					% (e.g. floating point fields for the
					% MLDS->C and MLDS->asm back-ends).
					% In such cases, the type here
					% should be mlds__generic_type;
					% it is the responsibility of the
					% HLDS->MLDS code generator
					% to insert code to box/unbox
					% the arguments.
					% 
		)

	;	gc_check
			% Check to see if we need to do a garbage collection,
			% and if so, do it.
			% This is used for accurate garbage collection
			% with the MLDS->C back-end.

	;	mark_hp(mlds__lval)
			% Tell the heap sub-system to store a marker
			% (for later use in restore_hp/1 instructions)
			% in the specified lval.
			%
			% It's OK for the target to treat this as a no-op,
			% and probably that is what most targets will do.

	;	restore_hp(mlds__rval)
			% The rval must be a marker as returned by mark_hp/1.
			% The effect is to deallocate all the memory which
			% was allocated since that call to mark_hp.
			%
			% It's OK for the target to treat this as a no-op,
			% and probably that is what most targets will do.

	%
	% trail management
	%

	;	trail_op(trail_op)

	%
	% foreign language interfacing
	%

	;	inline_target_code(target_lang, list(target_code_component))
			% Do whatever is specified by the
			% target_code_components, which can be any piece
			% of code in the specified target language (C,
			% assembler, or whatever) that does not have any
			% non-local flow of control.
			% This is implemented by embedding the target
			% code in the output stream of instructions or
			% statements.
	;	outline_foreign_proc(
				foreign_language,
					% the foreign language this code is
					% written in.
				list(outline_arg),
				list(mlds__lval),
					% where to store return value(s)
				string
					% the user's foreign language code
					% fragment
		)
			% Do whatever is specified by the string, which
			% can be any piece of code in the specified
			% foreign language (C#, managed C++, or
			% whatever).
			% This is implemented by calling an externally
			% defined function, which the backend must
			% generate the definition for (in some other
			% file perhaps) and calling it.
			% The lvals are use to generate the appropriate
			% forwarding code.
			% XXX we should also store the list of mlds__rvals
			% where the input values come from, and use
	.

	% Stores information about each argument to an outline_foreign_proc.
:- type outline_arg
	--->	in(
			mlds__type,
				% The type of the argument.
			string,
				% The name of the argument in the foreign code.
			mlds__rval
				% The rval which holds the value of this
				% argument.
		)
	;	out(
			mlds__type,
				% The type of the argument.
			string,
				% The name of the argument in the foreign code.
			mlds__lval
				% The lval where we are to place the result
				% calculated by the foreign code into.
		)
	;	unused.
		
	%
	% This is just a random selection of possible languages
	% that we might want to target...
	%
:- type target_lang
	--->	lang_C
	;	lang_GNU_C
	;	lang_C_minus_minus
	;	lang_asm
	;	lang_il
	;	lang_java
	;	lang_java_asm
	;	lang_java_bytecode
	.

:- type target_code_component
	--->	user_target_code(string, maybe(prog_context),
				target_code_attributes)
			% user_target_code holds C code from
			% the user's `pragma c_code' declaration
	;	raw_target_code(string, target_code_attributes)
			% raw_target_code holds C code that the
			% compiler has generated.  To ensure that
			% following `#line' directives work OK,
			% either the string in a raw_target_code must
			% end in `\n' (or `\n' followed by whitespace),
			% or the following target_code_component must be 
			% a `name(Name)' component, for which we do not
			% output #line directives.
	;	target_code_input(mlds__rval)
	;	target_code_output(mlds__lval)
	;	name(mlds__qualified_entity_name)
	.

:- type target_code_attributes == list(target_code_attribute).

:- type target_code_attribute
	--->	max_stack_size(int).
		% max_stack_size(Size):
		% This attribute declares the maximum stack usage of a
		% particular piece of code.  The unit that `Size' is measured
		% in depends upon foreign language being used.  Currently this
		% attribute is only used (and is in fact required) by the
		% `IL' foreign language interface, and is measured in units
		% of stack items.

	%
	% constructor id
	%
:- type ctor_name == mlds__qualified_ctor_id.
:- type mlds__ctor_id ---> ctor_id(mlds__class_name, arity).
:- type mlds__qualified_ctor_id ==
	mlds__fully_qualified_name(mlds__ctor_id).

	%
	% trail management
	% For documentation, see the corresponding LLDS instructions
	% in llds.m.
	%
:- type trail_op
	--->	store_ticket(mlds__lval)
	;	reset_ticket(mlds__rval, mlds__reset_trail_reason)
	;	discard_ticket
	;	prune_ticket
	;	mark_ticket_stack(mlds__lval)
	;	prune_tickets_to(mlds__rval)
% 	;	discard_tickets_to(mlds__rval)	% used only by the library
	.

%-----------------------------------------------------------------------------%

	%
	% A field_id represents some data within an object
	%

:- type field_id 
	--->		% offset(N) represents the field
			% at offset N Words.
	 	offset(mlds__rval)
	;		% named_field(Name, CtorType) represents the field
			% with the specified name.  The CtorType gives the
			% MLDS type for this particular constructor.
			% The type of the object is given by the PtrType
			% in the field(..) lval; CtorType may either be
			% the same as PtrType, or it may be a pointer to
			% a derived class.  In the latter case, the
			% MLDS->target code back-end is responsible
			% for inserting a downcast from PtrType to CtorType
			% before accessing the field.
		named_field(mlds__fully_qualified_name(field_name), mlds__type)
	.

:- type field_name == string.

	%
	% An mlds__var represents a variable or constant.
	%
:- type mlds__var == mlds__fully_qualified_name(mlds__var_name).
:- type mlds__var_name ---> 
		mlds__var_name(string, maybe(int)). 
		% var name and perhaps a unique number to be added as a
		% suffix where necessary.

	%
	% An lval represents a data location or variable that can be used
	% as the target of an assignment.
	%
:- type mlds__lval 

	%
	% values on the heap
	% or fields of a structure
	%
	--->	field(maybe(mlds__tag), mlds__rval, field_id, 
			mlds__type, mlds__type)
				% field(Tag, Address, FieldId, FieldType,
				%	PtrType)
				% selects a field of a compound term.
				% Address is a tagged pointer to a cell
				% on the heap; the position in the cell, 
				% FieldId, is represented either as a field 
				% name or a number of words offset. If Tag is 
				% yes, the arg gives the value of the tag; if 
				% it is no, the tag bits will have to be masked
				% off. The value of the tag should be given if
				% it is known, since this will lead to
				% faster code.
				% The FieldType is the type of the field.
				% The PtrType is the type of the pointer
				% from which we are fetching the field.
				%
				% Note that for --low-level-data, we box
				% all fields of objects created with
				% new_object, i.e. they are reprsented
				% with type mlds__generic_type.
				% We also do that for some fields
				% even for --high-level-data
				% (e.g. floating point fields for the
				% MLDS->C and MLDS->asm back-ends)
				% In such cases,
				% the type here should be mlds__generic_type,
				% not the actual type of the field.
				% If the actual type is different, then it
				% is the HLDS->MLDS code generator's
				% responsibility to insert the necessary
				% code to handle boxing/unboxing.

	%
	% values somewhere in memory
	% this is the deference operator (e.g. unary `*' in C)
	%
	;	mem_ref(mlds__rval, mlds__type)	
				% The rval should have originally come
				% from a mem_addr rval.
				% The type is the type of the value being
				% dereferenced

	%
	% variables
	% these may be local or they may come from some enclosing scope
	% the variable name should be fully qualified
	;	var(mlds__var, mlds__type) 
	
	.

%-----------------------------------------------------------------------------%
%
% Expressions
%

	% An rval is an expression that represents a value.
:- type mlds__rval	
	--->	lval(mlds__lval)
		% The value of an `lval' rval is just the value stored in
		% the specified lval.

	;	mkword(mlds__tag, mlds__rval)
		% Given a pointer and a tag, mkword returns a tagged pointer.
		%
		% (XXX It might be more consistent to make this a binary_op,
		% with the tag argument just being an rval, rather than
		% having `mkword' be a separate kind of rval.)

	;	const(mlds__rval_const)

	;	unop(mlds__unary_op, mlds__rval)

	;	binop(binary_op, mlds__rval, mlds__rval)

	;	mem_addr(mlds__lval)
		% The address of a variable, etc.

	;	self(mlds__type).
		% The equivalent of the `this' pointer in C++ with the
		% type of the object.  Note that this rval is valid iff
		% we are targetting an object oriented backend and we
		% are in an instance method (procedures which have the
		% per_instance flag set).

:- type mlds__unary_op
			% box(MLDSType)
			% convert from MLDSType to mlds__generic_type,
			% by boxing if necessary, or just casting if not
	--->	box(mlds__type)	

			% unbox(MLDSType)
			% convert from mlds__generic_type to MLDSType,
			% applying the inverse transformation to box/1,
			% i.e. unboxing if boxing was necessary,
			% and just casting otherwise.
	;	unbox(mlds__type)
		
			% cast(MLDSType):
			% Coerce the type of the rval to be MLDSType.
			% XXX it might be worthwhile adding the 
			% type that we cast from.
	;	cast(mlds__type)
	;	std_unop(builtin_ops__unary_op).

:- type mlds__rval_const
	--->	true
	;	false
	;	int_const(int)
	;	float_const(float)
	;	string_const(string)
			% A multi_string_const is a string containing
			% embedded NULs, whose real length is given
			% by the integer, and not the location of the
			% first null character.
	;	multi_string_const(int, string)
	;	code_addr_const(mlds__code_addr)
	;	data_addr_const(mlds__data_addr)
		% A null value, of the given type. 
		% Usually the type will be a pointer (mlds__ptr_type)
		% but it could also be string or a func_type.
		% (Null is not a valid value of type string
		% or func_type, but null values of those types
		% may be useful as placeholders in cases where the
		% value will never be used.)
	; 	null(mlds__type).


:- type mlds__code_addr
	--->	proc(mlds__qualified_proc_label, mlds__func_signature)
	;	internal(mlds__qualified_proc_label, mlds__func_sequence_num,
			mlds__func_signature).

:- type mlds__data_addr
	--->	data_addr(mlds_module_name, mlds__data_name).
			% module name; which var

:- type mlds__data == mlds__fully_qualified_name(mlds__data_name).

:- type mlds__data_name
	--->	var(mlds__var_name)
			% ordinary variables
	;	common(int)
			% Compiler-introduced constants representing
			% global constants.  These are called "common"
			% because they may be common sub-expressions.
	%
	% Stuff for handling polymorphism/RTTI and type classes.
	%
	;	rtti(rtti_id)
	%
	% Stuff for handling debugging and accurate garbage collection.
	% (Those features are not yet implemented for the MLDS back-end,
	% so these data_names are not yet used.)
	%
	;	module_layout
			% Layout information for the current module.
	;	proc_layout(mlds__proc_label)
			% Layout structure for the given procedure.
	;	internal_layout(mlds__proc_label, mlds__func_sequence_num)
			% Layout structure for the given internal MLDS func.
	%
	% Stuff for tabling
	%
	;	tabling_pointer(mlds__proc_label).
			% A variable that contains a pointer that points to
			% the table used to implement memoization, loopcheck
			% or minimal model semantics for the given procedure.

%-----------------------------------------------------------------------------%

%
% Note: the types `tag' and `reset_trail_reason' here are all
% defined exactly the same as the ones in llds.m.  The definitions are
% duplicated here because we don't want mlds.m to depend on llds.m.
% (Alternatively, we could move both these definitions into a new module
% imported by both mlds.m and llds.m, but these definitions are small enough
% and simple enough that I don't think it is worth creating a new module
% just for them.)
%

	% A tag should be a small non-negative integer.
:- type tag == int.

	% see runtime/mercury_trail.h
:- type reset_trail_reason
	--->	undo
	;	commit
	;	solve
	;	exception
	;	gc
	.

%-----------------------------------------------------------------------------%

:- type mlds__qualified_proc_label
	==	mlds__fully_qualified_name(mlds__proc_label).
:- type mlds__proc_label
	==	pair(mlds__pred_label, proc_id).

:- type mlds__qualified_pred_label
	==	mlds__fully_qualified_name(mlds__pred_label).

	% An mlds__pred_label is a structure containing information that
	% uniquely identifies a HLDS predicate within a given module.
	%
	% Note that a predicate can have both a declaring and a defining module.
	% The defining module is the module that provides the code for the
	% predicate, the declaring module contains the `:- pred' declaration.
	% When these are different, as for specialised versions of predicates
	% from `.opt' files, the defining module's name is added as a
	% qualifier to the pred name.
:- type mlds__pred_label
	--->	pred(
			pred_or_func,		% predicate/function
				% The declaring module,
				% if different to the defining module
			maybe(mercury_module_name),
			string,			% name
			arity,			% arity
			code_model,		% code model
			bool			% function without return value
						% (i.e. non-default mode)
		)
			
	;	special_pred(
			string,			% pred name
			maybe(mercury_module_name),
				% The module declaring the type,
				% if this is different to module defining
				% the special_pred.
			string,			% the type name
			arity			% the type arity
		).

%-----------------------------------------------------------------------------%
%-----------------------------------------------------------------------------%

:- implementation.

:- import_module backend_libs__foreign.
:- import_module hlds__error_util.
:- import_module hlds__hlds_data.
:- import_module libs__globals.
:- import_module parse_tree__modules.

:- import_module int, term, string, require.

%-----------------------------------------------------------------------------%

mlds__get_module_name(mlds(ModuleName, _, _, _)) = ModuleName.

%-----------------------------------------------------------------------------%

% Currently mlds__contexts just contain a prog_context.

:- type mlds__context ---> mlds__context(prog_context).

mlds__make_context(Context) = mlds__context(Context).

mlds__get_prog_context(mlds__context(Context)) = Context.

%-----------------------------------------------------------------------------%

% There is some special-case handling for arrays and foreign_types here.
% But apart from that,
% currently we return mlds__types that are just the same as Mercury types,
% except that we also store the type category, so that we
% can tell if the type is an enumeration or not, without
% needing to refer to the HLDS type_table.
% XXX It might be a better idea to get rid of the mercury_type/2
% MLDS type and instead fully convert all Mercury types to MLDS types.

mercury_type_to_mlds_type(ModuleInfo, Type) = MLDSType :-
	( 
		type_to_ctor_and_args(Type, TypeCtor, [ElemType]),
		TypeCtor = qualified(unqualified("array"), "array") - 1
	->
		MLDSElemType = mercury_type_to_mlds_type(ModuleInfo, ElemType),
		MLDSType = mlds__mercury_array_type(MLDSElemType)
	;
		type_to_ctor_and_args(Type, TypeCtor, _),
		module_info_types(ModuleInfo, Types),
		map__search(Types, TypeCtor, TypeDefn),
		hlds_data__get_type_defn_body(TypeDefn, Body),
		Body = foreign_type(foreign_type_body(MaybeIL, MaybeC,
			MaybeJava), _)
	->
		module_info_globals(ModuleInfo, Globals),
		globals__get_target(Globals, Target),
		( Target = c,
			( MaybeC = yes(CForeignType - _),
				ForeignType = c(CForeignType)
			; MaybeC = no,
				% This is checked by check_foreign_type
				% in make_hlds.
				unexpected(this_file,
				"mercury_type_to_mlds_type: No C foreign type")
			)
		; Target = il,
			( MaybeIL = yes(ILForeignType - _),
				ForeignType = il(ILForeignType)
			; MaybeIL = no,
				% This is checked by check_foreign_type
				% in make_hlds.
				unexpected(this_file,
				"mercury_type_to_mlds_type: No IL foreign type")
			)
		; Target = java,
			( MaybeJava = yes(JavaForeignType - _),
				ForeignType = java(JavaForeignType)
			; MaybeJava = no,
				% This is checked by check_foreign_type
				% in make_hlds.
				unexpected(this_file,
				"mercury_type_to_mlds_type: No Java foreign type")
			)	
		; Target = asm,
			( MaybeC = yes(CForeignType - _),
				ForeignType = c(CForeignType)
			; MaybeC = no,
				% XXX This ought to be checked by the
				% front-end, e.g. check_foreign_type
				% in make_hlds.
				sorry(this_file,
				"mercury_type_to_mlds_type: No C foreign type")
			)
		),
		MLDSType = mlds__foreign_type(ForeignType)
	;
		classify_type(ModuleInfo, Type) = Category,
		ExportedType = to_exported_type(ModuleInfo, Type),
		MLDSType = mercury_type(Type, Category, ExportedType)
	).

%-----------------------------------------------------------------------------%

mlds__get_func_signature(func_params(Parameters, RetTypes)) =
		func_signature(ParamTypes, RetTypes) :-
	ParamTypes = mlds__get_arg_types(Parameters).

mlds__get_arg_types(Parameters) = ArgTypes :-
	GetArgType = (func(mlds__argument(_, Type, _)) = Type),
	ArgTypes = list__map(GetArgType, Parameters).

%-----------------------------------------------------------------------------%

% A mercury module name consists of two parts.  One part is the package
% which the module name is defined in, and the other part is the actual
% module name.  For example the module name System.XML could be defined
% in the package XML.
%
% Note that modules in the Mercury standard library map get a `mercury'
% prefix e.g. `mercury.builtin', `mercury.io', `mercury.std_util', etc.,
% when mapped to MLDS package names.

% :- type mlds_module_name == prog_data__module_name.
:- type mlds_module_name
	---> name(
		package_name	:: prog_data__module_name,
		module_name	:: prog_data__module_name
	).

mercury_module_and_package_name_to_mlds(MLDS_Package, MercuryModule)
	= name(MLDS_Package, MercuryModule).

mercury_module_name_to_mlds(MercuryModule) = name(MLDS_Package, MLDS_Package) :-
	(
		MercuryModule = unqualified(ModuleName),
		mercury_std_library_module(ModuleName)
	->
		MLDS_Package = qualified(unqualified("mercury"), ModuleName)
	;
		MLDS_Package = MercuryModule
	).

is_std_lib_module(Module, UnqualifiedName) :-
	Name = Module ^ module_name,
	( Name = unqualified(UnqualifiedName)
	; Name = qualified(unqualified("mercury"), UnqualifiedName)
	),
	mercury_std_library_module(UnqualifiedName).

mlds_module_name_to_sym_name(Module) = Module ^ module_name.

mlds_module_name_to_package_name(Module) = Module ^ package_name.

mlds__append_class_qualifier(name(Package, Module), ClassName, ClassArity) =
		name(Package, qualified(Module, ClassQualifier)) :-
	string__format("%s_%d", [s(ClassName), i(ClassArity)],
		ClassQualifier).

mlds__append_wrapper_class(Name) = mlds__append_name(Name, wrapper_class_name).

mlds__append_name(name(Package, Module), Name)
	= name(Package, qualified(Module, Name)).

wrapper_class_name = "mercury_code".

%-----------------------------------------------------------------------------%

%
% We represent the set of declaration flags as a bunch of bit-fields packed
% into a single int.
%
:- type mlds__decl_flags == int.

%
% Here we define which bits are used to store each bitfield.
%
% It would be nicer to use a language builtin, e.g. index/2, for these.
% But currently builtin__index/2 does not work in the reverse mode,
% and you can't use std_util__construct/4 since that numbers the
% alternatives in a different order than builtin__index/2.
%
% It would also be nice to use a typeclass:
%	:- typeclass bitfield(T) where [
%		func bits(T) = int,
%		func mask(T::unused) = (int::out) is det
%	].
% But currently that is too cumbersome, since you can't define class
% methods inline.
%
% On the other hand, doing it manually may be more efficient than either
% of those two approaches.
%

:- func access_bits(access) = int.
:- mode access_bits(in) = out is det.
:- mode access_bits(out) = in is semidet.
access_bits(public)  	= 0x00.
access_bits(private) 	= 0x01.
access_bits(protected)	= 0x02.
access_bits(default)	= 0x03.
access_bits(local) 	= 0x04.
% 0x5 - 0x7 reserved

:- func access_mask = int.
access_mask = 0x07.

:- func per_instance_bits(per_instance) = int.
:- mode per_instance_bits(in) = out is det.
:- mode per_instance_bits(out) = in is semidet.
per_instance_bits(one_copy) 	= 0x00.
per_instance_bits(per_instance)	= 0x08.

:- func per_instance_mask = int.
per_instance_mask = per_instance_bits(per_instance).

:- func virtuality_bits(virtuality) = int.
:- mode virtuality_bits(in) = out is det.
:- mode virtuality_bits(out) = in is semidet.
virtuality_bits(non_virtual) 	= 0x00.
virtuality_bits(virtual)	= 0x10.

:- func virtuality_mask = int.
virtuality_mask = virtuality_bits(virtual).

:- func finality_bits(finality) = int.
:- mode finality_bits(in) = out is det.
:- mode finality_bits(out) = in is semidet.
finality_bits(overridable) 	= 0x00.
finality_bits(final)		= 0x20.

:- func finality_mask = int.
finality_mask = finality_bits(final).

:- func constness_bits(constness) = int.
:- mode constness_bits(in) = out is det.
:- mode constness_bits(out) = in is semidet.
constness_bits(modifiable) 	= 0x00.
constness_bits(const)		= 0x40.

:- func constness_mask = int.
constness_mask = constness_bits(const).

:- func abstractness_bits(abstractness) = int.
:- mode abstractness_bits(in) = out is det.
:- mode abstractness_bits(out) = in is semidet.
abstractness_bits(abstract) 	= 0x00.
abstractness_bits(concrete)	= 0x80.

:- func abstractness_mask = int.
abstractness_mask = abstractness_bits(concrete).

%
% Here we define the functions to lookup a member of the set.
%

access(Flags) = promise_det(pred(Access::out) is semidet :-
	Flags /\ access_mask = access_bits(Access)).

per_instance(Flags) = promise_det(pred(PerInstance::out) is semidet :-
	Flags /\ per_instance_mask = per_instance_bits(PerInstance)).

virtuality(Flags) = promise_det(pred(Virtuality::out) is semidet :-
	Flags /\ virtuality_mask = virtuality_bits(Virtuality)).

finality(Flags) = promise_det(pred(Finality::out) is semidet :-
	Flags /\ finality_mask = finality_bits(Finality)).

constness(Flags) = promise_det(pred(Constness::out) is semidet :-
	Flags /\ constness_mask = constness_bits(Constness)).

abstractness(Flags) = promise_det(pred(Abstractness::out) is semidet :-
	Flags /\ abstractness_mask = abstractness_bits(Abstractness)).

:- func promise_det(pred(T)) = T.
:- mode promise_det(pred(out) is semidet) = out is det.

promise_det(Pred) = X :-
	(if Pred(X0) then X = X0 else error("promise_det failed")).


%
% Here we define the functions to set a member of the set.
%

set_access(Flags, Access) =
	Flags /\ \access_mask \/ access_bits(Access).

set_per_instance(Flags, PerInstance) =
	Flags /\ \per_instance_mask \/ per_instance_bits(PerInstance).

set_virtuality(Flags, Virtuality) =
	Flags /\ \virtuality_mask \/ virtuality_bits(Virtuality).

set_finality(Flags, Finality) =
	Flags /\ \finality_mask \/ finality_bits(Finality).

set_constness(Flags, Constness) =
	Flags /\ \constness_mask \/ constness_bits(Constness).

set_abstractness(Flags, Abstractness) =
	Flags /\ \abstractness_mask \/ abstractness_bits(Abstractness).

init_decl_flags(Access, PerInstance, Virtuality, Finality, Constness,
		Abstractness) =
	access_bits(Access) \/
	per_instance_bits(PerInstance) \/
	virtuality_bits(Virtuality) \/
	finality_bits(Finality) \/
	constness_bits(Constness) \/
	abstractness_bits(Abstractness).

%-----------------------------------------------------------------------------%

:- func this_file = string.
this_file = "mlds.m".

%-----------------------------------------------------------------------------%