blob: 273d6eded93c40659a6788dd0e8c8682a6773121 [file] [log] [blame]
------------------------------------------------------------------------------
-- --
-- GNAT COMPILER COMPONENTS --
-- --
-- E X P _ D B U G --
-- --
-- S p e c --
-- --
-- Copyright (C) 1996-2022, Free Software Foundation, Inc. --
-- --
-- GNAT is free software; you can redistribute it and/or modify it under --
-- terms of the GNU General Public License as published by the Free Soft- --
-- ware Foundation; either version 3, or (at your option) any later ver- --
-- sion. GNAT is distributed in the hope that it will be useful, but WITH- --
-- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY --
-- or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License --
-- for more details. You should have received a copy of the GNU General --
-- Public License distributed with GNAT; see file COPYING3. If not, go to --
-- http://www.gnu.org/licenses for a complete copy of the license. --
-- --
-- GNAT was originally developed by the GNAT team at New York University. --
-- Extensive contributions were provided by Ada Core Technologies Inc. --
-- --
------------------------------------------------------------------------------
-- Expand routines for the generation of special declarations used by the
-- debugger. In accordance with the DWARF specification, certain type names
-- may also be encoded to provide additional information to the debugger, but
-- this practice is being deprecated and some encodings described below are no
-- longer generated by default (they are marked OBSOLETE).
with Namet; use Namet;
with Types; use Types;
with Uintp; use Uintp;
package Exp_Dbug is
-----------------------------------------------------
-- Encoding and Qualification of Names of Entities --
-----------------------------------------------------
-- This section describes how the names of entities are encoded in the
-- generated debugging information.
-- An entity in Ada has a name of the form X.Y.Z ... E where X,Y,Z are the
-- enclosing scopes (not including Standard at the start).
-- The encoding of the name follows this basic qualified naming scheme,
-- where the encoding of individual entity names is as described in Namet
-- (i.e. in particular names present in the original source are folded to
-- all lower case, with upper half and wide characters encoded as described
-- in Namet). Upper case letters are used only for entities generated by
-- the compiler.
-- There are two cases, global entities, and local entities. In more formal
-- terms, local entities are those which have a dynamic enclosing scope,
-- and global entities are at the library level, except that we always
-- consider procedures to be global entities, even if they are nested
-- (that's because at the debugger level a procedure name refers to the
-- code, and the code is indeed a global entity, including the case of
-- nested procedures.) In addition, we also consider all types to be global
-- entities, even if they are defined within a procedure.
-- The reason for treating all type names as global entities is that a
-- number of our type encodings work by having related type names, and we
-- need the full qualification to keep this unique.
-- For global entities, the encoded name includes all components of the
-- fully expanded name (but omitting Standard at the start). For example,
-- if a library-level child package P.Q has an embedded package R, and
-- there is an entity in this embedded package whose name is S, the encoded
-- name will include the components p.q.r.s.
-- For local entities, the encoded name only includes the components up to
-- the enclosing dynamic scope (other than a block). At run time, such a
-- dynamic scope is a subprogram, and the debugging formats know about
-- local variables of procedures, so it is not necessary to have full
-- qualification for such entities. In particular this means that direct
-- local variables of a procedure are not qualified.
-- For Ghost entities, the encoding adds a prefix "___ghost_" to aid the
-- detection of leaks of Ignored Ghost entities in the "living" space.
-- Ignored Ghost entities and any code associated with them should be
-- removed by the compiler in a post-processing pass. As a result,
-- object files should not contain any occurrences of this prefix.
-- As an example of the local name convention, consider a procedure V.W
-- with a local variable X, and a nested block Y containing an entity Z.
-- The fully qualified names of the entities X and Z are:
-- V.W.X
-- V.W.Y.Z
-- but since V.W is a subprogram, the encoded names will end up
-- encoding only
-- x
-- y.z
-- The separating dots are translated into double underscores
-----------------------------
-- Handling of Overloading --
-----------------------------
-- The above scheme is incomplete for overloaded subprograms, since
-- overloading can legitimately result in case of two entities with
-- exactly the same fully qualified names. To distinguish between
-- entries in a set of overloaded subprograms, the encoded names are
-- serialized by adding the suffix:
-- __nn (two underscores)
-- where nn is a serial number (2 for the second overloaded function,
-- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no
-- suffix implies the first instance).
-- These names are prefixed by the normal full qualification. So for
-- example, the third instance of the subprogram qrs in package yz
-- would have the name:
-- yz__qrs__3
-- A more subtle case arises with entities declared within overloaded
-- subprograms. If we have two overloaded subprograms, and both declare
-- an entity xyz, then the fully expanded name of the two xyz's is the
-- same. To distinguish these, we add the same __n suffix at the end of
-- the inner entity names.
-- In more complex cases, we can have multiple levels of overloading,
-- and we must make sure to distinguish which final declarative region
-- we are talking about. For this purpose, we use a more complex suffix
-- which has the form:
-- __nn_nn_nn ...
-- where the nn values are the homonym numbers as needed for any of the
-- qualifying entities, separated by a single underscore. If all the nn
-- values are 1, the suffix is omitted, Otherwise the suffix is present
-- (including any values of 1). The following example shows how this
-- suffixing works.
-- package body Yz is
-- procedure Qrs is -- Name is yz__qrs
-- procedure Tuv is ... end; -- Name is yz__qrs__tuv
-- begin ... end Qrs;
-- procedure Qrs (X: Int) is -- Name is yz__qrs__2
-- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1
-- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2
-- begin ... end Tuv;
-- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3
-- type m is new float; -- Name is yz__qrs__tuv__m__2_3
-- begin ... end Tuv;
-- begin ... end Qrs;
-- end Yz;
--------------------
-- Operator Names --
--------------------
-- The above rules applied to operator names would result in names with
-- quotation marks, which are not typically allowed by assemblers and
-- linkers, and even if allowed would be odd and hard to deal with. To
-- avoid this problem, operator names are encoded as follows:
-- Oabs abs
-- Oand and
-- Omod mod
-- Onot not
-- Oor or
-- Orem rem
-- Oxor xor
-- Oeq =
-- One /=
-- Olt <
-- Ole <=
-- Ogt >
-- Oge >=
-- Oadd +
-- Osubtract -
-- Oconcat &
-- Omultiply *
-- Odivide /
-- Oexpon **
-- These names are prefixed by the normal full qualification, and
-- suffixed by the overloading identification. So for example, the
-- second operator "=" defined in package Extra.Messages would have
-- the name:
-- extra__messages__Oeq__2
----------------------------------
-- Resolving Other Name Clashes --
----------------------------------
-- It might be thought that the above scheme is complete, but in Ada 95,
-- full qualification is insufficient to uniquely identify an entity in
-- the program, even if it is not an overloaded subprogram. There are
-- two possible confusions:
-- a.b
-- interpretation 1: entity b in body of package a
-- interpretation 2: child procedure b of package a
-- a.b.c
-- interpretation 1: entity c in child package a.b
-- interpretation 2: entity c in nested package b in body of a
-- It is perfectly legal in both cases for both interpretations to be
-- valid within a single program. This is a bit of a surprise since
-- certainly in Ada 83, full qualification was sufficient, but not in
-- Ada 95. The result is that the above scheme can result in duplicate
-- names. This would not be so bad if the effect were just restricted
-- to debugging information, but in fact in both the above cases, it
-- is possible for both symbols to be external names, and so we have
-- a real problem of name clashes.
-- To deal with this situation, we provide two additional encoding
-- rules for names:
-- First: all library subprogram names are preceded by the string
-- _ada_ (which causes no duplications, since normal Ada names can
-- never start with an underscore. This not only solves the first
-- case of duplication, but also solves another pragmatic problem
-- which is that otherwise Ada procedures can generate names that
-- clash with existing system function names. Most notably, we can
-- have clashes in the case of procedure Main with the C main that
-- in some systems is always present.
-- Second, for the case where nested packages declared in package
-- bodies can cause trouble, we add a suffix which shows which
-- entities in the list are body-nested packages, i.e. packages
-- whose spec is within a package body. The rules are as follows,
-- given a list of names in a qualified name name1.name2....
-- If none are body-nested package entities, then there is no suffix
-- If at least one is a body-nested package entity, then the suffix
-- is X followed by a string of b's and n's (b = body-nested package
-- entity, n = not a body-nested package).
-- There is one element in this string for each entity in the encoded
-- expanded name except the first (the rules are such that the first
-- entity of the encoded expanded name can never be a body-nested'
-- package. Trailing n's are omitted, as is the last b (there must
-- be at least one b, or we would not be generating a suffix at all).
-- For example, suppose we have
-- package x is
-- pragma Elaborate_Body;
-- m1 : integer; -- #1
-- end x;
-- package body x is
-- package y is m2 : integer; end y; -- #2
-- package body y is
-- package z is r : integer; end z; -- #3
-- end;
-- m3 : integer; -- #4
-- end x;
-- package x.y is
-- pragma Elaborate_Body;
-- m2 : integer; -- #5
-- end x.y;
-- package body x.y is
-- m3 : integer; -- #6
-- procedure j is -- #7
-- package k is
-- z : integer; -- #8
-- end k;
-- begin
-- null;
-- end j;
-- end x.y;
-- procedure x.m3 is begin null; end; -- #9
-- Then the encodings would be:
-- #1. x__m1 (no BNPE's in sight)
-- #2. x__y__m2X (y is a BNPE)
-- #3. x__y__z__rXb (y is a BNPE, so is z)
-- #4. x__m3 (no BNPE's in sight)
-- #5. x__y__m2 (no BNPE's in sight)
-- #6. x__y__m3 (no BNPE's in signt)
-- #7. x__y__j (no BNPE's in sight)
-- #8. k__z (no BNPE's, only up to procedure)
-- #9 _ada_x__m3 (library-level subprogram)
-- Note that we have instances here of both kind of potential name
-- clashes, and the above examples show how the encodings avoid the
-- clash as follows:
-- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library
-- level subprogram, so it is preceded by the string _ada_ which acts
-- to distinguish it from the package body entity.
-- Lines #2 and #5 both refer to the entity x.y.m2, but the first
-- instance is inside the body-nested package y, so there is an X
-- suffix to distinguish it from the child library entity.
-- Note that enumeration literals never need Xb type suffixes, since
-- they are never referenced using global external names.
---------------------
-- Interface Names --
---------------------
-- Note: if an interface name is present, then the external name is
-- taken from the specified interface name. Given current limitations of
-- the gcc backend, this means that the debugging name is also set to
-- the interface name, but conceptually, it would be possible (and
-- indeed desirable) to have the debugging information still use the Ada
-- name as qualified above, so we still fully qualify the name in the
-- front end.
-------------------------------------
-- Encodings Related to Task Types --
-------------------------------------
-- Each task object defined by a single task declaration is associated
-- with a prefix that is used to qualify procedures defined in that
-- task. Given
--
-- package body P is
-- task body TaskObj is
-- procedure F1 is ... end;
-- begin
-- B;
-- end TaskObj;
-- end P;
--
-- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1.
-- The body, B, is contained in a subprogram whose name is
-- p__taskobjTKB.
------------------------------------------
-- Encodings Related to Protected Types --
------------------------------------------
-- Each protected type has an associated record type, that describes
-- the actual layout of the private data. In addition to the private
-- components of the type, the Corresponding_Record_Type includes one
-- component of type Protection, which is the actual lock structure.
-- The run-time size of the protected type is the size of the corres-
-- ponding record.
-- For a protected type prot, the Corresponding_Record_Type is encoded
-- as protV.
-- The operations of a protected type are encoded as follows: each
-- operation results in two subprograms, a locking one that is called
-- from outside of the object, and a non-locking one that is used for
-- calls from other operations on the same object. The locking operation
-- simply acquires the lock, and then calls the non-locking version.
-- The names of all of these have a prefix constructed from the name of
-- the type, and a suffix which is P or N, depending on whether this is
-- the protected/non-locking version of the operation.
-- Operations generated for protected entries follow the same encoding.
-- Each entry results in two subprograms: a procedure that holds the
-- entry body, and a function that holds the evaluation of the barrier.
-- The names of these subprograms include the prefix '_E' or '_B' res-
-- pectively. The names also include a numeric suffix to render them
-- unique in the presence of overloaded entries.
-- Given the declaration:
-- protected type Lock is
-- function Get return Integer;
-- procedure Set (X: Integer);
-- entry Update (Val : Integer);
-- private
-- Value : Integer := 0;
-- end Lock;
-- the following operations are created:
-- lock_getN
-- lock_getP,
-- lock_setN
-- lock_setP
-- lock_update_E1s
-- lock_udpate_B2s
-- If the protected type implements at least one interface, the
-- following additional operations are created:
-- lock_get
-- lock_set
-- These operations are used to ensure overriding of interface level
-- subprograms and proper dispatching on interface class-wide objects.
-- The bodies of these operations contain calls to their respective
-- protected versions:
-- function lock_get return Integer is
-- begin
-- return lock_getP;
-- end lock_get;
-- procedure lock_set (X : Integer) is
-- begin
-- lock_setP (X);
-- end lock_set;
----------------------------------------------------
-- Conversion between Entities and External Names --
----------------------------------------------------
procedure Get_External_Name
(Entity : Entity_Id;
Has_Suffix : Boolean := False;
Suffix : String := "");
-- Set Name_Buffer and Name_Len to the external name of the entity. The
-- external name is the Interface_Name, if specified, unless the entity
-- has an address clause or Has_Suffix is true.
--
-- If the Interface is not present, or not used, the external name is the
-- concatenation of:
--
-- - the string "_ada_", if the entity is a library subprogram,
-- - the names of any enclosing scopes, each followed by "__",
-- or "X_" if the next entity is a subunit)
-- - the name of the entity
-- - the string "$" (or "__" if target does not allow "$"), followed
-- by homonym suffix, if the entity is an overloaded subprogram
-- or is defined within an overloaded subprogram.
-- - the string "___" followed by Suffix if Has_Suffix is true.
--
-- Note that a call to this procedure has no effect if we are not
-- generating code, since the necessary information for computing the
-- proper external name is not available in this case.
-- WARNING: There is a matching C declaration of this subprogram in fe.h
-------------------------------------
-- Encoding for translation into C --
-------------------------------------
-- In Modify_Tree_For_C mode we must add encodings to dismabiguate cases
-- where Ada block structure cannot be directly translated. These cases
-- are as follows:
-- a) A loop variable may hide a homonym in an enclosing block
-- b) A block-local variable may hide a homonym in an enclosing block
-- In C these constructs are not scopes and we must distinguish the names
-- explicitly. In the first case we create a qualified name with the suffix
-- 'L', in the second case with a suffix 'B'.
--------------------------------------------
-- Subprograms for Handling Qualification --
--------------------------------------------
procedure Qualify_Entity_Names (N : Node_Id);
-- Given a node N, that represents a block, subprogram body, or package
-- body or spec, or protected or task type, sets a fully qualified name
-- for the defining entity of given construct, and also sets fully
-- qualified names for all enclosed entities of the construct (using
-- First_Entity/Next_Entity). Note that the actual modifications of the
-- names is postponed till a subsequent call to Qualify_All_Entity_Names.
-- Note: this routine does not deal with prepending _ada_ to library
-- subprogram names. The reason for this is that we only prepend _ada_
-- to the library entity itself, and not to names built from this name.
procedure Qualify_All_Entity_Names;
-- When Qualify_Entity_Names is called, no actual name changes are made,
-- i.e. the actual calls to Qualify_Entity_Name are deferred until a call
-- is made to this procedure. The reason for this deferral is that when
-- names are changed semantic processing may be affected. By deferring
-- the changes till just before gigi is called, we avoid any concerns
-- about such effects. Gigi itself does not use the names except for
-- output of names for debugging purposes (which is why we are doing
-- the name changes in the first place).
-- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are
-- useful to remove qualification from a name qualified by the call to
-- Qualify_All_Entity_Names.
--------------------------------
-- Handling of Numeric Values --
--------------------------------
-- All numeric values here are encoded as strings of decimal digits. Only
-- integer values need to be encoded. A negative value is encoded as the
-- corresponding positive value followed by a lower case m for minus to
-- indicate that the value is negative (e.g. 2m for -2).
------------------------
-- Encapsulated Types --
------------------------
-- In some cases, the compiler may encapsulate a type by wrapping it in a
-- record. For example, this is used when a size or alignment specification
-- requires a larger type. Consider:
-- type x is mod 2 ** 64;
-- for x'size use 256;
-- In this case, the compiler generates a record type x___PAD, which has
-- a single field whose name is F. This single field is 64-bit long and
-- contains the actual value. This kind of padding is used when the logical
-- value to be stored is shorter than the object in which it is allocated.
-- A similar encapsulation is done for some packed array types, in which
-- case the record type is x___JM and the field name is OBJECT. This is
-- used in the case of a packed array stored using modular representation
-- (see the section on representation of packed array objects). In this
-- case the wrapping is used to achieve correct positioning of the packed
-- array value (left/right justified in its field depending on endianness).
-- When the debugger sees an object of a type whose name has a suffix of
-- ___PAD or ___JM, the type will be a record containing a single field,
-- and the name of that field will be all upper case. In this case, it
-- should look inside to get the value of the inner field, and neither
-- the outer structure name, nor the field name should appear when the
-- value is printed.
-- Similarly, when the debugger sees a record named REP being the type of
-- a field inside another record type, it should treat the fields inside
-- REP as being part of the outer record (this REP field is only present
-- for code generation purposes). The REP record should not appear in the
-- values printed by the debugger.
--------------------
-- Implicit Types --
--------------------
-- The compiler creates implicit type names in many situations where a
-- type is present semantically, but no specific name is present. For
-- example:
-- S : Integer range M .. N;
-- Here the subtype of S is not integer, but rather an anonymous subtype
-- of Integer. Where possible, the compiler generates names for such
-- anonymous types that are related to the type from which the subtype
-- is obtained as follows:
-- T name suffix
-- where name is the name from which the subtype is obtained, using
-- lower case letters and underscores, and suffix starts with an upper
-- case letter. For example the name for the above declaration might be:
-- TintegerS4b
-- If the debugger is asked to give the type of an entity and the type
-- has the form T name suffix, it is probably appropriate to just use
-- "name" in the response since this is what is meaningful to the
-- programmer.
-------------------
-- Modular Types --
-------------------
-- A type declared
-- type x is mod N;
-- is encoded as a subrange of an unsigned base type with lower bound zero
-- and upper bound N - 1. Thus we give these types a somewhat nonstandard
-- interpretation: the standard interpretation would not, in general, imply
-- that arithmetic operations on type x are performed modulo N (especially
-- not when N is not a power of 2).
--------------------------------------
-- Tagged Types and Type Extensions --
--------------------------------------
-- A type D derived from a tagged type P has a field named "_parent" of
-- type P that contains its inherited fields. The type of this field is
-- usually P, but may be a more distant ancestor, if P is a null extension
-- of that type.
-- The type tag of a tagged type is a field named "_tag" of a pointer type.
-- If the type is derived from another tagged type, its _tag field is found
-- in its _parent field.
------------------------------------
-- Type Name Encodings (OBSOLETE) --
------------------------------------
-- In the following typ is the name of the type as normally encoded by the
-- debugger rules, i.e. a non-qualified name, all in lower case, with
-- standard encoding of upper half and wide characters.
-----------------------
-- Fixed-Point Types --
-----------------------
-- Fixed-point types are encoded using a suffix that indicates the
-- delta and small values. The actual type itself is a normal integer
-- type.
-- typ___XF_nn_dd
-- typ___XF_nn_dd_nn_dd
-- The first form is used when small = delta. The value of delta (and
-- small) is given by the rational nn/dd, where nn and dd are decimal
-- integers.
--
-- The second form is used if the small value is different from the
-- delta. In this case, the first nn/dd rational value is for delta,
-- and the second value is for small.
--------------------
-- Discrete Types --
--------------------
-- Discrete types are coded with a suffix indicating the range in the
-- case where one or both of the bounds are discriminants or variable.
-- Note: at the current time, we also encode compile time known bounds
-- if they do not match the natural machine type bounds, but this may
-- be removed in the future, since it is redundant for most debugging
-- formats. However, we do not ever need XD encoding for enumeration
-- base types, since here it is always clear what the bounds are from
-- the total number of enumeration literals.
-- typ___XD
-- typ___XDL_lowerbound
-- typ___XDU_upperbound
-- typ___XDLU_lowerbound__upperbound
-- If a discrete type is a natural machine type (i.e. its bounds
-- correspond in a natural manner to its size), then it is left
-- unencoded. The above encoding forms are used when there is a
-- constrained range that does not correspond to the size or that
-- has discriminant references or other compile time known bounds.
-- The first form is used if both bounds are dynamic, in which case two
-- constant objects are present whose names are typ___L and typ___U in
-- the same scope as typ, and the values of these constants indicate
-- the bounds. As far as the debugger is concerned, these are simply
-- variables that can be accessed like any other variables. In the
-- enumeration case, these values correspond to the Enum_Rep values for
-- the lower and upper bounds.
-- The second form is used if the upper bound is dynamic, but the lower
-- bound is either constant or depends on a discriminant of the record
-- with which the type is associated. The upper bound is stored in a
-- constant object of name typ___U as previously described, but the
-- lower bound is encoded directly into the name as either a decimal
-- integer, or as the discriminant name.
-- The third form is similarly used if the lower bound is dynamic, but
-- the upper bound is compile time known or a discriminant reference,
-- in which case the lower bound is stored in a constant object of name
-- typ___L, and the upper bound is encoded directly into the name as
-- either a decimal integer, or as the discriminant name.
-- The fourth form is used if both bounds are discriminant references
-- or compile time known values, with the encoding first for the lower
-- bound, then for the upper bound, as previously described.
------------------
-- Biased Types --
------------------
-- Only discrete types can be biased, and the fact that they are biased
-- is indicated by a suffix of the form:
-- typ___XB_lowerbound__upperbound
-- Here lowerbound and upperbound are decimal integers, with the usual
-- (postfix "m") encoding for negative numbers. Biased types are only
-- possible where the bounds are compile time known, and the values are
-- represented as unsigned offsets from the lower bound given. For
-- example:
-- type Q is range 10 .. 15;
-- for Q'size use 3;
-- The size clause will force values of type Q in memory to be stored
-- in biased form (e.g. 11 will be represented by the bit pattern 001).
----------------------------------------------
-- Record Types with Variable-Length Fields --
----------------------------------------------
-- The debugging formats do not fully support these types, and indeed
-- some formats simply generate no useful information at all for such
-- types. In order to provide information for the debugger, gigi creates
-- a parallel type in the same scope with one of the names
-- type___XVE
-- type___XVU
-- The former name is used for a record and the latter for the union
-- that is made for a variant record (see below) if that record or union
-- has a field of variable size or if the record or union itself has a
-- variable size. These encodings suffix any other encodings that that
-- might be suffixed to the type name.
-- The idea here is to provide all the needed information to interpret
-- objects of the original type in the form of a "fixed up" type, which
-- is representable using the normal debugging information.
-- There are three cases to be dealt with. First, some fields may have
-- variable positions because they appear after variable-length fields.
-- To deal with this, we encode *all* the field bit positions of the
-- special ___XV type in a non-standard manner.
-- The idea is to encode not the position, but rather information that
-- allows computing the position of a field from the position of the
-- previous field. The algorithm for computing the actual positions of
-- all fields and the length of the record is as follows. In this
-- description, let P represent the current bit position in the record.
-- 1. Initialize P to 0
-- 2. For each field in the record:
-- 2a. If an alignment is given (see below), then round P up, if
-- needed, to the next multiple of that alignment.
-- 2b. If a bit position is given, then increment P by that amount
-- (that is, treat it as an offset from the end of the preceding
-- record).
-- 2c. Assign P as the actual position of the field
-- 2d. Compute the length, L, of the represented field (see below)
-- and compute P'=P+L. Unless the field represents a variant part
-- (see below and also Variant Record Encoding), set P to P'.
-- The alignment, if present, is encoded in the field name of the
-- record, which has a suffix:
-- fieldname___XVAnn
-- where the nn after the XVA indicates the alignment value in storage
-- units. This encoding is present only if an alignment is present.
-- The size of the record described by an XVE-encoded type (in bits) is
-- generally the maximum value attained by P' in step 2d above, rounded
-- up according to the record's alignment.
-- Second, the variable-length fields themselves are represented by
-- replacing the type by a special access type. The designated type of
-- this access type is the original variable-length type, and the fact
-- that this field has been transformed in this way is signalled by
-- encoding the field name as:
-- field___XVL
-- where field is the original field name. If a field is both
-- variable-length and also needs an alignment encoding, then the
-- encodings are combined using:
-- field___XVLnn
-- Note: the reason that we change the type is so that the resulting
-- type has no variable-length fields. At least some of the formats used
-- for debugging information simply cannot tolerate variable- length
-- fields, so the encoded information would get lost.
-- Third, in the case of a variant record, the special union that
-- contains the variants is replaced by a normal C union. In this case,
-- the positions are all zero.
-- Discriminants appear before any variable-length fields that depend on
-- them, with one exception. In some cases, a discriminant governing the
-- choice of a variant clause may appear in the list of fields of an XVE
-- type after the entry for the variant clause itself (this can happen
-- in the presence of a representation clause for the record type in the
-- source program). However, when this happens, the discriminant's
-- position may be determined by first applying the rules described in
-- this section, ignoring the variant clause. As a result, discriminants
-- can always be located independently of the variable-length fields
-- that depend on them.
-- The size of the ___XVE or ___XVU record or union is set to the
-- alignment (in bytes) of the original object so that the debugger
-- can calculate the size of the original type.
-- As an example of this encoding, consider the declarations:
-- type Q is array (1 .. V1) of Float; -- alignment 4
-- type R is array (1 .. V2) of Long_Float; -- alignment 8
-- type X is record
-- A : Character;
-- B : Float;
-- C : String (1 .. V3);
-- D : Float;
-- E : Q;
-- F : R;
-- G : Float;
-- end record;
-- The encoded type looks like:
-- type anonymousQ is access Q;
-- type anonymousR is access R;
-- type X___XVE is record
-- A : Character; -- position contains 0
-- B : Float; -- position contains 24
-- C___XVL : access String (1 .. V3); -- position contains 0
-- D___XVA4 : Float; -- position contains 0
-- E___XVL4 : anonymousQ; -- position contains 0
-- F___XVL8 : anonymousR; -- position contains 0
-- G : Float; -- position contains 0
-- end record;
-- Any bit sizes recorded for fields other than dynamic fields and
-- variants are honored as for ordinary records.
-- Notes:
-- 1) The B field could also have been encoded by using a position of
-- zero and an alignment of 4, but in such a case the coding by position
-- is preferred (since it takes up less space). We have used the
-- (illegal) notation access xxx as field types in the example above.
-- 2) The E field does not actually need the alignment indication but
-- this may not be detected in this case by the conversion routines.
-- 3) Our conventions do not cover all XVE-encoded records in which
-- some, but not all, fields have representation clauses. Such records
-- may, therefore, be displayed incorrectly by debuggers. This situation
-- is not common.
-----------------------
-- Base Record Types --
-----------------------
-- Under certain circumstances, debuggers need two descriptions of a
-- record type, one that gives the actual details of the base type's
-- structure (as described elsewhere in these comments) and one that may
-- be used to obtain information about the particular subtype and the
-- size of the objects being typed. In such cases the compiler will
-- substitute type whose name is typically compiler-generated and
-- irrelevant except as a key for obtaining the actual type.
-- Specifically, if this name is x, then we produce a record type named
-- x___XVS consisting of one field. The name of this field is that of
-- the actual type being encoded, which we'll call y. The type of this
-- single field can be either an arbitrary non-reference type, e.g. an
-- integer type, or a reference type; in the latter case, the referenced
-- type is also the actual type being encoded y. Both x and y may have
-- corresponding ___XVE types.
-- The size of the objects typed as x should be obtained from the
-- structure of x (and x___XVE, if applicable) as for ordinary types
-- unless there is a variable named x___XVZ, which, if present, will
-- hold the size (in bytes) of x. In this latter case, the size of the
-- x___XVS type will not be a constant but a reference to x___XVZ.
-- The type x will either be a subtype of y (see also Subtypes of
-- Variant Records, below) or will contain a single field of type y,
-- or no fields at all. The layout, types, and positions of these
-- fields will be accurate, if present. (Currently, however, the GDB
-- debugger makes no use of x except to determine its size).
-- Among other uses, XVS types are used to encode unconstrained types.
-- For example, given:
--
-- subtype Int is INTEGER range 0..10;
-- type T1 (N: Int := 0) is record
-- F1: String (1 .. N);
-- end record;
-- type AT1 is array (INTEGER range <>) of T1;
--
-- the element type for AT1 might have a type defined as if it had
-- been written:
--
-- type at1___PAD is record F : T1; end record;
-- for at1___PAD'Size use 16 * 8;
--
-- and there would also be:
--
-- type at1___PAD___XVS is record t1: reft1; end record;
-- type t1 is ...
-- type reft1 is <reference to t1>
--
-- Had the subtype Int been dynamic:
--
-- subtype Int is INTEGER range 0 .. M; -- M a variable
--
-- Then the compiler would also generate a declaration whose effect
-- would be
--
-- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term;
--
-- Not all unconstrained types are so encoded; the XVS convention may be
-- unnecessary for unconstrained types of fixed size. However, this
-- encoding is always necessary when a subcomponent type (array
-- element's type or record field's type) is an unconstrained record
-- type some of whose components depend on discriminant values.
-----------------
-- Array Types --
-----------------
-- Since there is no way for the debugger to obtain the index subtypes
-- for an array type, we produce a type that has the name of the array
-- type followed by "___XA" and is a record type whose field types are
-- the respective types for the bounds (and whose field names are the
-- names of these types).
-- To conserve space, we do not produce this type unless one of the
-- index types is either an enumeration type, has a variable lower or
-- upper bound or is a biased type.
-- Given the full encoding of these types (see above description for
-- the encoding of discrete types), this means that all necessary
-- information for addressing arrays is available. In some debugging
-- formats, some or all of the bounds information may be available
-- redundantly, particularly in the fixed-point case, but this
-- information can in any case be ignored by the debugger.
-------------------------------------------------
-- Subprograms for Handling Encoded Type Names --
-------------------------------------------------
procedure Get_Encoded_Name (E : Entity_Id);
-- If the entity is a typename, store the external name of the entity as in
-- Get_External_Name, followed by three underscores plus the type encoding
-- in Name_Buffer with the length in Name_Len, and an ASCII.NUL character
-- stored following the name. Otherwise set Name_Buffer and Name_Len to
-- hold the entity name. Note that a call to this procedure has no effect
-- if we are not generating code, since the necessary information for
-- computing the proper encoded name is not available in this case.
-- WARNING: There is a matching C declaration of this subprogram in fe.h
--------------
-- Renaming --
--------------
-- Debugging information is generated for exception, object, package, and
-- subprogram renaming (generic renamings are not significant, since
-- generic templates are not relevant at debugging time).
-- Consider a renaming declaration of the form
-- x : typ renames y;
-- There is one case in which no special debugging information is required,
-- namely the case of an object renaming where the back end allocates a
-- reference for the renamed variable, and the entity x is this reference.
-- The debugger can handle this case without any special processing or
-- encoding (it won't know it was a renaming, but that does not matter).
-- All other cases of renaming generate a dummy variable for an entity
-- whose name is of the form:
-- x___XR_... for an object renaming
-- x___XRE_... for an exception renaming
-- x___XRP_... for a package renaming
-- and where the "..." represents a suffix that describes the structure of
-- the object name given in the renaming (see details below).
-- The name is fully qualified in the usual manner, i.e. qualified in the
-- same manner as the entity x would be. In the case of a package renaming
-- where x is a child unit, the qualification includes the name of the
-- parent unit, to disambiguate child units with the same simple name and
-- (of necessity) different parents.
-- Note: subprogram renamings are not encoded at the present time
-- The suffix of the variable name describing the renamed object is defined
-- to use the following encoding:
-- For the simple entity case, where y is just an entity name, the suffix
-- is of the form:
-- y___XE
-- i.e. the suffix has a single field, the first part matching the
-- name y, followed by a "___" separator, ending with sequence XE.
-- The entity name portion is fully qualified in the usual manner.
-- This same naming scheme is followed for all forms of encoded
-- renamings that rename a simple entity.
-- For the object renaming case where y is a selected component or an
-- indexed component, the variable name is suffixed by additional fields
-- that give details of the components. The name starts as above with a
-- y___XE name indicating the outer level object entity. Then a series of
-- selections and indexing operations can be specified as follows:
-- Indexed component
-- A series of subscript values appear in sequence, the number
-- corresponds to the number of dimensions of the array. The
-- subscripts have one of the following two forms:
-- XSnnn
-- Here nnn is a constant value, encoded as a decimal integer
-- (pos value for enumeration type case). Negative values have
-- a trailing 'm' as usual.
-- XSe
-- Here e is the (unqualified) name of a constant entity in the
-- same scope as the renaming which contains the subscript value.
-- Slice
-- For the slice case, we have two entries. The first is for the
-- lower bound of the slice, and has the form:
-- XLnnn
-- XLe
-- Specifies the lower bound, using exactly the same encoding as
-- for an XS subscript as described above.
-- Then the upper bound appears in the usual XSnnn/XSe form
-- Selected component
-- For a selected component, we have a single entry
-- XRf
-- Here f is the field name for the selection
-- For an explicit dereference (.all), we have a single entry
-- XA
-- As an example, consider the declarations:
-- package p is
-- type q is record
-- m : string (2 .. 5);
-- end record;
--
-- type r is array (1 .. 10, 1 .. 20) of q;
--
-- g : r;
--
-- z : string renames g (1,5).m(2 ..3)
-- end p;
-- The generated variable entity would appear as
-- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type;
-- p__g___XE--------------------outer entity is g
-- XS1-----------------first subscript for g
-- XS5--------------second subscript for g
-- XRm-----------select field m
-- XL2--------lower bound of slice
-- XS3-----upper bound of slice
-- Note that the type of the variable is a special internal type named
-- _renaming_type. This type is an arbitrary type of zero size created
-- in package Standard (see cstand.adb) and is ignored by the debugger.
function Debug_Renaming_Declaration (N : Node_Id) return Node_Id;
-- The argument N is a renaming declaration. The result is a variable
-- declaration as described in the above paragraphs. If N is not a special
-- debug declaration, then Empty is returned. This function also takes care
-- of setting Materialize_Entity on the renamed entity where required.
-------------------------------------------
-- Packed Array Representation in Memory --
-------------------------------------------
-- Packed arrays are represented in tightly packed form, with no extra bits
-- between components. This is true even when the component size is not a
-- factor of the storage unit size, so that as a result it is possible for
-- components to cross storage unit boundaries.
-- The layout in storage is identical, regardless of whether the
-- implementation type is a modular type or an array-of-bytes type. See
-- Exp_Pakd for details of how these implementation types are used, but for
-- the purpose of the debugger, only the starting address of the object in
-- memory is significant.
-- The following example should show clearly how the packing works in
-- the little-endian and big-endian cases:
-- type B is range 0 .. 7;
-- for B'Size use 3;
-- type BA is array (0 .. 5) of B;
-- pragma Pack (BA);
-- BV : constant BA := (1,2,3,4,5,6);
-- Little endian case
-- BV'Address + 2 BV'Address + 1 BV'Address + 0
-- +-----------------+-----------------+-----------------+
-- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 |
-- +-----------------+-----------------+-----------------+
-- <---------> <-----> <---> <---> <-----> <---> <--->
-- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0)
--
-- Big endian case
--
-- BV'Address + 0 BV'Address + 1 BV'Address + 2
-- +-----------------+-----------------+-----------------+
-- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? |
-- +-----------------+-----------------+-----------------+
-- <---> <---> <-----> <---> <---> <-----> <--------->
-- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits
-- Note that if a modular type is used to represent the array, the
-- allocation in memory is not the same as a normal modular type. The
-- difference occurs when the allocated object is larger than the size of
-- the array. For a normal modular type, we extend the value on the left
-- with zeroes.
-- For example, in the normal modular case, if we have a 6-bit modular
-- type, declared as mod 2**6, and we allocate an 8-bit object for this
-- type, then we extend the value with two bits on the most significant
-- end, and in either the little-endian or big-endian case, the value 63
-- is represented as 00111111 in binary in memory.
-- For a modular type used to represent a packed array, the rule is
-- different. In this case, if we have to extend the value, then we do it
-- with undefined bits (which are not initialized and whose value is
-- irrelevant to any generated code). Furthermore these bits are on the
-- right (least significant bits) in the big-endian case, and on the left
-- (most significant bits) in the little-endian case.
-- For example, if we have a packed boolean array of 6 bits, all set to
-- True, stored in an 8-bit object, then the value in memory in binary is
-- ??111111 in the little-endian case, and 111111?? in the big-endian case.
-- This is done so that the representation of packed arrays does not
-- depend on whether we use a modular representation or array of bytes
-- as previously described. This ensures that we can pass such values by
-- reference in the case where a subprogram has to be able to handle values
-- stored in either form.
-- Note that when we extract the value of such a modular packed array, we
-- expect to retrieve only the relevant bits, so in this same example, when
-- we extract the value we get 111111 in both cases, and the code generated
-- by the front end assumes this although it does not assume that any high
-- order bits are defined.
-- There are opportunities for optimization based on the knowledge that the
-- unused bits are irrelevant for these type of packed arrays. For example
-- if we have two such 6-bit-in-8-bit values and we do an assignment:
-- a := b;
-- Then logically, we extract the 6 bits and store only 6 bits in the
-- result, but the back end is free to simply assign the entire 8-bits in
-- this case, since we don't actually care about the undefined bits.
-- However, in the equality case, it is important to ensure that the
-- undefined bits do not participate in an equality test.
-- If a modular packed array value is assigned to a register then logically
-- it could always be held right justified, to avoid any need to shift,
-- e.g. when doing comparisons. But probably this is a bad choice, as it
-- would mean that an assignment such as a := above would require shifts
-- when one value is in a register and the other value is in memory.
-------------------------------------------
-- Packed Array Name Encoding (OBSOLETE) --
-------------------------------------------
-- For every constrained packed array, two types are created, and both
-- appear in the debugging output:
-- The original declared array type is a perfectly normal array type, and
-- its index bounds indicate the original bounds of the array.
-- The corresponding packed array type, which may be a modular type, or
-- may be an array of bytes type (see Exp_Pakd for full details). This is
-- the type that is actually used in the generated code and for debugging
-- information for all objects of the packed type.
-- The name of the corresponding packed array type is:
-- ttt___XPnnn
-- where
-- ttt is the name of the original declared array
-- nnn is the component size in bits (1-31)
-- Note that if the packed array is not bit-packed, the name will simply
-- be tttP.
-- When the debugger sees that an object is of a type that is encoded in
-- this manner, it can use the original type to determine the bounds and
-- the component type, and the component size to determine the packing
-- details.
-- For an unconstrained packed array, the corresponding packed array type
-- is neither used in the generated code nor for debugging information,
-- only the original type is used. In order to convey the packing in the
-- debugging information, the compiler generates the associated fat- and
-- thin-pointer types (see the Pointers to Unconstrained Array section
-- below) using the name of the corresponding packed array type as the
-- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively.
-- When the debugger sees that an object is of a type that is encoded in
-- this manner, it can use the type of the fields to determine the bounds
-- and the component type, and the component size to determine the packing
-- details.
------------------------------------------------------
-- Subprograms for Handling Packed Array Type Names --
------------------------------------------------------
function Make_Packed_Array_Impl_Type_Name
(Typ : Entity_Id;
Csize : Uint) return Name_Id;
-- This function is used in Exp_Pakd to create the name that is encoded as
-- described above. The entity Typ provides the name ttt, and the value
-- Csize is the component size that provides the nnn value.
--------------------------------------
-- Pointers to Unconstrained Arrays --
--------------------------------------
-- There are two kinds of pointer to unconstrained arrays. The debugger can
-- tell which format is in use by the form of the type of the pointer.
-- Fat Pointers
-- Fat pointers are represented as a structure with two fields. This
-- structure has two distinguished field names:
-- P_ARRAY is a pointer to the array type. The name of this type is
-- the unconstrained type followed by "___XUA". The bounds of this
-- array will be obtained through dereferences of P_BOUNDS below.
-- P_BOUNDS is a pointer to a structure. The name of this type is
-- the unconstrained array name followed by "___XUB" and it has
-- fields of the form:
-- LBn (n a decimal integer) lower bound of n'th dimension
-- UBn (n a decimal integer) upper bound of n'th dimension
-- The bounds may be of any integral type. In the case of enumeration
-- types, Enum_Rep values are used.
-- For a given unconstrained array type, the compiler will generate a
-- fat pointer type whose name is the name of the array type, and use
-- it to represent the array type itself in the debugging information.
-- This name was historically followed by "___XUP" (OBSOLETE).
-- For each pointer to this unconstrained array type, the compiler will
-- generate a typedef that points to the above fat pointer type. As a
-- consequence, when it comes to fat pointer types:
-- 1. The type name is given by the typedef, if any
-- 2. If the debugger is asked to output the type, the appropriate
-- form is "access arr" if there is the typedef, otherwise it is
-- the array definition.
-- Thin Pointers
-- The value of a thin pointer is a pointer to the second field of a
-- structure with two fields. The first field of the structure is of
-- the type ___XUB described for fat pointer types above. The second
-- field of the structure contains the actual array.
-- Thin pointers are represented as a regular pointer to array in the
-- debugging information. The bounds of this array will be the contents
-- of the first field above obtained through (shifted) dereferences.
-- Thin Pointers (OBSOLETE)
-- The value of a thin pointer is a pointer to the second field of a
-- structure with two fields. The name of this structure's type is
-- "arr___XUT", where "arr" is the name of the unconstrained array
-- type. Even though it points into the middle of this structure,
-- the type in the debugging information is pointer to structure.
-- The first field of the structure is named BOUNDS and is of the type
-- ___XUB described for fat pointer types above.
-- The second field of the structure is named ARRAY, and contains the
-- actual array. Because this array has a dynamic size, determined by
-- the BOUNDS field that precedes it, all of the information about
-- arr___XUT is encoded in a parallel type named arr___XUT___XVE, with
-- fields BOUNDS and ARRAY___XVL. As for previously described ___XVE
-- types, ARRAY___XVL has a pointer-to-array type. However, the array
-- type in this case is named arr___XUA and only its element type is
-- meaningful, just as described for fat pointers.
-----------------------------
-- Variant Record Encoding --
-----------------------------
-- The variant part of a variant record is encoded as a single field in the
-- enclosing record, whose name is:
-- discrim___XVN
-- where discrim is the unqualified name of the variant. This field name is
-- built by gigi (not by code in this unit). For Unchecked_Union record,
-- this discriminant will not appear in the record (see Unchecked Unions,
-- below).
-- The type corresponding to this field has a name that is obtained by
-- concatenating the type name with the above string and is similar to a C
-- union, in which each member of the union corresponds to one variant.
-- However, unlike a C union, the size of the type may be variable even if
-- each of the components are fixed size, since it includes a computation
-- of which variant is present.
-- The name of the union member is encoded to indicate the choices, and
-- is a string given by the following grammar:
-- member_name ::= {choice} | others_choice
-- choice ::= simple_choice | range_choice
-- simple_choice ::= S number
-- range_choice ::= R number T number
-- number ::= {decimal_digit} [m]
-- others_choice ::= O (upper case letter O)
-- The m in a number indicates a negative value. As an example of this
-- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by
-- R1T4S7S10m
-- In the case of enumeration values, the values used are the actual
-- representation values in the case where an enumeration type has an
-- enumeration representation spec (i.e. they are values that correspond
-- to the use of the Enum_Rep attribute).
-- The type of the inner record is given by the name of the union type (as
-- above) concatenated with the above string.
-- As an example, consider:
-- type Var (Disc : Boolean := True) is record
-- M : Integer;
-- case Disc is
-- when True =>
-- R : Integer;
-- S : Integer;
-- when False =>
-- T : Integer;
-- end case;
-- end record;
-- V1 : Var;
-- In this case, the type var is represented as a struct with three fields.
-- The first two are "disc" and "m", representing the values of these
-- record components. The third field is a union of two types, with field
-- names S1 and O. S1 is a struct with fields "r" and "s", and O is a
-- struct with field "t".
----------------------
-- Unchecked Unions --
----------------------
-- The encoding for variant records changes somewhat under the influence
-- of a "pragma Unchecked_Union" clause:
-- 1. The discriminant will not be present in the record, although its
-- name is still used in the encodings.
-- 2. Variants containing a single component named "x" of type "T" may
-- be encoded, as in ordinary C unions, as a single field of the
-- enclosing union type named "x" of type "T", dispensing with the
-- enclosing struct. In this case, of course, the discriminant values
-- corresponding to the variant are unavailable.
-- For example, the type Var in the preceding section, if followed by
-- "pragma Unchecked_Union (Var);" may be encoded as a struct with two
-- fields. The first is "m". The second field is a union of two types,
-- with field names S1 and "t". As before, S1 is a struct with fields
-- "r" and "s". "t" is a field of type Integer.
------------------------------------------------
-- Subprograms for Handling Variant Encodings --
------------------------------------------------
procedure Get_Variant_Encoding (V : Node_Id);
-- This procedure is called by Gigi with V being the variant node. The
-- corresponding encoding string is returned in Name_Buffer with the length
-- of the string in Name_Len, and an ASCII.NUL character stored following
-- the name.
-- WARNING: There is a matching C declaration of this subprogram in fe.h
---------------------------------
-- Subtypes of Variant Records --
---------------------------------
-- A subtype of a variant record is represented by a type in which the
-- union field from the base type is replaced by one of the possible
-- values. For example, if we have:
-- type Var (Disc : Boolean := True) is record
-- M : Integer;
-- case Disc is
-- when True =>
-- R : Integer;
-- S : Integer;
-- when False =>
-- T : Integer;
-- end case;
-- end record;
-- V1 : Var;
-- V2 : Var (True);
-- V3 : Var (False);
-- Here V2, for example, is represented with a subtype whose name is
-- something like TvarS3b, which is a struct with three fields. The first
-- two fields are "disc" and "m" as for the base type, and the third field
-- is S1, which contains the fields "r" and "s".
-- The debugger should simply ignore structs with names of the form
-- corresponding to variants, and consider the fields inside as belonging
-- to the containing record.
-----------------------------------------------
-- Extra renamings for subprogram instances --
-----------------------------------------------
procedure Build_Subprogram_Instance_Renamings
(N : Node_Id;
Wrapper : Entity_Id);
-- The debugger has difficulties in recovering the value of actuals of an
-- elementary type, from within the body of a subprogram instantiation.
-- This is because such actuals generate an object declaration that is
-- placed within the wrapper package of the instance, and the entity in
-- these declarations is encoded in a complex way that GDB does not handle
-- well. These new renaming declarations appear within the body of the
-- subprogram, and are redundant from a visibility point of view, but they
-- should have no measurable performance impact, and require no special
-- decoding in the debugger.
-------------------------------------------
-- Character literals in Character Types --
-------------------------------------------
-- Character types are enumeration types at least one of whose enumeration
-- literals is a character literal. Enumeration literals are usually simply
-- represented using their identifier names. If the enumeration literal is
-- a character literal, the name is encoded as described in the following
-- paragraph.
-- The characters 'a'..'z' and '0'..'9' are represented as Qc, where 'c'
-- stands for the character itself. A name QUhh, where each 'h' is a
-- lower-case hexadecimal digit, stands for a character whose Unicode
-- encoding is hh, and QWhhhh likewise stands for a wide character whose
-- encoding is hhhh. The representation values are encoded as for ordinary
-- enumeration literals (and have no necessary relationship to the values
-- encoded in the names).
-- For example, given the type declaration
-- type x is (A, 'C', 'b');
-- the second enumeration literal would be named QU43 and the value
-- assigned to it would be 1, and the third enumeration literal would be
-- named Qb and the value assigned to it would be 2.
-----------------------------------------------
-- Secondary Dispatch tables of tagged types --
-----------------------------------------------
procedure Get_Secondary_DT_External_Name
(Typ : Entity_Id;
Ancestor_Typ : Entity_Id;
Suffix_Index : Int);
-- Set Name_Buffer and Name_Len to the external name of one secondary
-- dispatch table of Typ. If the interface has been inherited from some
-- ancestor then Ancestor_Typ is such node (in this case the secondary DT
-- is needed to handle overridden primitives); if there is no such ancestor
-- then Ancestor_Typ is equal to Typ.
--
-- Internal rule followed for the generation of the external name:
--
-- Case 1. If the secondary dispatch has not been inherited from some
-- ancestor of Typ then the external name is composed as
-- follows:
-- External_Name (Typ) + Suffix_Number + 'P'
--
-- Case 2. if the secondary dispatch table has been inherited from some
-- ancestor then the external name is composed as follows:
-- External_Name (Typ) + '_' + External_Name (Ancestor_Typ)
-- + Suffix_Number + 'P'
--
-- Note: We have to use the external names (instead of simply their names)
-- to protect the frontend against programs that give the same name to all
-- the interfaces and use the expanded name to reference them. The
-- Suffix_Number is used to differentiate all the secondary dispatch
-- tables of a given type.
--
-- Examples:
--
-- package Pkg1 is | package Pkg2 is | package Pkg3 is
-- type Typ is | type Typ is | type Typ is
-- interface; | interface; | interface;
-- end Pkg1; | end Pkg; | end Pkg3;
--
-- with Pkg1, Pkg2, Pkg3;
-- package Case_1 is
-- type Typ is new Pkg1.Typ and Pkg2.Typ and Pkg3.Typ with ...
-- end Case_1;
--
-- with Case_1;
-- package Case_2 is
-- type Typ is new Case_1.Typ with ...
-- end Case_2;
--
-- These are the external names generated for Case_1.Typ (note that
-- Pkg1.Typ is associated with the Primary Dispatch Table, because it
-- is the parent of this type, and hence no external name is
-- generated for it).
-- case_1__typ0P (associated with Pkg2.Typ)
-- case_1__typ1P (associated with Pkg3.Typ)
--
-- These are the external names generated for Case_2.Typ:
-- case_2__typ_case_1__typ0P
-- case_2__typ_case_1__typ1P
----------------------------
-- Effect of Optimization --
----------------------------
-- If the program is compiled with optimization on (e.g. -O1 switch
-- specified), then there may be variations in the output from the above
-- specification. In particular, objects may disappear from the output.
-- This includes not only constants and variables that the program declares
-- at the source level, but also the x___L and x___U constants created to
-- describe the lower and upper bounds of subtypes with dynamic bounds.
-- This means for example, that array bounds may disappear if optimization
-- is turned on. The debugger is expected to recognize that these constants
-- are missing and deal as best as it can with the limited information
-- available.
-----------------------------------------
-- GNAT Extensions to DWARF (OBSOLETE) --
-----------------------------------------
-- DW_AT_use_GNAT_descriptive_type, encoded with value 0x2301
-- This extension has never been implemented in the compiler.
-- DW_AT_GNAT_descriptive_type, encoded with value 0x2302
-- Any debugging information entry representing a type may have a
-- DW_AT_GNAT_descriptive_type attribute whose value is a reference,
-- pointing to a debugging information entry representing another type
-- associated to the type.
end Exp_Dbug;