2003-12-26 Guilhem Lavaux <guilhem@kaffe.org>

[official-gcc.git] / gcc / ada / exp_dbug.ads

------------------------------------------------------------------------------
--                                                                          --
--                         GNAT COMPILER COMPONENTS                         --
--                                                                          --
--                             E X P _ D B U G                              --
--                                                                          --
--                                 S p e c                                  --
--                                                                          --
--          Copyright (C) 1996-2003 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 2,  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 COPYING.  If not, write --
-- to  the Free Software Foundation,  59 Temple Place - Suite 330,  Boston, --
-- MA 02111-1307, USA.                                                      --
--                                                                          --
-- GNAT was originally developed  by the GNAT team at  New York University. --
-- Extensive contributions were provided by Ada Core Technologies Inc.      --
--                                                                          --
------------------------------------------------------------------------------

--  Expand routines for generation of special declarations used by the
--  debugger. In accordance with the Dwarf 2.2 specification, certain
--  type names are encoded to provide information to the debugger.

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 embdded 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.

   --  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 with respect to overloaded
      --  subprograms, since overloading can legitimately result in a
      --  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 one of the suffixes:

      --    $n    (dollar sign)
      --    __nn  (two underscores)

      --  where nn is a serial number (2 for the second overloaded function,
      --  2 for the third, etc.). We use $ if this symbol is allowed, and
      --  double underscore if it is not. In the remaining examples in this
      --  section, we use a $ sign, but the $ is replaced by __ throughout
      --  these examples if $ sign is not available. 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 the 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, the string "PT", 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 suprograms: 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:

      --    lockPT_getN
      --    lockPT_getP,

      --    lockPT_setN
      --    lockPT_setP

      --    lockPT_update1sE
      --    lockPT_udpate2sB

   ----------------------------------------------------
   -- Conversion between Entities and External Names --
   ----------------------------------------------------

   No_Dollar_In_Label : constant Boolean := True;
   --  True iff the target does not allow dollar signs ("$") in external names
   --  ??? We want to migrate all platforms to use the same convention.
   --  As a first step, we force this constant to always be True. This
   --  constant will eventually be deleted after we have verified that
   --  the migration does not cause any unforseen adverse impact.
   --  We chose "__" because it is supported on all platforms, which is
   --  not the case of "$".

   procedure Get_External_Name
     (Entity     : Entity_Id;
      Has_Suffix : Boolean);
   --  Set Name_Buffer and Name_Len to the external name of entity E.
   --  The external name is the Interface_Name, if specified, unless
   --  the entity has an address clause or a suffix.
   --
   --  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.

   procedure Get_External_Name_With_Suffix
     (Entity : Entity_Id;
      Suffix : String);
   --  Set Name_Buffer and Name_Len to the external name of entity E.
   --  If Suffix is the empty string the external name is as above,
   --  otherwise 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 this procedure is called in the ASIS mode, it does nothing. See the
   --  comments in the body for more details.

   --------------------------------------------
   -- 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).

   -------------------------
   -- Type Name Encodings --
   -------------------------

   --  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

      ------------------------
      -- Encapsulated Types --
      ------------------------

      --  In some cases, the compiler encapsulates a type by wrapping it in
      --  a structure. For example, this is used when a size or alignment
      --  specification requires a larger type. Consider:

      --    type y is mod 2 ** 64;
      --    for y'size use 256;

      --  In this case the compile generates a structure type y___PAD, which
      --  has a single field whose name is F. This single field is 64 bits
      --  long and contains the actual value.

      --  A similar encapsulation is done for some packed array types,
      --  in which case the structure type is y___LJM and the field name
      --  is OBJECT.

      --  When the debugger sees an object of a type whose name has a
      --  suffix not otherwise mentioned in this specification, the type
      --  is a record containing a single field, and the name of that field
      --  is all upper-case letters, it should look inside to get the value
      --  of the field, and neither the outer structure name, nor the
      --  field name should appear when the value is printed.

      -----------------------
      -- 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.

      ------------------------------
      -- VAX Floating-Point Types --
      ------------------------------

      --   Vax floating-point types are represented at run time as integer
      --   types, which are treated specially by the code generator. Their
      --   type names are encoded with the following suffix:

      --     typ___XFF
      --     typ___XFD
      --     typ___XFG

      --   representing the Vax F Float, D Float, and G Float types. The
      --   debugger must treat these specially. In particular, printing
      --   these values can be achieved using the debug procedures that
      --   are provided in package System.Vax_Float_Operations:

      --     procedure Debug_Output_D (Arg : D);
      --     procedure Debug_Output_F (Arg : F);
      --     procedure Debug_Output_G (Arg : G);

      --   These three procedures take a Vax floating-point argument, and
      --   output a corresponding decimal representation to standard output
      --   with no terminating line return.

      --------------------
      -- 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, and of course we do not need to encode the dummy XR
      --   types generated for renamings.

      --     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.

      -------------------
      -- Modular Types --
      -------------------

      --  A type declared

      --    type x is mod N;

      --  Is encoded as a subrange of an unsigned base type with lower bound
      --  0 and upper bound N. That is, there is no name encoding. We use
      --  the standard encodings provided by the debugging format. Thus
      --  we give these types a non-standard interpretation: the standard
      --  interpretation of our encoding would not, in general, imply that
      --  arithmetic on type x was to be performed modulo N (especially not
      --  when N is not a power of 2).

      ------------------
      -- 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 a
      --  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 is arbitrary).
      --  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 the size (in bits) of x.

      --  The type x will either be a subtype of y (see also Subtypes
      --  of Variant Records, below) or will contain 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 sometimes 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___C_PAD is record null; end record;
      --     for at1___C_PAD'Size use 16 * 8;
      --
      --  and there would also be
      --
      --     type at1___C_PAD___XVS is record t1: Integer; end record;
      --     type t1 is ...
      --
      --  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___C_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 whose field names
      --  are the names of the types for the bounds. The types of these
      --  fields is an integer type which is meaningless.

      --  To conserve space, we do not produce this type unless one of
      --  the index types is either an enumeration type, has a variable
      --  upper bound, has a lower bound different from the constant 1,
      --  is a biased type, or is wider than "sizetype".

      --  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.

      ----------------------------
      -- Note on 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 of S
      --  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.

   -------------------------------------------------
   -- 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.

   --------------
   -- 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 backend 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 type definition for
   --  an entity whose name is:

   --    x___XR    for an object renaming
   --    x___XRE   for an exception renaming
   --    x___XRP   for a package renaming

   --  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 type is an enumeration type with a single enumeration literal
   --  that is an identifier which describes the renamed variable.

   --    For the simple entity case, where y is an entity name,
   --    the enumeration is of the form:

   --       (y___XE)

   --          i.e. the enumeration type has a single field, whose name
   --          matches the name y, with the XE suffix. The entity for this
   --          enumeration literal is fully qualified in the usual manner.
   --          All subprogram, exception, and package renamings fall into
   --          this category, as well as simple object renamings.

   --    For the object renaming case where y is a selected component or an
   --    indexed component, the literal name is suffixed by additional fields
   --    that give details of the components. The name starts as above with
   --    a y___XE entity indicating the outer level variable. 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 deference (.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 type definition would appear as

   --       type p__z___XR is
   --         (p__g___XEXS1XS5XRmXL2XS3);
   --          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

   function Debug_Renaming_Declaration (N : Node_Id) return Node_Id;
   --  The argument N is a renaming declaration. The result is a type
   --  declaration as described in the above paragraphs. If not special
   --  debug declaration, than Empty is returned.

   ---------------------------
   -- Packed Array Encoding --
   ---------------------------

   --  For every 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)

   --  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 size to determine the packing details.

   --  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
   --     +-----------------+-----------------+-----------------+
   --     | 0 0 0 0 0 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 0 0 0 0 0 0 |
   --     +-----------------+-----------------+-----------------+
   --       <---> <---> <-----> <---> <---> <-----> <--------->
   --       BV(0) BV(1)  BV(2)  BV(3) BV(4)  BV(5)  unused bits

   ------------------------------------------------------
   -- Subprograms for Handling Packed Array Type Names --
   ------------------------------------------------------

   function Make_Packed_Array_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 pointers to 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 struct with two fields. This
   --      struct 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". This
   --        array will have bounds which are the discriminants, and
   --        hence are unparsable, but will give the number of
   --        subscripts and the component type.

   --        P_BOUNDS is a pointer to a struct, the name of  whose type is the
   --        unconstrained array name followed by "___XUB" and which 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 any integral type. In the case of an
   --        enumeration type, Enum_Rep values are used.

   --      The debugging information will sometimes reference an anonymous
   --      fat pointer type. Such types are given the name xxx___XUP, where
   --      xxx is the name of the designated type. If the debugger is asked
   --      to output such a type name, the appropriate form is "access xxx".

   --    Thin Pointers

   --      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 actually points into
   --      middle of this structure, the thin pointer's type in debugging
   --      information is pointer-to-arr___XUT.

   --      The first field of arr___XUT is named BOUNDS, and has a type
   --      named arr___XUB, with the structure described for such types
   --      in fat pointers, as described above.

   --      The second field of arr___XUT 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.

   --------------------------------------
   -- Tagged Types and Type Extensions --
   --------------------------------------

   --  A type C 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 (encoded as usual if it has a dynamic size),
   --  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 type void*.
   --  If the type is derived from another tagged type, its _tag field is
   --  found in its _parent field.

   -----------------------------
   -- 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). In the case of an
   --  Unchecked_Union record, this discriminant will not appear in the
   --  record, and the debugger must proceed accordingly (basically it
   --  can treat this case as it would a C union).

   --  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. In that case,
   --  it will be encoded as above and a type with the suffix "___XVN___XVU"
   --  will be present.

   --  The name of the union member is encoded to indicate the choices, and
   --  is a string given by the following grammar:

   --    union_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. Since that
   --  type may itself be variable-sized, it may also be encoded as above
   --  with a new type with a further suffix of "___XVU".

   --  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
   --  fields "t".

   ------------------------------------------------
   -- 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.

   ---------------------------------
   -- 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.

   -------------------------------------------
   -- 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. In
   --  the case where an enumeration literal is a character literal, the
   --  name aencoded as described in the following paragraph.

   --  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.

   ----------------------------
   -- 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.

end Exp_Dbug;