-- --
-- S p e c --
-- --
--- Copyright (C) 1996-2005, Free Software Foundation, Inc. --
+-- Copyright (C) 1996-2010, 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- --
+-- 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 COPYING. If not, write --
--- to the Free Software Foundation, 51 Franklin Street, Fifth Floor, --
--- Boston, MA 02110-1301, USA. --
+-- 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. --
-- debugger. In accordance with the Dwarf 2.2 specification, certain
-- type names are encoded to provide information to the debugger.
+with Namet; use Namet;
with Types; use Types;
with Uintp; use Uintp;
-- Encoding and Qualification of Names of Entities --
-----------------------------------------------------
- -- This section describes how the names of entities are encoded in
- -- the generated debugging information.
+ -- 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).
+ -- 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.
+ -- 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
+ -- 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 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:
+ -- 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
-- 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 the suffix:
+ -- 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)
-- 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:
+ -- 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
-- __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.
+ -- 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
-- 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:
+ -- 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
-- 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:
+ -- second operator "=" defined in package Extra.Messages would have
+ -- the name:
-- extra__messages__Oeq__2
----------------------------------
-- 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:
+ -- 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 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
+ -- 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
-- a real problem of name clashes.
-- To deal with this situation, we provide two additional encoding
- -- rules for names
+ -- rules for names:
-- First: all library subprogram names are preceded by the string
-- _ada_ (which causes no duplications, since normal Ada names can
-- 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.
+ -- 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 --
-- end TaskObj;
-- end P;
--
- -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1,
+ -- 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.
-- 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
+ -- 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
No_Dollar_In_Label : constant Boolean := True;
-- True iff the target does not allow dollar signs ("$") in external names
- -- ??? We want to migrate all platforms to use the same convention.
- -- As a first step, we force this constant to always be True. This
- -- constant will eventually be deleted after we have verified that
- -- the migration does not cause any unforseen adverse impact.
- -- We chose "__" because it is supported on all platforms, which is
- -- not the case of "$".
+ -- ??? 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 unforeseen 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.
+ -- 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:
+ -- 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 "__",
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:
+ -- 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 "__",
-- 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.
+ -- 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).
+ -- 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
+ -- 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
+ -- 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;
-- a size of 256 for a signed integer value, then a typical choice is
-- to wrap a 64-bit integer in a 256 bit PAD structure.
- -- A similar encapsulation is done for some packed array types,
- -- in which case the structure type is y___JM and the field name
- -- is OBJECT. This is used in the case of a packed array stored
- -- in modular representation (see section on representation of
- -- packed array objects). In this case the JM wrapping is used to
- -- achieve correct positioning of the packed array value (left or
- -- 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.
+ -- A similar encapsulation is done for some packed array types, in which
+ -- case the structure type is y___JM and the field name is OBJECT.
+ -- This is used in the case of a packed array stored using modular
+ -- representation (see section on representation of packed array
+ -- objects). In this case the JM wrapping is used to achieve correct
+ -- positioning of the packed array value (left or 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.
-- When the debugger sees a record named REP being a field inside
- -- another record, 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.
+ -- another record, 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.
-----------------------
-- 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.
+ -- delta and small values. The actual type itself is a normal integer
+ -- type.
-- typ___XF_nn_dd
-- typ___XF_nn_dd_nn_dd
-- 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:
+ -- 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);
-- 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.
+ -- 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.
+ -- 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
-- 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 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
-- 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
+ -- zero 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:
+ -- 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:
+ -- 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).
+ -- 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 --
-- 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.
+ -- 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
-- 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.
+ -- 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.
+ -- 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).
+ -- 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
-- 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.
+ -- 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:
+ -- 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
-- 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.
+ -- 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
-- 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.
+ -- 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.
+ -- 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.
+ -- 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
+ -- 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 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
+
+ -- 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 sometimes used to encode
- -- unconstrained types. For example, given
+ -- 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
-- 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;
+ -- type at1___PAD is record F : T1; end record;
+ -- for at1___PAD'Size use 16 * 8;
--
- -- and there would also be
+ -- and there would also be:
--
- -- type at1___C_PAD___XVS is record t1: Integer; end record;
+ -- 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:
--
-- Then the compiler would also generate a declaration whose effect
-- would be
--
- -- at1___C_PAD___XVZ: constant Integer := 32 + M * 8 + padding term;
+ -- 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.
+ -- 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.
+ -- 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
- -- upper bound, has a lower bound different from the constant 1,
- -- is a biased type, or is wider than "sizetype".
+ -- 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.
+ -- 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:
+ -- 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:
+ -- 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:
+ -- case letter. For example the name for the above declaration might be:
-- TintegerS4b
-- Renaming --
--------------
- -- Debugging information is generated for exception, object, package,
- -- and subprogram renaming (generic renamings are not significant, since
+ -- 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;
+ -- 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
+ -- 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 type definition for
- -- an entity whose name is:
+ -- 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
- -- 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.
+ -- 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.
+ -- 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 an entity name,
- -- the enumeration is of the form:
+ -- For the simple entity case, where y is just an entity name, the suffix
+ -- is of the form:
- -- (y___XE)
+ -- 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.
+ -- 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 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, 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
-- 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.
+ -- 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.
+ -- 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
+ -- 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.
+ -- 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
-- Here f is the field name for the selection
- -- For an explicit deference (.all), we have a single entry
+ -- For an explicit dereference (.all), we have a single entry
-- XA
-- z : string renames g (1,5).m(2 ..3)
-- end p;
- -- The generated type definition would appear as
+ -- The generated variable entity 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
+ -- 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 type
- -- declaration as described in the above paragraphs. If not special
- -- debug declaration, than Empty is returned.
+ -- 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.
---------------------------
-- Packed Array Encoding --
---------------------------
- -- For every packed array, two types are created, and both appear in
- -- the debugging output.
+ -- 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 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.
+ -- 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.
+ -- 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.
-------------------------------------------
-- 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.
+ -- 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.
+ -- 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:
-- 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.
+ -- 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.
+ -- 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.
+ -- 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.
+ -- 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.
+ -- 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:
+ -- 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.
+ -- 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 := b
- -- above would require shifts when one value is in a register and the
- -- other value is in memory.
+ -- 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.
------------------------------------------------------
-- Subprograms for Handling Packed Array Type Names --
(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.
+ -- 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.
+ -- 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_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
-- 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 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".
+ -- For a given unconstrained array type, the compiler will generate one
+ -- fat-pointer type whose name is "arr___XUP", where "arr" is the name
+ -- of the array type, and use it to represent the array type itself in
+ -- the debugging information.
+
+ -- For each pointer to this unconstrained array type, the compiler will
+ -- generate a typedef that points to the above "arr___XUP" fat-pointer
+ -- type. As a consequence, when it comes to fat-pointer types:
+
+ -- 1. The type name is given by the typedef
+
+ -- 2. If the debugger is asked to output the type, the appropriate
+ -- form is "access arr", except if the type name is "arr___XUP"
+ -- for which 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 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.
+ -- 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.
+ -- 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.
+ -- 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:
+ -- 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.
+ -- 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. 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
+ -- member_name ::= {choice} | others_choice
-- choice ::= simple_choice | range_choice
-- simple_choice ::= S number
-- range_choice ::= R number T number
-- 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).
+ -- 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".
+ -- 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:
-- 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".
+ -- 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. As for normal
+ -- variants, the field name "x" may be suffixed with ___XVL if it
+ -- has dynamic size.
+
+ -- 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.
+ -- 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 --
-- 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".
+ -- 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.
+ -- 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.
+ -- 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.
- -- 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).
+ -- 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.
+ -- the second enumeration literal would be named QU43 and the value
+ -- assigned to it would be 1.
-----------------------------------------------
-- Secondary Dispatch tables of tagged types --
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 overriden primitives); if there is no such
- -- ancestor then Ancestor_Typ is equal to Typ.
+ -- 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:
--
-- 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.
+ -- 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:
--
--
-- 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 the parent of this type, and hence no external name is
+ -- 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)
----------------------------
-- 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.
+ -- 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 DWARF2/3 --
+ ---------------------------------
+
+ -- If the compiler switch "-gdwarf+" is specified, GNAT Vendor extensions
+ -- to DWARF2/3 are generated, with the following variations from the above
+ -- specification.
+
+ -- Change in the contents of the DW_AT_name attribute
+
+ -- The operators are represented in their natural form. (for example,
+ -- the addition operator is written as "+" instead of "Oadd"). The
+ -- component separator is "." instead of "__"
+
+ -- Introduction of DW_AT_GNAT_encoding, encoded with value 0x2301
+
+ -- Any debugging information entry representing a program entity, named
+ -- or implicit, may have a DW_AT_GNAT_encoding attribute. The value of
+ -- this attribute is a string representing the suffix internally added
+ -- by GNAT for various purposes, mainly for representing debug
+ -- information compatible with other formats. In particular this is
+ -- useful for IDEs which need to filter out information internal to
+ -- GNAT from their graphical interfaces.
+
+ -- If a debugging information entry has multiple encodings, all of them
+ -- will be listed in DW_AT_GNAT_encoding using the list separator ':'.
+
+ -- Introduction of 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.
+
+ -- Modification of the contents of the DW_AT_producer string
+
+ -- When emitting full GNAT Vendor extensions to DWARF2/3, "-gdwarf+"
+ -- is appended to the DW_AT_producer string.
+ --
+ -- When emitting only DW_AT_GNAT_descriptive_type, "-gdwarf+-" is
+ -- appended to the DW_AT_producer string.
end Exp_Dbug;