------------------------------------------------------------------------------ -- -- -- GNAT COMPILER COMPONENTS -- -- -- -- E X P _ D B U G -- -- -- -- S p e c -- -- -- -- Copyright (C) 1996-2019, Free Software Foundation, Inc. -- -- -- -- GNAT is free software; you can redistribute it and/or modify it under -- -- terms of the GNU General Public License as published by the Free Soft- -- -- ware Foundation; either version 3, or (at your option) any later ver- -- -- sion. GNAT is distributed in the hope that it will be useful, but WITH- -- -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -- -- or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License -- -- for more details. You should have received a copy of the GNU General -- -- Public License distributed with GNAT; see file COPYING3. If not, go to -- -- http://www.gnu.org/licenses for a complete copy of the license. -- -- -- -- GNAT was originally developed by the GNAT team at New York University. -- -- Extensive contributions were provided by Ada Core Technologies Inc. -- -- -- ------------------------------------------------------------------------------ -- Expand routines for 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 Namet; use Namet; with Types; use Types; with Uintp; use Uintp; package Exp_Dbug is ----------------------------------------------------- -- Encoding and Qualification of Names of Entities -- ----------------------------------------------------- -- This section describes how the names of entities are encoded in the -- generated debugging information. -- An entity in Ada has a name of the form X.Y.Z ... E where X,Y,Z are the -- enclosing scopes (not including Standard at the start). -- The encoding of the name follows this basic qualified naming scheme, -- where the encoding of individual entity names is as described in Namet -- (i.e. in particular names present in the original source are folded to -- all lower case, with upper half and wide characters encoded as described -- in Namet). Upper case letters are used only for entities generated by -- the compiler. -- There are two cases, global entities, and local entities. In more formal -- terms, local entities are those which have a dynamic enclosing scope, -- and global entities are at the library level, except that we always -- consider procedures to be global entities, even if they are nested -- (that's because at the debugger level a procedure name refers to the -- code, and the code is indeed a global entity, including the case of -- nested procedures.) In addition, we also consider all types to be global -- entities, even if they are defined within a procedure. -- The reason for treating all type names as global entities is that a -- number of our type encodings work by having related type names, and we -- need the full qualification to keep this unique. -- For global entities, the encoded name includes all components of the -- fully expanded name (but omitting Standard at the start). For example, -- if a library-level child package P.Q has an embedded package R, and -- there is an entity in this embedded package whose name is S, the encoded -- name will include the components p.q.r.s. -- For local entities, the encoded name only includes the components up to -- the enclosing dynamic scope (other than a block). At run time, such a -- dynamic scope is a subprogram, and the debugging formats know about -- local variables of procedures, so it is not necessary to have full -- qualification for such entities. In particular this means that direct -- local variables of a procedure are not qualified. -- For Ghost entities, the encoding adds a prefix "___ghost_" to aid the -- detection of leaks of Ignored Ghost entities in the "living" space. -- Ignored Ghost entities and any code associated with them should be -- removed by the compiler in a post-processing pass. As a result, -- object files should not contain any occurrences of this prefix. -- As an example of the local name convention, consider a procedure V.W -- with a local variable X, and a nested block Y containing an entity Z. -- The fully qualified names of the entities X and Z are: -- V.W.X -- V.W.Y.Z -- but since V.W is a subprogram, the encoded names will end up -- encoding only -- x -- y.z -- The separating dots are translated into double underscores ----------------------------- -- Handling of Overloading -- ----------------------------- -- The above scheme is incomplete for overloaded subprograms, since -- overloading can legitimately result in case of two entities with -- exactly the same fully qualified names. To distinguish between -- entries in a set of overloaded subprograms, the encoded names are -- serialized by adding the suffix: -- __nn (two underscores) -- where nn is a serial number (2 for the second overloaded function, -- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no -- suffix implies the first instance). -- These names are prefixed by the normal full qualification. So for -- example, the third instance of the subprogram qrs in package yz -- would have the name: -- yz__qrs__3 -- A more subtle case arises with entities declared within overloaded -- subprograms. If we have two overloaded subprograms, and both declare -- an entity xyz, then the fully expanded name of the two xyz's is the -- same. To distinguish these, we add the same __n suffix at the end of -- the inner entity names. -- In more complex cases, we can have multiple levels of overloading, -- and we must make sure to distinguish which final declarative region -- we are talking about. For this purpose, we use a more complex suffix -- which has the form: -- __nn_nn_nn ... -- where the nn values are the homonym numbers as needed for any of the -- qualifying entities, separated by a single underscore. If all the nn -- values are 1, the suffix is omitted, Otherwise the suffix is present -- (including any values of 1). The following example shows how this -- suffixing works. -- package body Yz is -- procedure Qrs is -- Name is yz__qrs -- procedure Tuv is ... end; -- Name is yz__qrs__tuv -- begin ... end Qrs; -- procedure Qrs (X: Int) is -- Name is yz__qrs__2 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1 -- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2 -- begin ... end Tuv; -- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3 -- type m is new float; -- Name is yz__qrs__tuv__m__2_3 -- begin ... end Tuv; -- begin ... end Qrs; -- end Yz; -------------------- -- Operator Names -- -------------------- -- The above rules applied to operator names would result in names with -- quotation marks, which are not typically allowed by assemblers and -- linkers, and even if allowed would be odd and hard to deal with. To -- avoid this problem, operator names are encoded as follows: -- Oabs abs -- Oand and -- Omod mod -- Onot not -- Oor or -- Orem rem -- Oxor xor -- Oeq = -- One /= -- Olt < -- Ole <= -- Ogt > -- Oge >= -- Oadd + -- Osubtract - -- Oconcat & -- Omultiply * -- Odivide / -- Oexpon ** -- These names are prefixed by the normal full qualification, and -- suffixed by the overloading identification. So for example, the -- second operator "=" defined in package Extra.Messages would have -- the name: -- extra__messages__Oeq__2 ---------------------------------- -- Resolving Other Name Clashes -- ---------------------------------- -- It might be thought that the above scheme is complete, but in Ada 95, -- full qualification is insufficient to uniquely identify an entity in -- the program, even if it is not an overloaded subprogram. There are -- two possible confusions: -- a.b -- interpretation 1: entity b in body of package a -- interpretation 2: child procedure b of package a -- a.b.c -- interpretation 1: entity c in child package a.b -- interpretation 2: entity c in nested package b in body of a -- It is perfectly legal in both cases for both interpretations to be -- valid within a single program. This is a bit of a surprise since -- certainly in Ada 83, full qualification was sufficient, but not in -- Ada 95. The result is that the above scheme can result in duplicate -- names. This would not be so bad if the effect were just restricted -- to debugging information, but in fact in both the above cases, it -- is possible for both symbols to be external names, and so we have -- a real problem of name clashes. -- To deal with this situation, we provide two additional encoding -- rules for names: -- First: all library subprogram names are preceded by the string -- _ada_ (which causes no duplications, since normal Ada names can -- never start with an underscore. This not only solves the first -- case of duplication, but also solves another pragmatic problem -- which is that otherwise Ada procedures can generate names that -- clash with existing system function names. Most notably, we can -- have clashes in the case of procedure Main with the C main that -- in some systems is always present. -- Second, for the case where nested packages declared in package -- bodies can cause trouble, we add a suffix which shows which -- entities in the list are body-nested packages, i.e. packages -- whose spec is within a package body. The rules are as follows, -- given a list of names in a qualified name name1.name2.... -- If none are body-nested package entities, then there is no suffix -- If at least one is a body-nested package entity, then the suffix -- is X followed by a string of b's and n's (b = body-nested package -- entity, n = not a body-nested package). -- There is one element in this string for each entity in the encoded -- expanded name except the first (the rules are such that the first -- entity of the encoded expanded name can never be a body-nested' -- package. Trailing n's are omitted, as is the last b (there must -- be at least one b, or we would not be generating a suffix at all). -- For example, suppose we have -- package x is -- pragma Elaborate_Body; -- m1 : integer; -- #1 -- end x; -- package body x is -- package y is m2 : integer; end y; -- #2 -- package body y is -- package z is r : integer; end z; -- #3 -- end; -- m3 : integer; -- #4 -- end x; -- package x.y is -- pragma Elaborate_Body; -- m2 : integer; -- #5 -- end x.y; -- package body x.y is -- m3 : integer; -- #6 -- procedure j is -- #7 -- package k is -- z : integer; -- #8 -- end k; -- begin -- null; -- end j; -- end x.y; -- procedure x.m3 is begin null; end; -- #9 -- Then the encodings would be: -- #1. x__m1 (no BNPE's in sight) -- #2. x__y__m2X (y is a BNPE) -- #3. x__y__z__rXb (y is a BNPE, so is z) -- #4. x__m3 (no BNPE's in sight) -- #5. x__y__m2 (no BNPE's in sight) -- #6. x__y__m3 (no BNPE's in signt) -- #7. x__y__j (no BNPE's in sight) -- #8. k__z (no BNPE's, only up to procedure) -- #9 _ada_x__m3 (library-level subprogram) -- Note that we have instances here of both kind of potential name -- clashes, and the above examples show how the encodings avoid the -- clash as follows: -- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library -- level subprogram, so it is preceded by the string _ada_ which acts -- to distinguish it from the package body entity. -- Lines #2 and #5 both refer to the entity x.y.m2, but the first -- instance is inside the body-nested package y, so there is an X -- suffix to distinguish it from the child library entity. -- Note that enumeration literals never need Xb type suffixes, since -- they are never referenced using global external names. --------------------- -- Interface Names -- --------------------- -- Note: if an interface name is present, then the external name is -- taken from the specified interface name. Given current limitations of -- the gcc backend, this means that the debugging name is also set to -- the interface name, but conceptually, it would be possible (and -- indeed desirable) to have the debugging information still use the Ada -- name as qualified above, so we still fully qualify the name in the -- front end. ------------------------------------- -- Encodings Related to Task Types -- ------------------------------------- -- Each task object defined by a single task declaration is associated -- with a prefix that is used to qualify procedures defined in that -- task. Given -- -- package body P is -- task body TaskObj is -- procedure F1 is ... end; -- begin -- B; -- end TaskObj; -- end P; -- -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1. -- The body, B, is contained in a subprogram whose name is -- p__taskobjTKB. ------------------------------------------ -- Encodings Related to Protected Types -- ------------------------------------------ -- Each protected type has an associated record type, that describes -- the actual layout of the private data. In addition to the private -- components of the type, the Corresponding_Record_Type includes one -- component of type Protection, which is the actual lock structure. -- The run-time size of the protected type is the size of the corres- -- ponding record. -- For a protected type prot, the Corresponding_Record_Type is encoded -- as protV. -- The operations of a protected type are encoded as follows: each -- operation results in two subprograms, a locking one that is called -- from outside of the object, and a non-locking one that is used for -- calls from other operations on the same object. The locking operation -- simply acquires the lock, and then calls the non-locking version. -- The names of all of these have a prefix constructed from the name of -- the type, and a suffix which is P or N, depending on whether this is -- the protected/non-locking version of the operation. -- Operations generated for protected entries follow the same encoding. -- Each entry results in two subprograms: a procedure that holds the -- entry body, and a function that holds the evaluation of the barrier. -- The names of these subprograms include the prefix '_E' or '_B' res- -- pectively. The names also include a numeric suffix to render them -- unique in the presence of overloaded entries. -- Given the declaration: -- protected type Lock is -- function Get return Integer; -- procedure Set (X: Integer); -- entry Update (Val : Integer); -- private -- Value : Integer := 0; -- end Lock; -- the following operations are created: -- lock_getN -- lock_getP, -- lock_setN -- lock_setP -- lock_update_E1s -- lock_udpate_B2s -- If the protected type implements at least one interface, the -- following additional operations are created: -- lock_get -- lock_set -- These operations are used to ensure overriding of interface level -- subprograms and proper dispatching on interface class-wide objects. -- The bodies of these operations contain calls to their respective -- protected versions: -- function lock_get return Integer is -- begin -- return lock_getP; -- end lock_get; -- procedure lock_set (X : Integer) is -- begin -- lock_setP (X); -- end lock_set; ---------------------------------------------------- -- Conversion between Entities and External Names -- ---------------------------------------------------- procedure Get_External_Name (Entity : Entity_Id; Has_Suffix : Boolean := False; Suffix : String := ""); -- Set Name_Buffer and Name_Len to the external name of the entity. The -- external name is the Interface_Name, if specified, unless the entity -- has an address clause or Has_Suffix is true. -- -- If the Interface is not present, or not used, the external name is the -- concatenation of: -- -- - the string "_ada_", if the entity is a library subprogram, -- - the names of any enclosing scopes, each followed by "__", -- or "X_" if the next entity is a subunit) -- - the name of the entity -- - the string "$" (or "__" if target does not allow "$"), followed -- by homonym suffix, if the entity is an overloaded subprogram -- or is defined within an overloaded subprogram. -- - the string "___" followed by Suffix if Has_Suffix is true. -- -- Note that a call to this procedure has no effect if we are not -- generating code, since the necessary information for computing the -- proper external name is not available in this case. -- WARNING: There is a matching C declaration of this subprogram in fe.h ------------------------------------- -- Encoding for translation into C -- ------------------------------------- -- In Modify_Tree_For_C mode we must add encodings to dismabiguate cases -- where Ada block structure cannot be directly translated. These cases -- are as follows: -- a) A loop variable may hide a homonym in an enclosing block -- b) A block-local variable may hide a homonym in an enclosing block -- In C these constructs are not scopes and we must distinguish the names -- explicitly. In the first case we create a qualified name with the suffix -- 'L', in the second case with a suffix 'B'. -------------------------------------------- -- Subprograms for Handling Qualification -- -------------------------------------------- procedure Qualify_Entity_Names (N : Node_Id); -- Given a node N, that represents a block, subprogram body, or package -- body or spec, or protected or task type, sets a fully qualified name -- for the defining entity of given construct, and also sets fully -- qualified names for all enclosed entities of the construct (using -- First_Entity/Next_Entity). Note that the actual modifications of the -- names is postponed till a subsequent call to Qualify_All_Entity_Names. -- Note: this routine does not deal with prepending _ada_ to library -- subprogram names. The reason for this is that we only prepend _ada_ -- to the library entity itself, and not to names built from this name. procedure Qualify_All_Entity_Names; -- When Qualify_Entity_Names is called, no actual name changes are made, -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call -- is made to this procedure. The reason for this deferral is that when -- names are changed semantic processing may be affected. By deferring -- the changes till just before gigi is called, we avoid any concerns -- about such effects. Gigi itself does not use the names except for -- output of names for debugging purposes (which is why we are doing -- the name changes in the first place). -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are -- useful to remove qualification from a name qualified by the call to -- Qualify_All_Entity_Names. -------------------------------- -- Handling of Numeric Values -- -------------------------------- -- All numeric values here are encoded as strings of decimal digits. Only -- integer values need to be encoded. A negative value is encoded as the -- corresponding positive value followed by a lower case m for minus to -- indicate that the value is negative (e.g. 2m for -2). ------------------------- -- 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. This kind of padding is used -- when the logical value to be stored is shorter than the object in -- which it is allocated. For example if a size clause is used to set -- 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 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. ----------------------- -- Fixed-Point Types -- ----------------------- -- Fixed-point types are encoded using a suffix that indicates the -- delta and small values. The actual type itself is a normal integer -- type. -- typ___XF_nn_dd -- typ___XF_nn_dd_nn_dd -- The first form is used when small = delta. The value of delta (and -- small) is given by the rational nn/dd, where nn and dd are decimal -- integers. -- -- The second form is used if the small value is different from the -- delta. In this case, the first nn/dd rational value is for delta, -- and the second value is for small. -------------------- -- Discrete Types -- -------------------- -- Discrete types are coded with a suffix indicating the range in the -- case where one or both of the bounds are discriminants or variable. -- Note: at the current time, we also encode compile time known bounds -- if they do not match the natural machine type bounds, but this may -- be removed in the future, since it is redundant for most debugging -- formats. However, we do not ever need XD encoding for enumeration -- base types, since here it is always clear what the bounds are from -- the total number of enumeration literals. -- typ___XD -- typ___XDL_lowerbound -- typ___XDU_upperbound -- typ___XDLU_lowerbound__upperbound -- If a discrete type is a natural machine type (i.e. its bounds -- correspond in a natural manner to its size), then it is left -- unencoded. The above encoding forms are used when there is a -- constrained range that does not correspond to the size or that -- has discriminant references or other compile time known bounds. -- The first form is used if both bounds are dynamic, in which case two -- constant objects are present whose names are typ___L and typ___U in -- the same scope as typ, and the values of these constants indicate -- the bounds. As far as the debugger is concerned, these are simply -- variables that can be accessed like any other variables. In the -- enumeration case, these values correspond to the Enum_Rep values for -- the lower and upper bounds. -- The second form is used if the upper bound is dynamic, but the lower -- bound is either constant or depends on a discriminant of the record -- with which the type is associated. The upper bound is stored in a -- constant object of name typ___U as previously described, but the -- lower bound is encoded directly into the name as either a decimal -- integer, or as the discriminant name. -- The third form is similarly used if the lower bound is dynamic, but -- the upper bound is compile time known or a discriminant reference, -- in which case the lower bound is stored in a constant object of name -- typ___L, and the upper bound is encoded directly into the name as -- either a decimal integer, or as the discriminant name. -- The fourth form is used if both bounds are discriminant references -- or compile time known values, with the encoding first for the lower -- bound, then for the upper bound, as previously described. ------------------- -- Modular Types -- ------------------- -- A type declared -- type x is mod N; -- Is encoded as a subrange of an unsigned base type with lower bound -- zero and upper bound N. 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 type whose name is typically compiler-generated and -- irrelevant except as a key for obtaining the actual type. -- Specifically, if this name is x, then we produce a record type named -- x___XVS consisting of one field. The name of this field is that of -- the actual type being encoded, which we'll call y. The type of this -- single field can be either an arbitrary non-reference type, e.g. an -- integer type, or a reference type; in the latter case, the referenced -- type is also the actual type being encoded y. Both x and y may have -- corresponding ___XVE types. -- The size of the objects typed as x should be obtained from the -- structure of x (and x___XVE, if applicable) as for ordinary types -- unless there is a variable named x___XVZ, which, if present, will -- hold the size (in bytes) of x. In this latter case, the size of the -- x___XVS type will not be a constant but a reference to x___XVZ. -- The type x will either be a subtype of y (see also Subtypes of -- Variant Records, below) or will contain a single field of type y, -- or no fields at all. The layout, types, and positions of these -- fields will be accurate, if present. (Currently, however, the GDB -- debugger makes no use of x except to determine its size). -- Among other uses, XVS types are used to encode unconstrained types. -- For example, given: -- -- subtype Int is INTEGER range 0..10; -- type T1 (N: Int := 0) is record -- F1: String (1 .. N); -- end record; -- type AT1 is array (INTEGER range <>) of T1; -- -- the element type for AT1 might have a type defined as if it had -- been written: -- -- type at1___PAD is record F : T1; end record; -- for at1___PAD'Size use 16 * 8; -- -- and there would also be: -- -- type at1___PAD___XVS is record t1: reft1; end record; -- type t1 is ... -- type reft1 is -- -- Had the subtype Int been dynamic: -- -- subtype Int is INTEGER range 0 .. M; -- M a variable -- -- Then the compiler would also generate a declaration whose effect -- would be -- -- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term; -- -- Not all unconstrained types are so encoded; the XVS convention may be -- unnecessary for unconstrained types of fixed size. However, this -- encoding is always necessary when a subcomponent type (array -- element's type or record field's type) is an unconstrained record -- type some of whose components depend on discriminant values. ----------------- -- Array Types -- ----------------- -- Since there is no way for the debugger to obtain the index subtypes -- for an array type, we produce a type that has the name of the array -- type followed by "___XA" and is a record type whose field types are -- the respective types for the bounds (and whose field names are the -- names of these types). -- To conserve space, we do not produce this type unless one of the -- index types is either an enumeration type, has a variable lower or -- upper bound or is a biased type. -- Given the full encoding of these types (see above description for -- the encoding of discrete types), this means that all necessary -- information for addressing arrays is available. In some debugging -- formats, some or all of the bounds information may be available -- redundantly, particularly in the fixed-point case, but this -- information can in any case be ignored by the debugger. ---------------------------- -- 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 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. Note that a call to this procedure has no effect -- if we are not generating code, since the necessary information for -- computing the proper encoded name is not available in this case. -- WARNING: There is a matching C declaration of this subprogram in fe.h -------------- -- Renaming -- -------------- -- Debugging information is generated for exception, object, package, and -- subprogram renaming (generic renamings are not significant, since -- generic templates are not relevant at debugging time). -- Consider a renaming declaration of the form -- x : typ renames y; -- There is one case in which no special debugging information is required, -- namely the case of an object renaming where the back end allocates a -- reference for the renamed variable, and the entity x is this reference. -- The debugger can handle this case without any special processing or -- encoding (it won't know it was a renaming, but that does not matter). -- All other cases of renaming generate a dummy variable for an entity -- whose name is of the form: -- x___XR_... for an object renaming -- x___XRE_... for an exception renaming -- x___XRP_... for a package renaming -- and where the "..." represents a suffix that describes the structure of -- the object name given in the renaming (see details below). -- The name is fully qualified in the usual manner, i.e. qualified in the -- same manner as the entity x would be. In the case of a package renaming -- where x is a child unit, the qualification includes the name of the -- parent unit, to disambiguate child units with the same simple name and -- (of necessity) different parents. -- Note: subprogram renamings are not encoded at the present time -- The suffix of the variable name describing the renamed object is defined -- to use the following encoding: -- For the simple entity case, where y is just an entity name, the suffix -- is of the form: -- y___XE -- i.e. the suffix has a single field, the first part matching the -- name y, followed by a "___" separator, ending with sequence XE. -- The entity name portion is fully qualified in the usual manner. -- This same naming scheme is followed for all forms of encoded -- renamings that rename a simple entity. -- For the object renaming case where y is a selected component or an -- indexed component, the variable name is suffixed by additional fields -- that give details of the components. The name starts as above with a -- y___XE name indicating the outer level object entity. Then a series of -- selections and indexing operations can be specified as follows: -- Indexed component -- A series of subscript values appear in sequence, the number -- corresponds to the number of dimensions of the array. The -- subscripts have one of the following two forms: -- XSnnn -- Here nnn is a constant value, encoded as a decimal integer -- (pos value for enumeration type case). Negative values have -- a trailing 'm' as usual. -- XSe -- Here e is the (unqualified) name of a constant entity in the -- same scope as the renaming which contains the subscript value. -- Slice -- For the slice case, we have two entries. The first is for the -- lower bound of the slice, and has the form: -- XLnnn -- XLe -- Specifies the lower bound, using exactly the same encoding as -- for an XS subscript as described above. -- Then the upper bound appears in the usual XSnnn/XSe form -- Selected component -- For a selected component, we have a single entry -- XRf -- Here f is the field name for the selection -- For an explicit dereference (.all), we have a single entry -- XA -- As an example, consider the declarations: -- package p is -- type q is record -- m : string (2 .. 5); -- end record; -- -- type r is array (1 .. 10, 1 .. 20) of q; -- -- g : r; -- -- z : string renames g (1,5).m(2 ..3) -- end p; -- The generated variable entity would appear as -- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type; -- p__g___XE--------------------outer entity is g -- XS1-----------------first subscript for g -- XS5--------------second subscript for g -- XRm-----------select field m -- XL2--------lower bound of slice -- XS3-----upper bound of slice -- Note that the type of the variable is a special internal type named -- _renaming_type. This type is an arbitrary type of zero size created -- in package Standard (see cstand.adb) and is ignored by the debugger. function Debug_Renaming_Declaration (N : Node_Id) return Node_Id; -- The argument N is a renaming declaration. The result is a variable -- declaration as described in the above paragraphs. If N is not a special -- debug declaration, then Empty is returned. This function also takes care -- of setting Materialize_Entity on the renamed entity where required. --------------------------- -- Packed Array Encoding -- --------------------------- -- For every constrained packed array, two types are created, and both -- appear in the debugging output: -- The original declared array type is a perfectly normal array type, and -- its index bounds indicate the original bounds of the array. -- The corresponding packed array type, which may be a modular type, or -- may be an array of bytes type (see Exp_Pakd for full details). This is -- the type that is actually used in the generated code and for debugging -- information for all objects of the packed type. -- The name of the corresponding packed array type is: -- ttt___XPnnn -- where -- ttt is the name of the original declared array -- nnn is the component size in bits (1-31) -- Note that if the packed array is not bit-packed, the name will simply -- be tttP. -- When the debugger sees that an object is of a type that is encoded in -- this manner, it can use the original type to determine the bounds and -- the component type, and the component size to determine the packing -- details. -- For an unconstrained packed array, the corresponding packed array type -- is neither used in the generated code nor for debugging information, -- only the original type is used. In order to convey the packing in the -- debugging information, the compiler generates the associated fat- and -- thin-pointer types (see the Pointers to Unconstrained Array section -- below) using the name of the corresponding packed array type as the -- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively. -- When the debugger sees that an object is of a type that is encoded in -- this manner, it can use the type of the fields to determine the bounds -- and the component type, and the component size to determine the packing -- details. ------------------------------------------- -- Packed Array Representation in Memory -- ------------------------------------------- -- Packed arrays are represented in tightly packed form, with no extra bits -- between components. This is true even when the component size is not a -- factor of the storage unit size, so that as a result it is possible for -- components to cross storage unit boundaries. -- The layout in storage is identical, regardless of whether the -- implementation type is a modular type or an array-of-bytes type. See -- Exp_Pakd for details of how these implementation types are used, but for -- the purpose of the debugger, only the starting address of the object in -- memory is significant. -- The following example should show clearly how the packing works in -- the little-endian and big-endian cases: -- type B is range 0 .. 7; -- for B'Size use 3; -- type BA is array (0 .. 5) of B; -- pragma Pack (BA); -- BV : constant BA := (1,2,3,4,5,6); -- Little endian case -- BV'Address + 2 BV'Address + 1 BV'Address + 0 -- +-----------------+-----------------+-----------------+ -- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 | -- +-----------------+-----------------+-----------------+ -- <---------> <-----> <---> <---> <-----> <---> <---> -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0) -- -- Big endian case -- -- BV'Address + 0 BV'Address + 1 BV'Address + 2 -- +-----------------+-----------------+-----------------+ -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? | -- +-----------------+-----------------+-----------------+ -- <---> <---> <-----> <---> <---> <-----> <---------> -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits -- Note that if a modular type is used to represent the array, the -- allocation in memory is not the same as a normal modular type. The -- difference occurs when the allocated object is larger than the size of -- the array. For a normal modular type, we extend the value on the left -- with zeroes. -- For example, in the normal modular case, if we have a 6-bit modular -- type, declared as mod 2**6, and we allocate an 8-bit object for this -- type, then we extend the value with two bits on the most significant -- end, and in either the little-endian or big-endian case, the value 63 -- is represented as 00111111 in binary in memory. -- For a modular type used to represent a packed array, the rule is -- different. In this case, if we have to extend the value, then we do it -- with undefined bits (which are not initialized and whose value is -- irrelevant to any generated code). Furthermore these bits are on the -- right (least significant bits) in the big-endian case, and on the left -- (most significant bits) in the little-endian case. -- For example, if we have a packed boolean array of 6 bits, all set to -- True, stored in an 8-bit object, then the value in memory in binary is -- ??111111 in the little-endian case, and 111111?? in the big-endian case. -- This is done so that the representation of packed arrays does not -- depend on whether we use a modular representation or array of bytes -- as previously described. This ensures that we can pass such values by -- reference in the case where a subprogram has to be able to handle values -- stored in either form. -- Note that when we extract the value of such a modular packed array, we -- expect to retrieve only the relevant bits, so in this same example, when -- we extract the value we get 111111 in both cases, and the code generated -- by the front end assumes this although it does not assume that any high -- order bits are defined. -- There are opportunities for optimization based on the knowledge that the -- unused bits are irrelevant for these type of packed arrays. For example -- if we have two such 6-bit-in-8-bit values and we do an assignment: -- a := b; -- Then logically, we extract the 6 bits and store only 6 bits in the -- result, but the back end is free to simply assign the entire 8-bits in -- this case, since we don't actually care about the undefined bits. -- However, in the equality case, it is important to ensure that the -- undefined bits do not participate in an equality test. -- If a modular packed array value is assigned to a register then logically -- it could always be held right justified, to avoid any need to shift, -- e.g. when doing comparisons. But probably this is a bad choice, as it -- would mean that an assignment such as a := above would require shifts -- when one value is in a register and the other value is in memory. ------------------------------------------------------ -- Subprograms for Handling Packed Array Type Names -- ------------------------------------------------------ function Make_Packed_Array_Impl_Type_Name (Typ : Entity_Id; Csize : Uint) return Name_Id; -- This function is used in Exp_Pakd to create the name that is encoded as -- described above. The entity Typ provides the name ttt, and the value -- Csize is the component size that provides the nnn value. -------------------------------------- -- Pointers to Unconstrained Arrays -- -------------------------------------- -- There are two kinds of 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. -- 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. -------------------------------------- -- 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). 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: -- member_name ::= {choice} | others_choice -- choice ::= simple_choice | range_choice -- simple_choice ::= S number -- range_choice ::= R number T number -- number ::= {decimal_digit} [m] -- others_choice ::= O (upper case letter O) -- The m in a number indicates a negative value. As an example of this -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by -- R1T4S7S10m -- In the case of enumeration values, the values used are the actual -- representation values in the case where an enumeration type has an -- enumeration representation spec (i.e. they are values that correspond -- to the use of the Enum_Rep attribute). -- The type of the inner record is given by the name of the union type (as -- above) concatenated with the above string. 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 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. -- WARNING: There is a matching C declaration of this subprogram in fe.h --------------------------------- -- Subtypes of Variant Records -- --------------------------------- -- A subtype of a variant record is represented by a type in which the -- union field from the base type is replaced by one of the possible -- values. For example, if we have: -- type Var (Disc : Boolean := True) is record -- M : Integer; -- case Disc is -- when True => -- R : Integer; -- S : Integer; -- when False => -- T : Integer; -- end case; -- end record; -- V1 : Var; -- V2 : Var (True); -- V3 : Var (False); -- Here V2, for example, is represented with a subtype whose name is -- something like TvarS3b, which is a struct with three fields. The first -- two fields are "disc" and "m" as for the base type, and the third field -- is S1, which contains the fields "r" and "s". -- The debugger should simply ignore structs with names of the form -- corresponding to variants, and consider the fields inside as belonging -- to the containing record. ----------------------------------------------- -- Extra renamings for subprogram instances -- ----------------------------------------------- procedure Build_Subprogram_Instance_Renamings (N : Node_Id; Wrapper : Entity_Id); -- The debugger has difficulties in recovering the value of actuals of an -- elementary type, from within the body of a subprogram instantiation. -- This is because such actuals generate an object declaration that is -- placed within the wrapper package of the instance, and the entity in -- these declarations is encoded in a complex way that GDB does not handle -- well. These new renaming declarations appear within the body of the -- subprogram, and are redundant from a visibility point of view, but They -- should have no measurable performance impact, and require no special -- decoding in the debugger. ------------------------------------------- -- Character literals in Character Types -- ------------------------------------------- -- Character types are enumeration types at least one of whose enumeration -- literals is a character literal. Enumeration literals are usually simply -- represented using their identifier names. If the enumeration literal is -- a character literal, the name is encoded as described in the following -- paragraph. -- The characters 'a'..'z' and '0'..'9' are represented as Qc, where 'c' -- stands for the character itself. A name QUhh, where each 'h' is a -- lower-case hexadecimal digit, stands for a character whose Unicode -- encoding is hh, and QWhhhh likewise stands for a wide character whose -- encoding is hhhh. The representation values are encoded as for ordinary -- enumeration literals (and have no necessary relationship to the values -- encoded in the names). -- For example, given the type declaration -- type x is (A, 'C', 'b'); -- the second enumeration literal would be named QU43 and the value -- assigned to it would be 1, and the third enumeration literal would be -- named Qb and the value assigned to it would be 2. ----------------------------------------------- -- Secondary Dispatch tables of tagged types -- ----------------------------------------------- procedure Get_Secondary_DT_External_Name (Typ : Entity_Id; Ancestor_Typ : Entity_Id; Suffix_Index : Int); -- Set Name_Buffer and Name_Len to the external name of one secondary -- dispatch table of Typ. If the interface has been inherited from some -- ancestor then Ancestor_Typ is such node (in this case the secondary DT -- is needed to handle overridden primitives); if there is no such ancestor -- then Ancestor_Typ is equal to Typ. -- -- Internal rule followed for the generation of the external name: -- -- Case 1. If the secondary dispatch has not been inherited from some -- ancestor of Typ then the external name is composed as -- follows: -- External_Name (Typ) + Suffix_Number + 'P' -- -- Case 2. if the secondary dispatch table has been inherited from some -- ancestor then the external name is composed as follows: -- External_Name (Typ) + '_' + External_Name (Ancestor_Typ) -- + Suffix_Number + 'P' -- -- Note: We have to use the external names (instead of simply their names) -- to protect the frontend against programs that give the same name to all -- the interfaces and use the expanded name to reference them. The -- Suffix_Number is used to differentiate all the secondary dispatch -- tables of a given type. -- -- Examples: -- -- package Pkg1 is | package Pkg2 is | package Pkg3 is -- type Typ is | type Typ is | type Typ is -- interface; | interface; | interface; -- end Pkg1; | end Pkg; | end Pkg3; -- -- with Pkg1, Pkg2, Pkg3; -- package Case_1 is -- type Typ is new Pkg1.Typ and Pkg2.Typ and Pkg3.Typ with ... -- end Case_1; -- -- with Case_1; -- package Case_2 is -- type Typ is new Case_1.Typ with ... -- end Case_2; -- -- These are the external names generated for Case_1.Typ (note that -- Pkg1.Typ is associated with the Primary Dispatch Table, because it -- is the parent of this type, and hence no external name is -- generated for it). -- case_1__typ0P (associated with Pkg2.Typ) -- case_1__typ1P (associated with Pkg3.Typ) -- -- These are the external names generated for Case_2.Typ: -- case_2__typ_case_1__typ0P -- case_2__typ_case_1__typ1P ---------------------------- -- Effect of Optimization -- ---------------------------- -- If the program is compiled with optimization on (e.g. -O1 switch -- specified), then there may be variations in the output from the above -- specification. In particular, objects may disappear from the output. -- This includes not only constants and variables that the program declares -- at the source level, but also the x___L and x___U constants created to -- describe the lower and upper bounds of subtypes with dynamic bounds. -- This means for example, that array bounds may disappear if optimization -- is turned on. The debugger is expected to recognize that these constants -- are missing and deal as best as it can with the limited information -- available. --------------------------------- -- GNAT Extensions to 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;