1 ------------------------------------------------------------------------------
3 -- GNAT COMPILER COMPONENTS --
9 -- Copyright (C) 1996-2020, Free Software Foundation, Inc. --
11 -- GNAT is free software; you can redistribute it and/or modify it under --
12 -- terms of the GNU General Public License as published by the Free Soft- --
13 -- ware Foundation; either version 3, or (at your option) any later ver- --
14 -- sion. GNAT is distributed in the hope that it will be useful, but WITH- --
15 -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY --
16 -- or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License --
17 -- for more details. You should have received a copy of the GNU General --
18 -- Public License distributed with GNAT; see file COPYING3. If not, go to --
19 -- http://www.gnu.org/licenses for a complete copy of the license. --
21 -- GNAT was originally developed by the GNAT team at New York University. --
22 -- Extensive contributions were provided by Ada Core Technologies Inc. --
24 ------------------------------------------------------------------------------
26 -- Expand routines for generation of special declarations used by the
27 -- debugger. In accordance with the Dwarf 2.2 specification, certain
28 -- type names are encoded to provide information to the debugger.
30 with Namet; use Namet;
31 with Types; use Types;
32 with Uintp; use Uintp;
36 -----------------------------------------------------
37 -- Encoding and Qualification of Names of Entities --
38 -----------------------------------------------------
40 -- This section describes how the names of entities are encoded in the
41 -- generated debugging information.
43 -- An entity in Ada has a name of the form X.Y.Z ... E where X,Y,Z are the
44 -- enclosing scopes (not including Standard at the start).
46 -- The encoding of the name follows this basic qualified naming scheme,
47 -- where the encoding of individual entity names is as described in Namet
48 -- (i.e. in particular names present in the original source are folded to
49 -- all lower case, with upper half and wide characters encoded as described
50 -- in Namet). Upper case letters are used only for entities generated by
53 -- There are two cases, global entities, and local entities. In more formal
54 -- terms, local entities are those which have a dynamic enclosing scope,
55 -- and global entities are at the library level, except that we always
56 -- consider procedures to be global entities, even if they are nested
57 -- (that's because at the debugger level a procedure name refers to the
58 -- code, and the code is indeed a global entity, including the case of
59 -- nested procedures.) In addition, we also consider all types to be global
60 -- entities, even if they are defined within a procedure.
62 -- The reason for treating all type names as global entities is that a
63 -- number of our type encodings work by having related type names, and we
64 -- need the full qualification to keep this unique.
66 -- For global entities, the encoded name includes all components of the
67 -- fully expanded name (but omitting Standard at the start). For example,
68 -- if a library-level child package P.Q has an embedded package R, and
69 -- there is an entity in this embedded package whose name is S, the encoded
70 -- name will include the components p.q.r.s.
72 -- For local entities, the encoded name only includes the components up to
73 -- the enclosing dynamic scope (other than a block). At run time, such a
74 -- dynamic scope is a subprogram, and the debugging formats know about
75 -- local variables of procedures, so it is not necessary to have full
76 -- qualification for such entities. In particular this means that direct
77 -- local variables of a procedure are not qualified.
79 -- For Ghost entities, the encoding adds a prefix "___ghost_" to aid the
80 -- detection of leaks of Ignored Ghost entities in the "living" space.
81 -- Ignored Ghost entities and any code associated with them should be
82 -- removed by the compiler in a post-processing pass. As a result,
83 -- object files should not contain any occurrences of this prefix.
85 -- As an example of the local name convention, consider a procedure V.W
86 -- with a local variable X, and a nested block Y containing an entity Z.
87 -- The fully qualified names of the entities X and Z are:
92 -- but since V.W is a subprogram, the encoded names will end up
98 -- The separating dots are translated into double underscores
100 -----------------------------
101 -- Handling of Overloading --
102 -----------------------------
104 -- The above scheme is incomplete for overloaded subprograms, since
105 -- overloading can legitimately result in case of two entities with
106 -- exactly the same fully qualified names. To distinguish between
107 -- entries in a set of overloaded subprograms, the encoded names are
108 -- serialized by adding the suffix:
110 -- __nn (two underscores)
112 -- where nn is a serial number (2 for the second overloaded function,
113 -- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no
114 -- suffix implies the first instance).
116 -- These names are prefixed by the normal full qualification. So for
117 -- example, the third instance of the subprogram qrs in package yz
118 -- would have the name:
122 -- A more subtle case arises with entities declared within overloaded
123 -- subprograms. If we have two overloaded subprograms, and both declare
124 -- an entity xyz, then the fully expanded name of the two xyz's is the
125 -- same. To distinguish these, we add the same __n suffix at the end of
126 -- the inner entity names.
128 -- In more complex cases, we can have multiple levels of overloading,
129 -- and we must make sure to distinguish which final declarative region
130 -- we are talking about. For this purpose, we use a more complex suffix
131 -- which has the form:
135 -- where the nn values are the homonym numbers as needed for any of the
136 -- qualifying entities, separated by a single underscore. If all the nn
137 -- values are 1, the suffix is omitted, Otherwise the suffix is present
138 -- (including any values of 1). The following example shows how this
141 -- package body Yz is
142 -- procedure Qrs is -- Name is yz__qrs
143 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv
144 -- begin ... end Qrs;
146 -- procedure Qrs (X: Int) is -- Name is yz__qrs__2
147 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1
148 -- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2
149 -- begin ... end Tuv;
151 -- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3
152 -- type m is new float; -- Name is yz__qrs__tuv__m__2_3
153 -- begin ... end Tuv;
154 -- begin ... end Qrs;
161 -- The above rules applied to operator names would result in names with
162 -- quotation marks, which are not typically allowed by assemblers and
163 -- linkers, and even if allowed would be odd and hard to deal with. To
164 -- avoid this problem, operator names are encoded as follows:
186 -- These names are prefixed by the normal full qualification, and
187 -- suffixed by the overloading identification. So for example, the
188 -- second operator "=" defined in package Extra.Messages would have
191 -- extra__messages__Oeq__2
193 ----------------------------------
194 -- Resolving Other Name Clashes --
195 ----------------------------------
197 -- It might be thought that the above scheme is complete, but in Ada 95,
198 -- full qualification is insufficient to uniquely identify an entity in
199 -- the program, even if it is not an overloaded subprogram. There are
200 -- two possible confusions:
204 -- interpretation 1: entity b in body of package a
205 -- interpretation 2: child procedure b of package a
209 -- interpretation 1: entity c in child package a.b
210 -- interpretation 2: entity c in nested package b in body of a
212 -- It is perfectly legal in both cases for both interpretations to be
213 -- valid within a single program. This is a bit of a surprise since
214 -- certainly in Ada 83, full qualification was sufficient, but not in
215 -- Ada 95. The result is that the above scheme can result in duplicate
216 -- names. This would not be so bad if the effect were just restricted
217 -- to debugging information, but in fact in both the above cases, it
218 -- is possible for both symbols to be external names, and so we have
219 -- a real problem of name clashes.
221 -- To deal with this situation, we provide two additional encoding
224 -- First: all library subprogram names are preceded by the string
225 -- _ada_ (which causes no duplications, since normal Ada names can
226 -- never start with an underscore. This not only solves the first
227 -- case of duplication, but also solves another pragmatic problem
228 -- which is that otherwise Ada procedures can generate names that
229 -- clash with existing system function names. Most notably, we can
230 -- have clashes in the case of procedure Main with the C main that
231 -- in some systems is always present.
233 -- Second, for the case where nested packages declared in package
234 -- bodies can cause trouble, we add a suffix which shows which
235 -- entities in the list are body-nested packages, i.e. packages
236 -- whose spec is within a package body. The rules are as follows,
237 -- given a list of names in a qualified name name1.name2....
239 -- If none are body-nested package entities, then there is no suffix
241 -- If at least one is a body-nested package entity, then the suffix
242 -- is X followed by a string of b's and n's (b = body-nested package
243 -- entity, n = not a body-nested package).
245 -- There is one element in this string for each entity in the encoded
246 -- expanded name except the first (the rules are such that the first
247 -- entity of the encoded expanded name can never be a body-nested'
248 -- package. Trailing n's are omitted, as is the last b (there must
249 -- be at least one b, or we would not be generating a suffix at all).
251 -- For example, suppose we have
254 -- pragma Elaborate_Body;
255 -- m1 : integer; -- #1
259 -- package y is m2 : integer; end y; -- #2
261 -- package z is r : integer; end z; -- #3
263 -- m3 : integer; -- #4
267 -- pragma Elaborate_Body;
268 -- m2 : integer; -- #5
271 -- package body x.y is
272 -- m3 : integer; -- #6
273 -- procedure j is -- #7
275 -- z : integer; -- #8
282 -- procedure x.m3 is begin null; end; -- #9
284 -- Then the encodings would be:
286 -- #1. x__m1 (no BNPE's in sight)
287 -- #2. x__y__m2X (y is a BNPE)
288 -- #3. x__y__z__rXb (y is a BNPE, so is z)
289 -- #4. x__m3 (no BNPE's in sight)
290 -- #5. x__y__m2 (no BNPE's in sight)
291 -- #6. x__y__m3 (no BNPE's in signt)
292 -- #7. x__y__j (no BNPE's in sight)
293 -- #8. k__z (no BNPE's, only up to procedure)
294 -- #9 _ada_x__m3 (library-level subprogram)
296 -- Note that we have instances here of both kind of potential name
297 -- clashes, and the above examples show how the encodings avoid the
300 -- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library
301 -- level subprogram, so it is preceded by the string _ada_ which acts
302 -- to distinguish it from the package body entity.
304 -- Lines #2 and #5 both refer to the entity x.y.m2, but the first
305 -- instance is inside the body-nested package y, so there is an X
306 -- suffix to distinguish it from the child library entity.
308 -- Note that enumeration literals never need Xb type suffixes, since
309 -- they are never referenced using global external names.
311 ---------------------
312 -- Interface Names --
313 ---------------------
315 -- Note: if an interface name is present, then the external name is
316 -- taken from the specified interface name. Given current limitations of
317 -- the gcc backend, this means that the debugging name is also set to
318 -- the interface name, but conceptually, it would be possible (and
319 -- indeed desirable) to have the debugging information still use the Ada
320 -- name as qualified above, so we still fully qualify the name in the
323 -------------------------------------
324 -- Encodings Related to Task Types --
325 -------------------------------------
327 -- Each task object defined by a single task declaration is associated
328 -- with a prefix that is used to qualify procedures defined in that
332 -- task body TaskObj is
333 -- procedure F1 is ... end;
339 -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1.
340 -- The body, B, is contained in a subprogram whose name is
343 ------------------------------------------
344 -- Encodings Related to Protected Types --
345 ------------------------------------------
347 -- Each protected type has an associated record type, that describes
348 -- the actual layout of the private data. In addition to the private
349 -- components of the type, the Corresponding_Record_Type includes one
350 -- component of type Protection, which is the actual lock structure.
351 -- The run-time size of the protected type is the size of the corres-
354 -- For a protected type prot, the Corresponding_Record_Type is encoded
357 -- The operations of a protected type are encoded as follows: each
358 -- operation results in two subprograms, a locking one that is called
359 -- from outside of the object, and a non-locking one that is used for
360 -- calls from other operations on the same object. The locking operation
361 -- simply acquires the lock, and then calls the non-locking version.
362 -- The names of all of these have a prefix constructed from the name of
363 -- the type, and a suffix which is P or N, depending on whether this is
364 -- the protected/non-locking version of the operation.
366 -- Operations generated for protected entries follow the same encoding.
367 -- Each entry results in two subprograms: a procedure that holds the
368 -- entry body, and a function that holds the evaluation of the barrier.
369 -- The names of these subprograms include the prefix '_E' or '_B' res-
370 -- pectively. The names also include a numeric suffix to render them
371 -- unique in the presence of overloaded entries.
373 -- Given the declaration:
375 -- protected type Lock is
376 -- function Get return Integer;
377 -- procedure Set (X: Integer);
378 -- entry Update (Val : Integer);
380 -- Value : Integer := 0;
383 -- the following operations are created:
394 -- If the protected type implements at least one interface, the
395 -- following additional operations are created:
401 -- These operations are used to ensure overriding of interface level
402 -- subprograms and proper dispatching on interface class-wide objects.
403 -- The bodies of these operations contain calls to their respective
404 -- protected versions:
406 -- function lock_get return Integer is
411 -- procedure lock_set (X : Integer) is
416 ----------------------------------------------------
417 -- Conversion between Entities and External Names --
418 ----------------------------------------------------
420 procedure Get_External_Name
422 Has_Suffix : Boolean := False;
423 Suffix : String := "");
424 -- Set Name_Buffer and Name_Len to the external name of the entity. The
425 -- external name is the Interface_Name, if specified, unless the entity
426 -- has an address clause or Has_Suffix is true.
428 -- If the Interface is not present, or not used, the external name is the
431 -- - the string "_ada_", if the entity is a library subprogram,
432 -- - the names of any enclosing scopes, each followed by "__",
433 -- or "X_" if the next entity is a subunit)
434 -- - the name of the entity
435 -- - the string "$" (or "__" if target does not allow "$"), followed
436 -- by homonym suffix, if the entity is an overloaded subprogram
437 -- or is defined within an overloaded subprogram.
438 -- - the string "___" followed by Suffix if Has_Suffix is true.
440 -- Note that a call to this procedure has no effect if we are not
441 -- generating code, since the necessary information for computing the
442 -- proper external name is not available in this case.
444 -- WARNING: There is a matching C declaration of this subprogram in fe.h
446 -------------------------------------
447 -- Encoding for translation into C --
448 -------------------------------------
450 -- In Modify_Tree_For_C mode we must add encodings to dismabiguate cases
451 -- where Ada block structure cannot be directly translated. These cases
454 -- a) A loop variable may hide a homonym in an enclosing block
455 -- b) A block-local variable may hide a homonym in an enclosing block
457 -- In C these constructs are not scopes and we must distinguish the names
458 -- explicitly. In the first case we create a qualified name with the suffix
459 -- 'L', in the second case with a suffix 'B'.
461 --------------------------------------------
462 -- Subprograms for Handling Qualification --
463 --------------------------------------------
465 procedure Qualify_Entity_Names (N : Node_Id);
466 -- Given a node N, that represents a block, subprogram body, or package
467 -- body or spec, or protected or task type, sets a fully qualified name
468 -- for the defining entity of given construct, and also sets fully
469 -- qualified names for all enclosed entities of the construct (using
470 -- First_Entity/Next_Entity). Note that the actual modifications of the
471 -- names is postponed till a subsequent call to Qualify_All_Entity_Names.
472 -- Note: this routine does not deal with prepending _ada_ to library
473 -- subprogram names. The reason for this is that we only prepend _ada_
474 -- to the library entity itself, and not to names built from this name.
476 procedure Qualify_All_Entity_Names;
477 -- When Qualify_Entity_Names is called, no actual name changes are made,
478 -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call
479 -- is made to this procedure. The reason for this deferral is that when
480 -- names are changed semantic processing may be affected. By deferring
481 -- the changes till just before gigi is called, we avoid any concerns
482 -- about such effects. Gigi itself does not use the names except for
483 -- output of names for debugging purposes (which is why we are doing
484 -- the name changes in the first place).
486 -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are
487 -- useful to remove qualification from a name qualified by the call to
488 -- Qualify_All_Entity_Names.
490 --------------------------------
491 -- Handling of Numeric Values --
492 --------------------------------
494 -- All numeric values here are encoded as strings of decimal digits. Only
495 -- integer values need to be encoded. A negative value is encoded as the
496 -- corresponding positive value followed by a lower case m for minus to
497 -- indicate that the value is negative (e.g. 2m for -2).
499 -------------------------
500 -- Type Name Encodings --
501 -------------------------
503 -- In the following typ is the name of the type as normally encoded by the
504 -- debugger rules, i.e. a non-qualified name, all in lower case, with
505 -- standard encoding of upper half and wide characters
507 ------------------------
508 -- Encapsulated Types --
509 ------------------------
511 -- In some cases, the compiler encapsulates a type by wrapping it in a
512 -- structure. For example, this is used when a size or alignment
513 -- specification requires a larger type. Consider:
515 -- type y is mod 2 ** 64;
516 -- for y'size use 256;
518 -- In this case the compile generates a structure type y___PAD, which
519 -- has a single field whose name is F. This single field is 64 bits
520 -- long and contains the actual value. This kind of padding is used
521 -- when the logical value to be stored is shorter than the object in
522 -- which it is allocated. For example if a size clause is used to set
523 -- a size of 256 for a signed integer value, then a typical choice is
524 -- to wrap a 64-bit integer in a 256 bit PAD structure.
526 -- A similar encapsulation is done for some packed array types, in which
527 -- case the structure type is y___JM and the field name is OBJECT.
528 -- This is used in the case of a packed array stored using modular
529 -- representation (see section on representation of packed array
530 -- objects). In this case the JM wrapping is used to achieve correct
531 -- positioning of the packed array value (left or right justified in its
532 -- field depending on endianness.
534 -- When the debugger sees an object of a type whose name has a suffix of
535 -- ___PAD or ___JM, the type will be a record containing a single field,
536 -- and the name of that field will be all upper case. In this case, it
537 -- should look inside to get the value of the inner field, and neither
538 -- the outer structure name, nor the field name should appear when the
541 -- When the debugger sees a record named REP being a field inside
542 -- another record, it should treat the fields inside REP as being part
543 -- of the outer record (this REP field is only present for code
544 -- generation purposes). The REP record should not appear in the values
545 -- printed by the debugger.
547 -----------------------
548 -- Fixed-Point Types --
549 -----------------------
551 -- Fixed-point types are encoded using a suffix that indicates the
552 -- delta and small values. The actual type itself is a normal integer
556 -- typ___XF_nn_dd_nn_dd
558 -- The first form is used when small = delta. The value of delta (and
559 -- small) is given by the rational nn/dd, where nn and dd are decimal
562 -- The second form is used if the small value is different from the
563 -- delta. In this case, the first nn/dd rational value is for delta,
564 -- and the second value is for small.
570 -- Discrete types are coded with a suffix indicating the range in the
571 -- case where one or both of the bounds are discriminants or variable.
573 -- Note: at the current time, we also encode compile time known bounds
574 -- if they do not match the natural machine type bounds, but this may
575 -- be removed in the future, since it is redundant for most debugging
576 -- formats. However, we do not ever need XD encoding for enumeration
577 -- base types, since here it is always clear what the bounds are from
578 -- the total number of enumeration literals.
581 -- typ___XDL_lowerbound
582 -- typ___XDU_upperbound
583 -- typ___XDLU_lowerbound__upperbound
585 -- If a discrete type is a natural machine type (i.e. its bounds
586 -- correspond in a natural manner to its size), then it is left
587 -- unencoded. The above encoding forms are used when there is a
588 -- constrained range that does not correspond to the size or that
589 -- has discriminant references or other compile time known bounds.
591 -- The first form is used if both bounds are dynamic, in which case two
592 -- constant objects are present whose names are typ___L and typ___U in
593 -- the same scope as typ, and the values of these constants indicate
594 -- the bounds. As far as the debugger is concerned, these are simply
595 -- variables that can be accessed like any other variables. In the
596 -- enumeration case, these values correspond to the Enum_Rep values for
597 -- the lower and upper bounds.
599 -- The second form is used if the upper bound is dynamic, but the lower
600 -- bound is either constant or depends on a discriminant of the record
601 -- with which the type is associated. The upper bound is stored in a
602 -- constant object of name typ___U as previously described, but the
603 -- lower bound is encoded directly into the name as either a decimal
604 -- integer, or as the discriminant name.
606 -- The third form is similarly used if the lower bound is dynamic, but
607 -- the upper bound is compile time known or a discriminant reference,
608 -- in which case the lower bound is stored in a constant object of name
609 -- typ___L, and the upper bound is encoded directly into the name as
610 -- either a decimal integer, or as the discriminant name.
612 -- The fourth form is used if both bounds are discriminant references
613 -- or compile time known values, with the encoding first for the lower
614 -- bound, then for the upper bound, as previously described.
624 -- Is encoded as a subrange of an unsigned base type with lower bound
625 -- zero and upper bound N. That is, there is no name encoding. We use
626 -- the standard encodings provided by the debugging format. Thus we
627 -- give these types a non-standard interpretation: the standard
628 -- interpretation of our encoding would not, in general, imply that
629 -- arithmetic on type x was to be performed modulo N (especially not
630 -- when N is not a power of 2).
636 -- Only discrete types can be biased, and the fact that they are biased
637 -- is indicated by a suffix of the form:
639 -- typ___XB_lowerbound__upperbound
641 -- Here lowerbound and upperbound are decimal integers, with the usual
642 -- (postfix "m") encoding for negative numbers. Biased types are only
643 -- possible where the bounds are compile time known, and the values are
644 -- represented as unsigned offsets from the lower bound given. For
647 -- type Q is range 10 .. 15;
650 -- The size clause will force values of type Q in memory to be stored
651 -- in biased form (e.g. 11 will be represented by the bit pattern 001).
653 ----------------------------------------------
654 -- Record Types with Variable-Length Fields --
655 ----------------------------------------------
657 -- The debugging formats do not fully support these types, and indeed
658 -- some formats simply generate no useful information at all for such
659 -- types. In order to provide information for the debugger, gigi creates
660 -- a parallel type in the same scope with one of the names
665 -- The former name is used for a record and the latter for the union
666 -- that is made for a variant record (see below) if that record or union
667 -- has a field of variable size or if the record or union itself has a
668 -- variable size. These encodings suffix any other encodings that that
669 -- might be suffixed to the type name.
671 -- The idea here is to provide all the needed information to interpret
672 -- objects of the original type in the form of a "fixed up" type, which
673 -- is representable using the normal debugging information.
675 -- There are three cases to be dealt with. First, some fields may have
676 -- variable positions because they appear after variable-length fields.
677 -- To deal with this, we encode *all* the field bit positions of the
678 -- special ___XV type in a non-standard manner.
680 -- The idea is to encode not the position, but rather information that
681 -- allows computing the position of a field from the position of the
682 -- previous field. The algorithm for computing the actual positions of
683 -- all fields and the length of the record is as follows. In this
684 -- description, let P represent the current bit position in the record.
686 -- 1. Initialize P to 0
688 -- 2. For each field in the record:
690 -- 2a. If an alignment is given (see below), then round P up, if
691 -- needed, to the next multiple of that alignment.
693 -- 2b. If a bit position is given, then increment P by that amount
694 -- (that is, treat it as an offset from the end of the preceding
697 -- 2c. Assign P as the actual position of the field
699 -- 2d. Compute the length, L, of the represented field (see below)
700 -- and compute P'=P+L. Unless the field represents a variant part
701 -- (see below and also Variant Record Encoding), set P to P'.
703 -- The alignment, if present, is encoded in the field name of the
704 -- record, which has a suffix:
708 -- where the nn after the XVA indicates the alignment value in storage
709 -- units. This encoding is present only if an alignment is present.
711 -- The size of the record described by an XVE-encoded type (in bits) is
712 -- generally the maximum value attained by P' in step 2d above, rounded
713 -- up according to the record's alignment.
715 -- Second, the variable-length fields themselves are represented by
716 -- replacing the type by a special access type. The designated type of
717 -- this access type is the original variable-length type, and the fact
718 -- that this field has been transformed in this way is signalled by
719 -- encoding the field name as:
723 -- where field is the original field name. If a field is both
724 -- variable-length and also needs an alignment encoding, then the
725 -- encodings are combined using:
729 -- Note: the reason that we change the type is so that the resulting
730 -- type has no variable-length fields. At least some of the formats used
731 -- for debugging information simply cannot tolerate variable- length
732 -- fields, so the encoded information would get lost.
734 -- Third, in the case of a variant record, the special union that
735 -- contains the variants is replaced by a normal C union. In this case,
736 -- the positions are all zero.
738 -- Discriminants appear before any variable-length fields that depend on
739 -- them, with one exception. In some cases, a discriminant governing the
740 -- choice of a variant clause may appear in the list of fields of an XVE
741 -- type after the entry for the variant clause itself (this can happen
742 -- in the presence of a representation clause for the record type in the
743 -- source program). However, when this happens, the discriminant's
744 -- position may be determined by first applying the rules described in
745 -- this section, ignoring the variant clause. As a result, discriminants
746 -- can always be located independently of the variable-length fields
747 -- that depend on them.
749 -- The size of the ___XVE or ___XVU record or union is set to the
750 -- alignment (in bytes) of the original object so that the debugger
751 -- can calculate the size of the original type.
753 -- As an example of this encoding, consider the declarations:
755 -- type Q is array (1 .. V1) of Float; -- alignment 4
756 -- type R is array (1 .. V2) of Long_Float; -- alignment 8
761 -- C : String (1 .. V3);
768 -- The encoded type looks like:
770 -- type anonymousQ is access Q;
771 -- type anonymousR is access R;
773 -- type X___XVE is record
774 -- A : Character; -- position contains 0
775 -- B : Float; -- position contains 24
776 -- C___XVL : access String (1 .. V3); -- position contains 0
777 -- D___XVA4 : Float; -- position contains 0
778 -- E___XVL4 : anonymousQ; -- position contains 0
779 -- F___XVL8 : anonymousR; -- position contains 0
780 -- G : Float; -- position contains 0
783 -- Any bit sizes recorded for fields other than dynamic fields and
784 -- variants are honored as for ordinary records.
788 -- 1) The B field could also have been encoded by using a position of
789 -- zero and an alignment of 4, but in such a case the coding by position
790 -- is preferred (since it takes up less space). We have used the
791 -- (illegal) notation access xxx as field types in the example above.
793 -- 2) The E field does not actually need the alignment indication but
794 -- this may not be detected in this case by the conversion routines.
796 -- 3) Our conventions do not cover all XVE-encoded records in which
797 -- some, but not all, fields have representation clauses. Such records
798 -- may, therefore, be displayed incorrectly by debuggers. This situation
801 -----------------------
802 -- Base Record Types --
803 -----------------------
805 -- Under certain circumstances, debuggers need two descriptions of a
806 -- record type, one that gives the actual details of the base type's
807 -- structure (as described elsewhere in these comments) and one that may
808 -- be used to obtain information about the particular subtype and the
809 -- size of the objects being typed. In such cases the compiler will
810 -- substitute type whose name is typically compiler-generated and
811 -- irrelevant except as a key for obtaining the actual type.
813 -- Specifically, if this name is x, then we produce a record type named
814 -- x___XVS consisting of one field. The name of this field is that of
815 -- the actual type being encoded, which we'll call y. The type of this
816 -- single field can be either an arbitrary non-reference type, e.g. an
817 -- integer type, or a reference type; in the latter case, the referenced
818 -- type is also the actual type being encoded y. Both x and y may have
819 -- corresponding ___XVE types.
821 -- The size of the objects typed as x should be obtained from the
822 -- structure of x (and x___XVE, if applicable) as for ordinary types
823 -- unless there is a variable named x___XVZ, which, if present, will
824 -- hold the size (in bytes) of x. In this latter case, the size of the
825 -- x___XVS type will not be a constant but a reference to x___XVZ.
827 -- The type x will either be a subtype of y (see also Subtypes of
828 -- Variant Records, below) or will contain a single field of type y,
829 -- or no fields at all. The layout, types, and positions of these
830 -- fields will be accurate, if present. (Currently, however, the GDB
831 -- debugger makes no use of x except to determine its size).
833 -- Among other uses, XVS types are used to encode unconstrained types.
834 -- For example, given:
836 -- subtype Int is INTEGER range 0..10;
837 -- type T1 (N: Int := 0) is record
838 -- F1: String (1 .. N);
840 -- type AT1 is array (INTEGER range <>) of T1;
842 -- the element type for AT1 might have a type defined as if it had
845 -- type at1___PAD is record F : T1; end record;
846 -- for at1___PAD'Size use 16 * 8;
848 -- and there would also be:
850 -- type at1___PAD___XVS is record t1: reft1; end record;
852 -- type reft1 is <reference to t1>
854 -- Had the subtype Int been dynamic:
856 -- subtype Int is INTEGER range 0 .. M; -- M a variable
858 -- Then the compiler would also generate a declaration whose effect
861 -- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term;
863 -- Not all unconstrained types are so encoded; the XVS convention may be
864 -- unnecessary for unconstrained types of fixed size. However, this
865 -- encoding is always necessary when a subcomponent type (array
866 -- element's type or record field's type) is an unconstrained record
867 -- type some of whose components depend on discriminant values.
873 -- Since there is no way for the debugger to obtain the index subtypes
874 -- for an array type, we produce a type that has the name of the array
875 -- type followed by "___XA" and is a record type whose field types are
876 -- the respective types for the bounds (and whose field names are the
877 -- names of these types).
879 -- To conserve space, we do not produce this type unless one of the
880 -- index types is either an enumeration type, has a variable lower or
881 -- upper bound or is a biased type.
883 -- Given the full encoding of these types (see above description for
884 -- the encoding of discrete types), this means that all necessary
885 -- information for addressing arrays is available. In some debugging
886 -- formats, some or all of the bounds information may be available
887 -- redundantly, particularly in the fixed-point case, but this
888 -- information can in any case be ignored by the debugger.
890 ----------------------------
891 -- Note on Implicit Types --
892 ----------------------------
894 -- The compiler creates implicit type names in many situations where a
895 -- type is present semantically, but no specific name is present. For
898 -- S : Integer range M .. N;
900 -- Here the subtype of S is not integer, but rather an anonymous subtype
901 -- of Integer. Where possible, the compiler generates names for such
902 -- anonymous types that are related to the type from which the subtype
903 -- is obtained as follows:
907 -- where name is the name from which the subtype is obtained, using
908 -- lower case letters and underscores, and suffix starts with an upper
909 -- case letter. For example the name for the above declaration might be:
913 -- If the debugger is asked to give the type of an entity and the type
914 -- has the form T name suffix, it is probably appropriate to just use
915 -- "name" in the response since this is what is meaningful to the
918 -------------------------------------------------
919 -- Subprograms for Handling Encoded Type Names --
920 -------------------------------------------------
922 procedure Get_Encoded_Name (E : Entity_Id);
923 -- If the entity is a typename, store the external name of the entity as in
924 -- Get_External_Name, followed by three underscores plus the type encoding
925 -- in Name_Buffer with the length in Name_Len, and an ASCII.NUL character
926 -- stored following the name. Otherwise set Name_Buffer and Name_Len to
927 -- hold the entity name. Note that a call to this procedure has no effect
928 -- if we are not generating code, since the necessary information for
929 -- computing the proper encoded name is not available in this case.
931 -- WARNING: There is a matching C declaration of this subprogram in fe.h
937 -- Debugging information is generated for exception, object, package, and
938 -- subprogram renaming (generic renamings are not significant, since
939 -- generic templates are not relevant at debugging time).
941 -- Consider a renaming declaration of the form
943 -- x : typ renames y;
945 -- There is one case in which no special debugging information is required,
946 -- namely the case of an object renaming where the back end allocates a
947 -- reference for the renamed variable, and the entity x is this reference.
948 -- The debugger can handle this case without any special processing or
949 -- encoding (it won't know it was a renaming, but that does not matter).
951 -- All other cases of renaming generate a dummy variable for an entity
952 -- whose name is of the form:
954 -- x___XR_... for an object renaming
955 -- x___XRE_... for an exception renaming
956 -- x___XRP_... for a package renaming
958 -- and where the "..." represents a suffix that describes the structure of
959 -- the object name given in the renaming (see details below).
961 -- The name is fully qualified in the usual manner, i.e. qualified in the
962 -- same manner as the entity x would be. In the case of a package renaming
963 -- where x is a child unit, the qualification includes the name of the
964 -- parent unit, to disambiguate child units with the same simple name and
965 -- (of necessity) different parents.
967 -- Note: subprogram renamings are not encoded at the present time
969 -- The suffix of the variable name describing the renamed object is defined
970 -- to use the following encoding:
972 -- For the simple entity case, where y is just an entity name, the suffix
977 -- i.e. the suffix has a single field, the first part matching the
978 -- name y, followed by a "___" separator, ending with sequence XE.
979 -- The entity name portion is fully qualified in the usual manner.
980 -- This same naming scheme is followed for all forms of encoded
981 -- renamings that rename a simple entity.
983 -- For the object renaming case where y is a selected component or an
984 -- indexed component, the variable name is suffixed by additional fields
985 -- that give details of the components. The name starts as above with a
986 -- y___XE name indicating the outer level object entity. Then a series of
987 -- selections and indexing operations can be specified as follows:
991 -- A series of subscript values appear in sequence, the number
992 -- corresponds to the number of dimensions of the array. The
993 -- subscripts have one of the following two forms:
997 -- Here nnn is a constant value, encoded as a decimal integer
998 -- (pos value for enumeration type case). Negative values have
999 -- a trailing 'm' as usual.
1003 -- Here e is the (unqualified) name of a constant entity in the
1004 -- same scope as the renaming which contains the subscript value.
1008 -- For the slice case, we have two entries. The first is for the
1009 -- lower bound of the slice, and has the form:
1014 -- Specifies the lower bound, using exactly the same encoding as
1015 -- for an XS subscript as described above.
1017 -- Then the upper bound appears in the usual XSnnn/XSe form
1019 -- Selected component
1021 -- For a selected component, we have a single entry
1025 -- Here f is the field name for the selection
1027 -- For an explicit dereference (.all), we have a single entry
1031 -- As an example, consider the declarations:
1035 -- m : string (2 .. 5);
1038 -- type r is array (1 .. 10, 1 .. 20) of q;
1042 -- z : string renames g (1,5).m(2 ..3)
1045 -- The generated variable entity would appear as
1047 -- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type;
1048 -- p__g___XE--------------------outer entity is g
1049 -- XS1-----------------first subscript for g
1050 -- XS5--------------second subscript for g
1051 -- XRm-----------select field m
1052 -- XL2--------lower bound of slice
1053 -- XS3-----upper bound of slice
1055 -- Note that the type of the variable is a special internal type named
1056 -- _renaming_type. This type is an arbitrary type of zero size created
1057 -- in package Standard (see cstand.adb) and is ignored by the debugger.
1059 function Debug_Renaming_Declaration (N : Node_Id) return Node_Id;
1060 -- The argument N is a renaming declaration. The result is a variable
1061 -- declaration as described in the above paragraphs. If N is not a special
1062 -- debug declaration, then Empty is returned. This function also takes care
1063 -- of setting Materialize_Entity on the renamed entity where required.
1065 ---------------------------
1066 -- Packed Array Encoding --
1067 ---------------------------
1069 -- For every constrained packed array, two types are created, and both
1070 -- appear in the debugging output:
1072 -- The original declared array type is a perfectly normal array type, and
1073 -- its index bounds indicate the original bounds of the array.
1075 -- The corresponding packed array type, which may be a modular type, or
1076 -- may be an array of bytes type (see Exp_Pakd for full details). This is
1077 -- the type that is actually used in the generated code and for debugging
1078 -- information for all objects of the packed type.
1080 -- The name of the corresponding packed array type is:
1086 -- ttt is the name of the original declared array
1087 -- nnn is the component size in bits (1-31)
1089 -- Note that if the packed array is not bit-packed, the name will simply
1092 -- When the debugger sees that an object is of a type that is encoded in
1093 -- this manner, it can use the original type to determine the bounds and
1094 -- the component type, and the component size to determine the packing
1097 -- For an unconstrained packed array, the corresponding packed array type
1098 -- is neither used in the generated code nor for debugging information,
1099 -- only the original type is used. In order to convey the packing in the
1100 -- debugging information, the compiler generates the associated fat- and
1101 -- thin-pointer types (see the Pointers to Unconstrained Array section
1102 -- below) using the name of the corresponding packed array type as the
1103 -- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively.
1105 -- When the debugger sees that an object is of a type that is encoded in
1106 -- this manner, it can use the type of the fields to determine the bounds
1107 -- and the component type, and the component size to determine the packing
1110 -------------------------------------------
1111 -- Packed Array Representation in Memory --
1112 -------------------------------------------
1114 -- Packed arrays are represented in tightly packed form, with no extra bits
1115 -- between components. This is true even when the component size is not a
1116 -- factor of the storage unit size, so that as a result it is possible for
1117 -- components to cross storage unit boundaries.
1119 -- The layout in storage is identical, regardless of whether the
1120 -- implementation type is a modular type or an array-of-bytes type. See
1121 -- Exp_Pakd for details of how these implementation types are used, but for
1122 -- the purpose of the debugger, only the starting address of the object in
1123 -- memory is significant.
1125 -- The following example should show clearly how the packing works in
1126 -- the little-endian and big-endian cases:
1128 -- type B is range 0 .. 7;
1129 -- for B'Size use 3;
1131 -- type BA is array (0 .. 5) of B;
1132 -- pragma Pack (BA);
1134 -- BV : constant BA := (1,2,3,4,5,6);
1136 -- Little endian case
1138 -- BV'Address + 2 BV'Address + 1 BV'Address + 0
1139 -- +-----------------+-----------------+-----------------+
1140 -- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 |
1141 -- +-----------------+-----------------+-----------------+
1142 -- <---------> <-----> <---> <---> <-----> <---> <--->
1143 -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0)
1147 -- BV'Address + 0 BV'Address + 1 BV'Address + 2
1148 -- +-----------------+-----------------+-----------------+
1149 -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? |
1150 -- +-----------------+-----------------+-----------------+
1151 -- <---> <---> <-----> <---> <---> <-----> <--------->
1152 -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits
1154 -- Note that if a modular type is used to represent the array, the
1155 -- allocation in memory is not the same as a normal modular type. The
1156 -- difference occurs when the allocated object is larger than the size of
1157 -- the array. For a normal modular type, we extend the value on the left
1160 -- For example, in the normal modular case, if we have a 6-bit modular
1161 -- type, declared as mod 2**6, and we allocate an 8-bit object for this
1162 -- type, then we extend the value with two bits on the most significant
1163 -- end, and in either the little-endian or big-endian case, the value 63
1164 -- is represented as 00111111 in binary in memory.
1166 -- For a modular type used to represent a packed array, the rule is
1167 -- different. In this case, if we have to extend the value, then we do it
1168 -- with undefined bits (which are not initialized and whose value is
1169 -- irrelevant to any generated code). Furthermore these bits are on the
1170 -- right (least significant bits) in the big-endian case, and on the left
1171 -- (most significant bits) in the little-endian case.
1173 -- For example, if we have a packed boolean array of 6 bits, all set to
1174 -- True, stored in an 8-bit object, then the value in memory in binary is
1175 -- ??111111 in the little-endian case, and 111111?? in the big-endian case.
1177 -- This is done so that the representation of packed arrays does not
1178 -- depend on whether we use a modular representation or array of bytes
1179 -- as previously described. This ensures that we can pass such values by
1180 -- reference in the case where a subprogram has to be able to handle values
1181 -- stored in either form.
1183 -- Note that when we extract the value of such a modular packed array, we
1184 -- expect to retrieve only the relevant bits, so in this same example, when
1185 -- we extract the value we get 111111 in both cases, and the code generated
1186 -- by the front end assumes this although it does not assume that any high
1187 -- order bits are defined.
1189 -- There are opportunities for optimization based on the knowledge that the
1190 -- unused bits are irrelevant for these type of packed arrays. For example
1191 -- if we have two such 6-bit-in-8-bit values and we do an assignment:
1195 -- Then logically, we extract the 6 bits and store only 6 bits in the
1196 -- result, but the back end is free to simply assign the entire 8-bits in
1197 -- this case, since we don't actually care about the undefined bits.
1198 -- However, in the equality case, it is important to ensure that the
1199 -- undefined bits do not participate in an equality test.
1201 -- If a modular packed array value is assigned to a register then logically
1202 -- it could always be held right justified, to avoid any need to shift,
1203 -- e.g. when doing comparisons. But probably this is a bad choice, as it
1204 -- would mean that an assignment such as a := above would require shifts
1205 -- when one value is in a register and the other value is in memory.
1207 ------------------------------------------------------
1208 -- Subprograms for Handling Packed Array Type Names --
1209 ------------------------------------------------------
1211 function Make_Packed_Array_Impl_Type_Name
1213 Csize : Uint) return Name_Id;
1214 -- This function is used in Exp_Pakd to create the name that is encoded as
1215 -- described above. The entity Typ provides the name ttt, and the value
1216 -- Csize is the component size that provides the nnn value.
1218 --------------------------------------
1219 -- Pointers to Unconstrained Arrays --
1220 --------------------------------------
1222 -- There are two kinds of pointers to arrays. The debugger can tell which
1223 -- format is in use by the form of the type of the pointer.
1227 -- Fat pointers are represented as a struct with two fields. This
1228 -- struct has two distinguished field names:
1230 -- P_ARRAY is a pointer to the array type. The name of this type is
1231 -- the unconstrained type followed by "___XUA". This array will have
1232 -- bounds which are the discriminants, and hence are unparsable, but
1233 -- will give the number of subscripts and the component type.
1235 -- P_BOUNDS is a pointer to a struct, the name of whose type is the
1236 -- unconstrained array name followed by "___XUB" and which has
1237 -- fields of the form
1239 -- LBn (n a decimal integer) lower bound of n'th dimension
1240 -- UBn (n a decimal integer) upper bound of n'th dimension
1242 -- The bounds may be any integral type. In the case of an enumeration
1243 -- type, Enum_Rep values are used.
1245 -- For a given unconstrained array type, the compiler will generate one
1246 -- fat-pointer type whose name is "arr___XUP", where "arr" is the name
1247 -- of the array type, and use it to represent the array type itself in
1248 -- the debugging information.
1250 -- For each pointer to this unconstrained array type, the compiler will
1251 -- generate a typedef that points to the above "arr___XUP" fat-pointer
1252 -- type. As a consequence, when it comes to fat-pointer types:
1254 -- 1. The type name is given by the typedef
1256 -- 2. If the debugger is asked to output the type, the appropriate
1257 -- form is "access arr", except if the type name is "arr___XUP"
1258 -- for which it is the array definition.
1262 -- The value of a thin pointer is a pointer to the second field of a
1263 -- structure with two fields. The name of this structure's type is
1264 -- "arr___XUT", where "arr" is the name of the unconstrained array
1265 -- type. Even though it actually points into middle of this structure,
1266 -- the thin pointer's type in debugging information is
1267 -- pointer-to-arr___XUT.
1269 -- The first field of arr___XUT is named BOUNDS, and has a type named
1270 -- arr___XUB, with the structure described for such types in fat
1271 -- pointers, as described above.
1273 -- The second field of arr___XUT is named ARRAY, and contains the
1274 -- actual array. Because this array has a dynamic size, determined by
1275 -- the BOUNDS field that precedes it, all of the information about
1276 -- arr___XUT is encoded in a parallel type named arr___XUT___XVE, with
1277 -- fields BOUNDS and ARRAY___XVL. As for previously described ___XVE
1278 -- types, ARRAY___XVL has a pointer-to-array type. However, the array
1279 -- type in this case is named arr___XUA and only its element type is
1280 -- meaningful, just as described for fat pointers.
1282 --------------------------------------
1283 -- Tagged Types and Type Extensions --
1284 --------------------------------------
1286 -- A type C derived from a tagged type P has a field named "_parent" of
1287 -- type P that contains its inherited fields. The type of this field is
1288 -- usually P (encoded as usual if it has a dynamic size), but may be a more
1289 -- distant ancestor, if P is a null extension of that type.
1291 -- The type tag of a tagged type is a field named _tag, of type void*. If
1292 -- the type is derived from another tagged type, its _tag field is found in
1293 -- its _parent field.
1295 -----------------------------
1296 -- Variant Record Encoding --
1297 -----------------------------
1299 -- The variant part of a variant record is encoded as a single field in the
1300 -- enclosing record, whose name is:
1304 -- where discrim is the unqualified name of the variant. This field name is
1305 -- built by gigi (not by code in this unit). For Unchecked_Union record,
1306 -- this discriminant will not appear in the record (see Unchecked Unions,
1309 -- The type corresponding to this field has a name that is obtained by
1310 -- concatenating the type name with the above string and is similar to a C
1311 -- union, in which each member of the union corresponds to one variant.
1312 -- However, unlike a C union, the size of the type may be variable even if
1313 -- each of the components are fixed size, since it includes a computation
1314 -- of which variant is present. In that case, it will be encoded as above
1315 -- and a type with the suffix "___XVN___XVU" will be present.
1317 -- The name of the union member is encoded to indicate the choices, and
1318 -- is a string given by the following grammar:
1320 -- member_name ::= {choice} | others_choice
1321 -- choice ::= simple_choice | range_choice
1322 -- simple_choice ::= S number
1323 -- range_choice ::= R number T number
1324 -- number ::= {decimal_digit} [m]
1325 -- others_choice ::= O (upper case letter O)
1327 -- The m in a number indicates a negative value. As an example of this
1328 -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by
1332 -- In the case of enumeration values, the values used are the actual
1333 -- representation values in the case where an enumeration type has an
1334 -- enumeration representation spec (i.e. they are values that correspond
1335 -- to the use of the Enum_Rep attribute).
1337 -- The type of the inner record is given by the name of the union type (as
1338 -- above) concatenated with the above string. Since that type may itself be
1339 -- variable-sized, it may also be encoded as above with a new type with a
1340 -- further suffix of "___XVU".
1342 -- As an example, consider:
1344 -- type Var (Disc : Boolean := True) is record
1359 -- In this case, the type var is represented as a struct with three fields.
1360 -- The first two are "disc" and "m", representing the values of these
1361 -- record components. The third field is a union of two types, with field
1362 -- names S1 and O. S1 is a struct with fields "r" and "s", and O is a
1363 -- struct with field "t".
1365 ----------------------
1366 -- Unchecked Unions --
1367 ----------------------
1369 -- The encoding for variant records changes somewhat under the influence
1370 -- of a "pragma Unchecked_Union" clause:
1372 -- 1. The discriminant will not be present in the record, although its
1373 -- name is still used in the encodings.
1374 -- 2. Variants containing a single component named "x" of type "T" may
1375 -- be encoded, as in ordinary C unions, as a single field of the
1376 -- enclosing union type named "x" of type "T", dispensing with the
1377 -- enclosing struct. In this case, of course, the discriminant values
1378 -- corresponding to the variant are unavailable. As for normal
1379 -- variants, the field name "x" may be suffixed with ___XVL if it
1380 -- has dynamic size.
1382 -- For example, the type Var in the preceding section, if followed by
1383 -- "pragma Unchecked_Union (Var);" may be encoded as a struct with two
1384 -- fields. The first is "m". The second field is a union of two types,
1385 -- with field names S1 and "t". As before, S1 is a struct with fields
1386 -- "r" and "s". "t" is a field of type Integer.
1388 ------------------------------------------------
1389 -- Subprograms for Handling Variant Encodings --
1390 ------------------------------------------------
1392 procedure Get_Variant_Encoding (V : Node_Id);
1393 -- This procedure is called by Gigi with V being the variant node. The
1394 -- corresponding encoding string is returned in Name_Buffer with the length
1395 -- of the string in Name_Len, and an ASCII.NUL character stored following
1398 -- WARNING: There is a matching C declaration of this subprogram in fe.h
1400 ---------------------------------
1401 -- Subtypes of Variant Records --
1402 ---------------------------------
1404 -- A subtype of a variant record is represented by a type in which the
1405 -- union field from the base type is replaced by one of the possible
1406 -- values. For example, if we have:
1408 -- type Var (Disc : Boolean := True) is record
1423 -- V3 : Var (False);
1425 -- Here V2, for example, is represented with a subtype whose name is
1426 -- something like TvarS3b, which is a struct with three fields. The first
1427 -- two fields are "disc" and "m" as for the base type, and the third field
1428 -- is S1, which contains the fields "r" and "s".
1430 -- The debugger should simply ignore structs with names of the form
1431 -- corresponding to variants, and consider the fields inside as belonging
1432 -- to the containing record.
1434 -----------------------------------------------
1435 -- Extra renamings for subprogram instances --
1436 -----------------------------------------------
1438 procedure Build_Subprogram_Instance_Renamings
1440 Wrapper : Entity_Id);
1441 -- The debugger has difficulties in recovering the value of actuals of an
1442 -- elementary type, from within the body of a subprogram instantiation.
1443 -- This is because such actuals generate an object declaration that is
1444 -- placed within the wrapper package of the instance, and the entity in
1445 -- these declarations is encoded in a complex way that GDB does not handle
1446 -- well. These new renaming declarations appear within the body of the
1447 -- subprogram, and are redundant from a visibility point of view, but They
1448 -- should have no measurable performance impact, and require no special
1449 -- decoding in the debugger.
1451 -------------------------------------------
1452 -- Character literals in Character Types --
1453 -------------------------------------------
1455 -- Character types are enumeration types at least one of whose enumeration
1456 -- literals is a character literal. Enumeration literals are usually simply
1457 -- represented using their identifier names. If the enumeration literal is
1458 -- a character literal, the name is encoded as described in the following
1461 -- The characters 'a'..'z' and '0'..'9' are represented as Qc, where 'c'
1462 -- stands for the character itself. A name QUhh, where each 'h' is a
1463 -- lower-case hexadecimal digit, stands for a character whose Unicode
1464 -- encoding is hh, and QWhhhh likewise stands for a wide character whose
1465 -- encoding is hhhh. The representation values are encoded as for ordinary
1466 -- enumeration literals (and have no necessary relationship to the values
1467 -- encoded in the names).
1469 -- For example, given the type declaration
1471 -- type x is (A, 'C', 'b');
1473 -- the second enumeration literal would be named QU43 and the value
1474 -- assigned to it would be 1, and the third enumeration literal would be
1475 -- named Qb and the value assigned to it would be 2.
1477 -----------------------------------------------
1478 -- Secondary Dispatch tables of tagged types --
1479 -----------------------------------------------
1481 procedure Get_Secondary_DT_External_Name
1483 Ancestor_Typ : Entity_Id;
1484 Suffix_Index : Int);
1485 -- Set Name_Buffer and Name_Len to the external name of one secondary
1486 -- dispatch table of Typ. If the interface has been inherited from some
1487 -- ancestor then Ancestor_Typ is such node (in this case the secondary DT
1488 -- is needed to handle overridden primitives); if there is no such ancestor
1489 -- then Ancestor_Typ is equal to Typ.
1491 -- Internal rule followed for the generation of the external name:
1493 -- Case 1. If the secondary dispatch has not been inherited from some
1494 -- ancestor of Typ then the external name is composed as
1496 -- External_Name (Typ) + Suffix_Number + 'P'
1498 -- Case 2. if the secondary dispatch table has been inherited from some
1499 -- ancestor then the external name is composed as follows:
1500 -- External_Name (Typ) + '_' + External_Name (Ancestor_Typ)
1501 -- + Suffix_Number + 'P'
1503 -- Note: We have to use the external names (instead of simply their names)
1504 -- to protect the frontend against programs that give the same name to all
1505 -- the interfaces and use the expanded name to reference them. The
1506 -- Suffix_Number is used to differentiate all the secondary dispatch
1507 -- tables of a given type.
1511 -- package Pkg1 is | package Pkg2 is | package Pkg3 is
1512 -- type Typ is | type Typ is | type Typ is
1513 -- interface; | interface; | interface;
1514 -- end Pkg1; | end Pkg; | end Pkg3;
1516 -- with Pkg1, Pkg2, Pkg3;
1517 -- package Case_1 is
1518 -- type Typ is new Pkg1.Typ and Pkg2.Typ and Pkg3.Typ with ...
1522 -- package Case_2 is
1523 -- type Typ is new Case_1.Typ with ...
1526 -- These are the external names generated for Case_1.Typ (note that
1527 -- Pkg1.Typ is associated with the Primary Dispatch Table, because it
1528 -- is the parent of this type, and hence no external name is
1529 -- generated for it).
1530 -- case_1__typ0P (associated with Pkg2.Typ)
1531 -- case_1__typ1P (associated with Pkg3.Typ)
1533 -- These are the external names generated for Case_2.Typ:
1534 -- case_2__typ_case_1__typ0P
1535 -- case_2__typ_case_1__typ1P
1537 ----------------------------
1538 -- Effect of Optimization --
1539 ----------------------------
1541 -- If the program is compiled with optimization on (e.g. -O1 switch
1542 -- specified), then there may be variations in the output from the above
1543 -- specification. In particular, objects may disappear from the output.
1544 -- This includes not only constants and variables that the program declares
1545 -- at the source level, but also the x___L and x___U constants created to
1546 -- describe the lower and upper bounds of subtypes with dynamic bounds.
1547 -- This means for example, that array bounds may disappear if optimization
1548 -- is turned on. The debugger is expected to recognize that these constants
1549 -- are missing and deal as best as it can with the limited information
1552 ---------------------------------
1553 -- GNAT Extensions to DWARF2/3 --
1554 ---------------------------------
1556 -- If the compiler switch "-gdwarf+" is specified, GNAT Vendor extensions
1557 -- to DWARF2/3 are generated, with the following variations from the above
1560 -- Change in the contents of the DW_AT_name attribute
1562 -- The operators are represented in their natural form. (for example,
1563 -- the addition operator is written as "+" instead of "Oadd"). The
1564 -- component separator is "." instead of "__"
1566 -- Introduction of DW_AT_GNAT_encoding, encoded with value 0x2301
1568 -- Any debugging information entry representing a program entity, named
1569 -- or implicit, may have a DW_AT_GNAT_encoding attribute. The value of
1570 -- this attribute is a string representing the suffix internally added
1571 -- by GNAT for various purposes, mainly for representing debug
1572 -- information compatible with other formats. In particular this is
1573 -- useful for IDEs which need to filter out information internal to
1574 -- GNAT from their graphical interfaces.
1576 -- If a debugging information entry has multiple encodings, all of them
1577 -- will be listed in DW_AT_GNAT_encoding using the list separator ':'.
1579 -- Introduction of DW_AT_GNAT_descriptive_type, encoded with value 0x2302
1581 -- Any debugging information entry representing a type may have a
1582 -- DW_AT_GNAT_descriptive_type attribute whose value is a reference,
1583 -- pointing to a debugging information entry representing another type
1584 -- associated to the type.
1586 -- Modification of the contents of the DW_AT_producer string
1588 -- When emitting full GNAT Vendor extensions to DWARF2/3, "-gdwarf+"
1589 -- is appended to the DW_AT_producer string.
1591 -- When emitting only DW_AT_GNAT_descriptive_type, "-gdwarf+-" is
1592 -- appended to the DW_AT_producer string.