1 ------------------------------------------------------------------------------
3 -- GNAT COMPILER COMPONENTS --
9 -- Copyright (C) 1996-2004 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 2, 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 COPYING. If not, write --
19 -- to the Free Software Foundation, 59 Temple Place - Suite 330, Boston, --
20 -- MA 02111-1307, USA. --
22 -- GNAT was originally developed by the GNAT team at New York University. --
23 -- Extensive contributions were provided by Ada Core Technologies Inc. --
25 ------------------------------------------------------------------------------
27 -- Expand routines for generation of special declarations used by the
28 -- debugger. In accordance with the Dwarf 2.2 specification, certain
29 -- type names are encoded to provide information to the debugger.
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
41 -- the generated debugging information.
43 -- An entity in Ada has a name of the form X.Y.Z ... E where X,Y,Z
44 -- are the 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
48 -- Namet (i.e. in particular names present in the original source are
49 -- folded to all lower case, with upper half and wide characters encoded
50 -- as described in Namet). Upper case letters are used only for entities
51 -- generated by the compiler.
53 -- There are two cases, global entities, and local entities. In more
54 -- formal terms, local entities are those which have a dynamic enclosing
55 -- scope, and global entities are at the library level, except that we
56 -- always consider procedures to be global entities, even if they are
57 -- nested (that's because at the debugger level a procedure name refers
58 -- to the code, and the code is indeed a global entity, including the
59 -- case of nested procedures.) In addition, we also consider all types
60 -- to be global entities, even if they are defined within a procedure.
62 -- The reason for treating all type names as global entities is that
63 -- a number of our type encodings work by having related type names,
64 -- and we 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 embdded 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
73 -- up to the enclosing dynamic scope (other than a block). At run time,
74 -- such a dynamic scope is a subprogram, and the debugging formats know
75 -- about local variables of procedures, so it is not necessary to have
76 -- full qualification for such entities. In particular this means that
77 -- direct local variables of a procedure are not qualified.
79 -- As an example of the local name convention, consider a procedure V.W
80 -- with a local variable X, and a nested block Y containing an entity
81 -- Z. The fully qualified names of the entities X and Z are:
86 -- but since V.W is a subprogram, the encoded names will end up
92 -- The separating dots are translated into double underscores.
94 -----------------------------
95 -- Handling of Overloading --
96 -----------------------------
98 -- The above scheme is incomplete with respect to overloaded
99 -- subprograms, since overloading can legitimately result in a
100 -- case of two entities with exactly the same fully qualified names.
101 -- To distinguish between entries in a set of overloaded subprograms,
102 -- the encoded names are serialized by adding the suffix:
104 -- __nn (two underscores)
106 -- where nn is a serial number (2 for the second overloaded function,
107 -- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no
108 -- suffix implies the first instance).
110 -- These names are prefixed by the normal full qualification. So
111 -- for example, the third instance of the subprogram qrs in package
112 -- yz would have the name:
116 -- A more subtle case arises with entities declared within overloaded
117 -- subprograms. If we have two overloaded subprograms, and both declare
118 -- an entity xyz, then the fully expanded name of the two xyz's is the
119 -- same. To distinguish these, we add the same __n suffix at the end of
120 -- the inner entity names.
122 -- In more complex cases, we can have multiple levels of overloading,
123 -- and we must make sure to distinguish which final declarative region
124 -- we are talking about. For this purpose, we use a more complex suffix
125 -- which has the form:
129 -- where the nn values are the homonym numbers as needed for any of
130 -- the qualifying entities, separated by a single underscore. If all
131 -- the nn values are 1, the suffix is omitted, Otherwise the suffix
132 -- is present (including any values of 1). The following example
133 -- shows how this suffixing works.
135 -- package body Yz is
136 -- procedure Qrs is -- Name is yz__qrs
137 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv
138 -- begin ... end Qrs;
140 -- procedure Qrs (X: Int) is -- Name is yz__qrs__2
141 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1
142 -- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2
143 -- begin ... end Tuv;
145 -- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3
146 -- type m is new float; -- Name is yz__qrs__tuv__m__2_3
147 -- begin ... end Tuv;
148 -- begin ... end Qrs;
155 -- The above rules applied to operator names would result in names
156 -- with quotation marks, which are not typically allowed by assemblers
157 -- and linkers, and even if allowed would be odd and hard to deal with.
158 -- To avoid this problem, operator names are encoded as follows:
180 -- These names are prefixed by the normal full qualification, and
181 -- suffixed by the overloading identification. So for example, the
182 -- second operator "=" defined in package Extra.Messages would
185 -- extra__messages__Oeq__2
187 ----------------------------------
188 -- Resolving Other Name Clashes --
189 ----------------------------------
191 -- It might be thought that the above scheme is complete, but in Ada 95,
192 -- full qualification is insufficient to uniquely identify an entity
193 -- in the program, even if it is not an overloaded subprogram. There
194 -- are two possible confusions:
198 -- interpretation 1: entity b in body of package a
199 -- interpretation 2: child procedure b of package a
203 -- interpretation 1: entity c in child package a.b
204 -- interpretation 2: entity c in nested package b in body of a
206 -- It is perfectly legal in both cases for both interpretations to
207 -- be valid within a single program. This is a bit of a surprise since
208 -- certainly in Ada 83, full qualification was sufficient, but not in
209 -- Ada 95. The result is that the above scheme can result in duplicate
210 -- names. This would not be so bad if the effect were just restricted
211 -- to debugging information, but in fact in both the above cases, it
212 -- is possible for both symbols to be external names, and so we have
213 -- a real problem of name clashes.
215 -- To deal with this situation, we provide two additional encoding
218 -- First: all library subprogram names are preceded by the string
219 -- _ada_ (which causes no duplications, since normal Ada names can
220 -- never start with an underscore. This not only solves the first
221 -- case of duplication, but also solves another pragmatic problem
222 -- which is that otherwise Ada procedures can generate names that
223 -- clash with existing system function names. Most notably, we can
224 -- have clashes in the case of procedure Main with the C main that
225 -- in some systems is always present.
227 -- Second, for the case where nested packages declared in package
228 -- bodies can cause trouble, we add a suffix which shows which
229 -- entities in the list are body-nested packages, i.e. packages
230 -- whose spec is within a package body. The rules are as follows,
231 -- given a list of names in a qualified name name1.name2....
233 -- If none are body-nested package entities, then there is no suffix
235 -- If at least one is a body-nested package entity, then the suffix
236 -- is X followed by a string of b's and n's (b = body-nested package
237 -- entity, n = not a body-nested package).
239 -- There is one element in this string for each entity in the encoded
240 -- expanded name except the first (the rules are such that the first
241 -- entity of the encoded expanded name can never be a body-nested'
242 -- package. Trailing n's are omitted, as is the last b (there must
243 -- be at least one b, or we would not be generating a suffix at all).
245 -- For example, suppose we have
248 -- pragma Elaborate_Body;
249 -- m1 : integer; -- #1
253 -- package y is m2 : integer; end y; -- #2
255 -- package z is r : integer; end z; -- #3
257 -- m3 : integer; -- #4
261 -- pragma Elaborate_Body;
262 -- m2 : integer; -- #5
265 -- package body x.y is
266 -- m3 : integer; -- #6
267 -- procedure j is -- #7
269 -- z : integer; -- #8
276 -- procedure x.m3 is begin null; end; -- #9
278 -- Then the encodings would be:
280 -- #1. x__m1 (no BNPE's in sight)
281 -- #2. x__y__m2X (y is a BNPE)
282 -- #3. x__y__z__rXb (y is a BNPE, so is z)
283 -- #4. x__m3 (no BNPE's in sight)
284 -- #5. x__y__m2 (no BNPE's in sight)
285 -- #6. x__y__m3 (no BNPE's in signt)
286 -- #7. x__y__j (no BNPE's in sight)
287 -- #8. k__z (no BNPE's, only up to procedure)
288 -- #9 _ada_x__m3 (library level subprogram)
290 -- Note that we have instances here of both kind of potential name
291 -- clashes, and the above examples show how the encodings avoid the
294 -- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library
295 -- level subprogram, so it is preceded by the string _ada_ which acts
296 -- to distinguish it from the package body entity.
298 -- Lines #2 and #5 both refer to the entity x.y.m2, but the first
299 -- instance is inside the body-nested package y, so there is an X
300 -- suffix to distinguish it from the child library entity.
302 -- Note that enumeration literals never need Xb type suffixes, since
303 -- they are never referenced using global external names.
305 ---------------------
306 -- Interface Names --
307 ---------------------
309 -- Note: if an interface name is present, then the external name
310 -- is taken from the specified interface name. Given the current
311 -- limitations of the gcc backend, this means that the debugging
312 -- name is also set to the interface name, but conceptually, it
313 -- would be possible (and indeed desirable) to have the debugging
314 -- information still use the Ada name as qualified above, so we
315 -- still fully qualify the name in the front end.
317 -------------------------------------
318 -- Encodings Related to Task Types --
319 -------------------------------------
321 -- Each task object defined by a single task declaration is associated
322 -- with a prefix that is used to qualify procedures defined in that
326 -- task body TaskObj is
327 -- procedure F1 is ... end;
333 -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1,
334 -- The body, B, is contained in a subprogram whose name is
337 ------------------------------------------
338 -- Encodings Related to Protected Types --
339 ------------------------------------------
341 -- Each protected type has an associated record type, that describes
342 -- the actual layout of the private data. In addition to the private
343 -- components of the type, the Corresponding_Record_Type includes one
344 -- component of type Protection, which is the actual lock structure.
345 -- The run-time size of the protected type is the size of the corres-
348 -- For a protected type prot, the Corresponding_Record_Type is encoded
351 -- The operations of a protected type are encoded as follows: each
352 -- operation results in two subprograms, a locking one that is called
353 -- from outside of the object, and a non-locking one that is used for
354 -- calls from other operations on the same object. The locking operation
355 -- simply acquires the lock, and then calls the non-locking version.
356 -- The names of all of these have a prefix constructed from the name of
357 -- the type, the string "PT", and a suffix which is P or N, depending on
358 -- whether this is the protected/non-locking version of the operation.
360 -- Operations generated for protected entries follow the same encoding.
361 -- Each entry results in two suprograms: a procedure that holds the
362 -- entry body, and a function that holds the evaluation of the barrier.
363 -- The names of these subprograms include the prefix 'E' or 'B' res-
364 -- pectively. The names also include a numeric suffix to render them
365 -- unique in the presence of overloaded entries.
367 -- Given the declaration:
369 -- protected type Lock is
370 -- function Get return Integer;
371 -- procedure Set (X: Integer);
372 -- entry Update (Val : Integer);
374 -- Value : Integer := 0;
377 -- the following operations are created:
388 ----------------------------------------------------
389 -- Conversion between Entities and External Names --
390 ----------------------------------------------------
392 No_Dollar_In_Label : constant Boolean := True;
393 -- True iff the target does not allow dollar signs ("$") in external names
394 -- ??? We want to migrate all platforms to use the same convention.
395 -- As a first step, we force this constant to always be True. This
396 -- constant will eventually be deleted after we have verified that
397 -- the migration does not cause any unforseen adverse impact.
398 -- We chose "__" because it is supported on all platforms, which is
399 -- not the case of "$".
401 procedure Get_External_Name
403 Has_Suffix : Boolean);
404 -- Set Name_Buffer and Name_Len to the external name of entity E.
405 -- The external name is the Interface_Name, if specified, unless
406 -- the entity has an address clause or a suffix.
408 -- If the Interface is not present, or not used, the external name
409 -- is the concatenation of:
411 -- - the string "_ada_", if the entity is a library subprogram,
412 -- - the names of any enclosing scopes, each followed by "__",
413 -- or "X_" if the next entity is a subunit)
414 -- - the name of the entity
415 -- - the string "$" (or "__" if target does not allow "$"), followed
416 -- by homonym suffix, if the entity is an overloaded subprogram
417 -- or is defined within an overloaded subprogram.
419 procedure Get_External_Name_With_Suffix
422 -- Set Name_Buffer and Name_Len to the external name of entity E.
423 -- If Suffix is the empty string the external name is as above,
424 -- otherwise the external name is the concatenation of:
426 -- - the string "_ada_", if the entity is a library subprogram,
427 -- - the names of any enclosing scopes, each followed by "__",
428 -- or "X_" if the next entity is a subunit)
429 -- - the name of the entity
430 -- - the string "$" (or "__" if target does not allow "$"), followed
431 -- by homonym suffix, if the entity is an overloaded subprogram
432 -- or is defined within an overloaded subprogram.
433 -- - the string "___" followed by Suffix
435 -- If this procedure is called in the ASIS mode, it does nothing. See the
436 -- comments in the body for more details.
438 --------------------------------------------
439 -- Subprograms for Handling Qualification --
440 --------------------------------------------
442 procedure Qualify_Entity_Names (N : Node_Id);
443 -- Given a node N, that represents a block, subprogram body, or package
444 -- body or spec, or protected or task type, sets a fully qualified name
445 -- for the defining entity of given construct, and also sets fully
446 -- qualified names for all enclosed entities of the construct (using
447 -- First_Entity/Next_Entity). Note that the actual modifications of the
448 -- names is postponed till a subsequent call to Qualify_All_Entity_Names.
449 -- Note: this routine does not deal with prepending _ada_ to library
450 -- subprogram names. The reason for this is that we only prepend _ada_
451 -- to the library entity itself, and not to names built from this name.
453 procedure Qualify_All_Entity_Names;
454 -- When Qualify_Entity_Names is called, no actual name changes are made,
455 -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call
456 -- is made to this procedure. The reason for this deferral is that when
457 -- names are changed semantic processing may be affected. By deferring
458 -- the changes till just before gigi is called, we avoid any concerns
459 -- about such effects. Gigi itself does not use the names except for
460 -- output of names for debugging purposes (which is why we are doing
461 -- the name changes in the first place.
463 -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet
464 -- are useful to remove qualification from a name qualified by the
465 -- call to Qualify_All_Entity_Names.
467 --------------------------------
468 -- Handling of Numeric Values --
469 --------------------------------
471 -- All numeric values here are encoded as strings of decimal digits.
472 -- Only integer values need to be encoded. A negative value is encoded
473 -- as the corresponding positive value followed by a lower case m for
474 -- minus to indicate that the value is negative (e.g. 2m for -2).
476 -------------------------
477 -- Type Name Encodings --
478 -------------------------
480 -- In the following typ is the name of the type as normally encoded by
481 -- the debugger rules, i.e. a non-qualified name, all in lower case,
482 -- with standard encoding of upper half and wide characters
484 ------------------------
485 -- Encapsulated Types --
486 ------------------------
488 -- In some cases, the compiler encapsulates a type by wrapping it in
489 -- a structure. For example, this is used when a size or alignment
490 -- specification requires a larger type. Consider:
492 -- type y is mod 2 ** 64;
493 -- for y'size use 256;
495 -- In this case the compile generates a structure type y___PAD, which
496 -- has a single field whose name is F. This single field is 64 bits
497 -- long and contains the actual value.
499 -- A similar encapsulation is done for some packed array types,
500 -- in which case the structure type is y___LJM and the field name
503 -- When the debugger sees an object of a type whose name has a
504 -- suffix not otherwise mentioned in this specification, the type
505 -- is a record containing a single field, and the name of that field
506 -- is all upper-case letters, it should look inside to get the value
507 -- of the field, and neither the outer structure name, nor the
508 -- field name should appear when the value is printed.
510 -----------------------
511 -- Fixed-Point Types --
512 -----------------------
514 -- Fixed-point types are encoded using a suffix that indicates the
515 -- delta and small values. The actual type itself is a normal
519 -- typ___XF_nn_dd_nn_dd
521 -- The first form is used when small = delta. The value of delta (and
522 -- small) is given by the rational nn/dd, where nn and dd are decimal
525 -- The second form is used if the small value is different from the
526 -- delta. In this case, the first nn/dd rational value is for delta,
527 -- and the second value is for small.
529 ------------------------------
530 -- VAX Floating-Point Types --
531 ------------------------------
533 -- Vax floating-point types are represented at run time as integer
534 -- types, which are treated specially by the code generator. Their
535 -- type names are encoded with the following suffix:
541 -- representing the Vax F Float, D Float, and G Float types. The
542 -- debugger must treat these specially. In particular, printing
543 -- these values can be achieved using the debug procedures that
544 -- are provided in package System.Vax_Float_Operations:
546 -- procedure Debug_Output_D (Arg : D);
547 -- procedure Debug_Output_F (Arg : F);
548 -- procedure Debug_Output_G (Arg : G);
550 -- These three procedures take a Vax floating-point argument, and
551 -- output a corresponding decimal representation to standard output
552 -- with no terminating line return.
558 -- Discrete types are coded with a suffix indicating the range in
559 -- the case where one or both of the bounds are discriminants or
562 -- Note: at the current time, we also encode compile time known
563 -- bounds if they do not match the natural machine type bounds,
564 -- but this may be removed in the future, since it is redundant
565 -- for most debugging formats. However, we do not ever need XD
566 -- encoding for enumeration base types, since here it is always
567 -- clear what the bounds are from the total number of enumeration
568 -- literals, and of course we do not need to encode the dummy XR
569 -- types generated for renamings.
572 -- typ___XDL_lowerbound
573 -- typ___XDU_upperbound
574 -- typ___XDLU_lowerbound__upperbound
576 -- If a discrete type is a natural machine type (i.e. its bounds
577 -- correspond in a natural manner to its size), then it is left
578 -- unencoded. The above encoding forms are used when there is a
579 -- constrained range that does not correspond to the size or that
580 -- has discriminant references or other compile time known bounds.
582 -- The first form is used if both bounds are dynamic, in which case
583 -- two constant objects are present whose names are typ___L and
584 -- typ___U in the same scope as typ, and the values of these constants
585 -- indicate the bounds. As far as the debugger is concerned, these
586 -- are simply variables that can be accessed like any other variables.
587 -- In the enumeration case, these values correspond to the Enum_Rep
588 -- values for the lower and upper bounds.
590 -- The second form is used if the upper bound is dynamic, but the
591 -- lower bound is either constant or depends on a discriminant of
592 -- the record with which the type is associated. The upper bound
593 -- is stored in a constant object of name typ___U as previously
594 -- described, but the lower bound is encoded directly into the
595 -- name as either a decimal integer, or as the discriminant name.
597 -- The third form is similarly used if the lower bound is dynamic,
598 -- but the upper bound is compile time known or a discriminant
599 -- reference, in which case the lower bound is stored in a constant
600 -- object of name typ___L, and the upper bound is encoded directly
601 -- into the name as either a decimal integer, or as the discriminant
604 -- The fourth form is used if both bounds are discriminant references
605 -- or compile time known values, with the encoding first for the lower
606 -- bound, then for the upper bound, as previously described.
616 -- Is encoded as a subrange of an unsigned base type with lower bound
617 -- 0 and upper bound N. That is, there is no name encoding. We use
618 -- the standard encodings provided by the debugging format. Thus
619 -- we give these types a non-standard interpretation: the standard
620 -- interpretation of our encoding would not, in general, imply that
621 -- arithmetic on type x was to be performed modulo N (especially not
622 -- when N is not a power of 2).
628 -- Only discrete types can be biased, and the fact that they are
629 -- biased is indicated by a suffix of the form:
631 -- typ___XB_lowerbound__upperbound
633 -- Here lowerbound and upperbound are decimal integers, with the
634 -- usual (postfix "m") encoding for negative numbers. Biased
635 -- types are only possible where the bounds are compile time
636 -- known, and the values are represented as unsigned offsets
637 -- from the lower bound given. For example:
639 -- type Q is range 10 .. 15;
642 -- The size clause will force values of type Q in memory to be
643 -- stored in biased form (e.g. 11 will be represented by the
646 ----------------------------------------------
647 -- Record Types with Variable-Length Fields --
648 ----------------------------------------------
650 -- The debugging formats do not fully support these types, and indeed
651 -- some formats simply generate no useful information at all for such
652 -- types. In order to provide information for the debugger, gigi creates
653 -- a parallel type in the same scope with one of the names
658 -- The former name is used for a record and the latter for the union
659 -- that is made for a variant record (see below) if that record or
660 -- union has a field of variable size or if the record or union itself
661 -- has a variable size. These encodings suffix any other encodings that
662 -- that might be suffixed to the type name.
664 -- The idea here is to provide all the needed information to interpret
665 -- objects of the original type in the form of a "fixed up" type, which
666 -- is representable using the normal debugging information.
668 -- There are three cases to be dealt with. First, some fields may have
669 -- variable positions because they appear after variable-length fields.
670 -- To deal with this, we encode *all* the field bit positions of the
671 -- special ___XV type in a non-standard manner.
673 -- The idea is to encode not the position, but rather information
674 -- that allows computing the position of a field from the position
675 -- of the previous field. The algorithm for computing the actual
676 -- positions of all fields and the length of the record is as
677 -- follows. In this description, let P represent the current
678 -- bit position in the record.
680 -- 1. Initialize P to 0.
682 -- 2. For each field in the record,
684 -- 2a. If an alignment is given (see below), then round P
685 -- up, if needed, to the next multiple of that alignment.
687 -- 2b. If a bit position is given, then increment P by that
688 -- amount (that is, treat it as an offset from the end of the
689 -- preceding record).
691 -- 2c. Assign P as the actual position of the field.
693 -- 2d. Compute the length, L, of the represented field (see below)
694 -- and compute P'=P+L. Unless the field represents a variant part
695 -- (see below and also Variant Record Encoding), set P to P'.
697 -- The alignment, if present, is encoded in the field name of the
698 -- record, which has a suffix:
702 -- where the nn after the XVA indicates the alignment value in storage
703 -- units. This encoding is present only if an alignment is present.
705 -- The size of the record described by an XVE-encoded type (in bits)
706 -- is generally the maximum value attained by P' in step 2d above,
707 -- rounded up according to the record's alignment.
709 -- Second, the variable-length fields themselves are represented by
710 -- replacing the type by a special access type. The designated type
711 -- of this access type is the original variable-length type, and the
712 -- fact that this field has been transformed in this way is signalled
713 -- by encoding the field name as:
717 -- where field is the original field name. If a field is both
718 -- variable-length and also needs an alignment encoding, then the
719 -- encodings are combined using:
723 -- Note: the reason that we change the type is so that the resulting
724 -- type has no variable-length fields. At least some of the formats
725 -- used for debugging information simply cannot tolerate variable-
726 -- length fields, so the encoded information would get lost.
728 -- Third, in the case of a variant record, the special union
729 -- that contains the variants is replaced by a normal C union.
730 -- In this case, the positions are all zero.
732 -- Discriminants appear before any variable-length fields that depend
733 -- on them, with one exception. In some cases, a discriminant
734 -- governing the choice of a variant clause may appear in the list
735 -- of fields of an XVE type after the entry for the variant clause
736 -- itself (this can happen in the presence of a representation clause
737 -- for the record type in the source program). However, when this
738 -- happens, the discriminant's position may be determined by first
739 -- applying the rules described in this section, ignoring the variant
740 -- clause. As a result, discriminants can always be located
741 -- independently of the variable-length fields that depend on them.
743 -- The size of the ___XVE or ___XVU record or union is set to the
744 -- alignment (in bytes) of the original object so that the debugger
745 -- can calculate the size of the original type.
747 -- As an example of this encoding, consider the declarations:
749 -- type Q is array (1 .. V1) of Float; -- alignment 4
750 -- type R is array (1 .. V2) of Long_Float; -- alignment 8
755 -- C : String (1 .. V3);
762 -- The encoded type looks like:
764 -- type anonymousQ is access Q;
765 -- type anonymousR is access R;
767 -- type X___XVE is record
768 -- A : Character; -- position contains 0
769 -- B : Float; -- position contains 24
770 -- C___XVL : access String (1 .. V3); -- position contains 0
771 -- D___XVA4 : Float; -- position contains 0
772 -- E___XVL4 : anonymousQ; -- position contains 0
773 -- F___XVL8 : anonymousR; -- position contains 0
774 -- G : Float; -- position contains 0
777 -- Any bit sizes recorded for fields other than dynamic fields and
778 -- variants are honored as for ordinary records.
782 -- 1) The B field could also have been encoded by using a position
783 -- of zero, and an alignment of 4, but in such a case, the coding by
784 -- position is preferred (since it takes up less space). We have used
785 -- the (illegal) notation access xxx as field types in the example
788 -- 2) The E field does not actually need the alignment indication
789 -- but this may not be detected in this case by the conversion
792 -- 3) Our conventions do not cover all XVE-encoded records in which
793 -- some, but not all, fields have representation clauses. Such
794 -- records may, therefore, be displayed incorrectly by debuggers.
795 -- This situation is not common.
797 -----------------------
798 -- Base Record Types --
799 -----------------------
801 -- Under certain circumstances, debuggers need two descriptions
802 -- of a record type, one that gives the actual details of the
803 -- base type's structure (as described elsewhere in these
804 -- comments) and one that may be used to obtain information
805 -- about the particular subtype and the size of the objects
806 -- being typed. In such cases the compiler will substitute a
807 -- type whose name is typically compiler-generated and
808 -- irrelevant except as a key for obtaining the actual type.
809 -- Specifically, if this name is x, then we produce a record
810 -- type named x___XVS consisting of one field. The name of
811 -- this field is that of the actual type being encoded, which
812 -- we'll call y (the type of this single field is arbitrary).
813 -- Both x and y may have corresponding ___XVE types.
815 -- The size of the objects typed as x should be obtained from
816 -- the structure of x (and x___XVE, if applicable) as for
817 -- ordinary types unless there is a variable named x___XVZ, which,
818 -- if present, will hold the the size (in bits) of x.
820 -- The type x will either be a subtype of y (see also Subtypes
821 -- of Variant Records, below) or will contain no fields at
822 -- all. The layout, types, and positions of these fields will
823 -- be accurate, if present. (Currently, however, the GDB
824 -- debugger makes no use of x except to determine its size).
826 -- Among other uses, XVS types are sometimes used to encode
827 -- unconstrained types. For example, given
829 -- subtype Int is INTEGER range 0..10;
830 -- type T1 (N: Int := 0) is record
831 -- F1: String (1 .. N);
833 -- type AT1 is array (INTEGER range <>) of T1;
835 -- the element type for AT1 might have a type defined as if it had
838 -- type at1___C_PAD is record null; end record;
839 -- for at1___C_PAD'Size use 16 * 8;
841 -- and there would also be
843 -- type at1___C_PAD___XVS is record t1: Integer; end record;
846 -- Had the subtype Int been dynamic:
848 -- subtype Int is INTEGER range 0 .. M; -- M a variable
850 -- Then the compiler would also generate a declaration whose effect
853 -- at1___C_PAD___XVZ: constant Integer := 32 + M * 8 + padding term;
855 -- Not all unconstrained types are so encoded; the XVS
856 -- convention may be unnecessary for unconstrained types of
857 -- fixed size. However, this encoding is always necessary when
858 -- a subcomponent type (array element's type or record field's
859 -- type) is an unconstrained record type some of whose
860 -- components depend on discriminant values.
866 -- Since there is no way for the debugger to obtain the index subtypes
867 -- for an array type, we produce a type that has the name of the
868 -- array type followed by "___XA" and is a record whose field names
869 -- are the names of the types for the bounds. The types of these
870 -- fields is an integer type which is meaningless.
872 -- To conserve space, we do not produce this type unless one of
873 -- the index types is either an enumeration type, has a variable
874 -- upper bound, has a lower bound different from the constant 1,
875 -- is a biased type, or is wider than "sizetype".
877 -- Given the full encoding of these types (see above description for
878 -- the encoding of discrete types), this means that all necessary
879 -- information for addressing arrays is available. In some
880 -- debugging formats, some or all of the bounds information may
881 -- be available redundantly, particularly in the fixed-point case,
882 -- but this information can in any case be ignored by the debugger.
884 ----------------------------
885 -- Note on Implicit Types --
886 ----------------------------
888 -- The compiler creates implicit type names in many situations where
889 -- a type is present semantically, but no specific name is present.
892 -- S : Integer range M .. N;
894 -- Here the subtype of S is not integer, but rather an anonymous
895 -- subtype of Integer. Where possible, the compiler generates names
896 -- for such anonymous types that are related to the type from which
897 -- the subtype is obtained as follows:
901 -- where name is the name from which the subtype is obtained, using
902 -- lower case letters and underscores, and suffix starts with an upper
903 -- case letter. For example, the name for the above declaration of S
908 -- If the debugger is asked to give the type of an entity and the type
909 -- has the form T name suffix, it is probably appropriate to just use
910 -- "name" in the response since this is what is meaningful to the
913 -------------------------------------------------
914 -- Subprograms for Handling Encoded Type Names --
915 -------------------------------------------------
917 procedure Get_Encoded_Name (E : Entity_Id);
918 -- If the entity is a typename, store the external name of
919 -- the entity as in Get_External_Name, followed by three underscores
920 -- plus the type encoding in Name_Buffer with the length in Name_Len,
921 -- and an ASCII.NUL character stored following the name.
922 -- Otherwise set Name_Buffer and Name_Len to hold the entity name.
928 -- Debugging information is generated for exception, object, package,
929 -- and subprogram renaming (generic renamings are not significant, since
930 -- generic templates are not relevant at debugging time).
932 -- Consider a renaming declaration of the form
936 -- There is one case in which no special debugging information is required,
937 -- namely the case of an object renaming where the backend allocates a
938 -- reference for the renamed variable, and the entity x is this reference.
939 -- The debugger can handle this case without any special processing or
940 -- encoding (it won't know it was a renaming, but that does not matter).
942 -- All other cases of renaming generate a dummy type definition for
943 -- an entity whose name is:
945 -- x___XR for an object renaming
946 -- x___XRE for an exception renaming
947 -- x___XRP for a package renaming
949 -- The name is fully qualified in the usual manner, i.e. qualified in
950 -- the same manner as the entity x would be. In the case of a package
951 -- renaming where x is a child unit, the qualification includes the
952 -- name of the parent unit, to disambiguate child units with the same
953 -- simple name and (of necessity) different parents.
955 -- Note: subprogram renamings are not encoded at the present time.
957 -- The type is an enumeration type with a single enumeration literal
958 -- that is an identifier which describes the renamed variable.
960 -- For the simple entity case, where y is an entity name,
961 -- the enumeration is of the form:
965 -- i.e. the enumeration type has a single field, whose name
966 -- matches the name y, with the XE suffix. The entity for this
967 -- enumeration literal is fully qualified in the usual manner.
968 -- All subprogram, exception, and package renamings fall into
969 -- this category, as well as simple object renamings.
971 -- For the object renaming case where y is a selected component or an
972 -- indexed component, the literal name is suffixed by additional fields
973 -- that give details of the components. The name starts as above with
974 -- a y___XE entity indicating the outer level variable. Then a series
975 -- of selections and indexing operations can be specified as follows:
979 -- A series of subscript values appear in sequence, the number
980 -- corresponds to the number of dimensions of the array. The
981 -- subscripts have one of the following two forms:
985 -- Here nnn is a constant value, encoded as a decimal
986 -- integer (pos value for enumeration type case). Negative
987 -- values have a trailing 'm' as usual.
991 -- Here e is the (unqualified) name of a constant entity in
992 -- the same scope as the renaming which contains the subscript
997 -- For the slice case, we have two entries. The first is for
998 -- the lower bound of the slice, and has the form
1003 -- Specifies the lower bound, using exactly the same encoding
1004 -- as for an XS subscript as described above.
1006 -- Then the upper bound appears in the usual XSnnn/XSe form
1008 -- Selected component
1010 -- For a selected component, we have a single entry
1014 -- Here f is the field name for the selection
1016 -- For an explicit deference (.all), we have a single entry
1020 -- As an example, consider the declarations:
1024 -- m : string (2 .. 5);
1027 -- type r is array (1 .. 10, 1 .. 20) of q;
1031 -- z : string renames g (1,5).m(2 ..3)
1034 -- The generated type definition would appear as
1036 -- type p__z___XR is
1037 -- (p__g___XEXS1XS5XRmXL2XS3);
1038 -- p__g___XE--------------------outer entity is g
1039 -- XS1-----------------first subscript for g
1040 -- XS5--------------second subscript for g
1041 -- XRm-----------select field m
1042 -- XL2--------lower bound of slice
1043 -- XS3-----upper bound of slice
1045 function Debug_Renaming_Declaration (N : Node_Id) return Node_Id;
1046 -- The argument N is a renaming declaration. The result is a type
1047 -- declaration as described in the above paragraphs. If not special
1048 -- debug declaration, than Empty is returned.
1050 ---------------------------
1051 -- Packed Array Encoding --
1052 ---------------------------
1054 -- For every packed array, two types are created, and both appear in
1055 -- the debugging output.
1057 -- The original declared array type is a perfectly normal array type,
1058 -- and its index bounds indicate the original bounds of the array.
1060 -- The corresponding packed array type, which may be a modular type, or
1061 -- may be an array of bytes type (see Exp_Pakd for full details). This
1062 -- is the type that is actually used in the generated code and for
1063 -- debugging information for all objects of the packed type.
1065 -- The name of the corresponding packed array type is:
1070 -- ttt is the name of the original declared array
1071 -- nnn is the component size in bits (1-31)
1073 -- When the debugger sees that an object is of a type that is encoded
1074 -- in this manner, it can use the original type to determine the bounds,
1075 -- and the component size to determine the packing details.
1077 -- Packed arrays are represented in tightly packed form, with no extra
1078 -- bits between components. This is true even when the component size
1079 -- is not a factor of the storage unit size, so that as a result it is
1080 -- possible for components to cross storage unit boundaries.
1082 -- The layout in storage is identical, regardless of whether the
1083 -- implementation type is a modular type or an array-of-bytes type.
1084 -- See Exp_Pakd for details of how these implementation types are used,
1085 -- but for the purpose of the debugger, only the starting address of
1086 -- the object in memory is significant.
1088 -- The following example should show clearly how the packing works in
1089 -- the little-endian and big-endian cases:
1091 -- type B is range 0 .. 7;
1092 -- for B'Size use 3;
1094 -- type BA is array (0 .. 5) of B;
1095 -- pragma Pack (BA);
1097 -- BV : constant BA := (1,2,3,4,5,6);
1099 -- Little endian case
1101 -- BV'Address + 2 BV'Address + 1 BV'Address + 0
1102 -- +-----------------+-----------------+-----------------+
1103 -- | 0 0 0 0 0 0 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 |
1104 -- +-----------------+-----------------+-----------------+
1105 -- <---------> <-----> <---> <---> <-----> <---> <--->
1106 -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0)
1110 -- BV'Address + 0 BV'Address + 1 BV'Address + 2
1111 -- +-----------------+-----------------+-----------------+
1112 -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 0 0 0 0 0 0 |
1113 -- +-----------------+-----------------+-----------------+
1114 -- <---> <---> <-----> <---> <---> <-----> <--------->
1115 -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits
1117 ------------------------------------------------------
1118 -- Subprograms for Handling Packed Array Type Names --
1119 ------------------------------------------------------
1121 function Make_Packed_Array_Type_Name
1125 -- This function is used in Exp_Pakd to create the name that is encoded
1126 -- as described above. The entity Typ provides the name ttt, and the
1127 -- value Csize is the component size that provides the nnn value.
1129 --------------------------------------
1130 -- Pointers to Unconstrained Arrays --
1131 --------------------------------------
1133 -- There are two kinds of pointers to arrays. The debugger can tell
1134 -- which format is in use by the form of the type of the pointer.
1138 -- Fat pointers are represented as a struct with two fields. This
1139 -- struct has two distinguished field names:
1141 -- P_ARRAY is a pointer to the array type. The name of this
1142 -- type is the unconstrained type followed by "___XUA". This
1143 -- array will have bounds which are the discriminants, and
1144 -- hence are unparsable, but will give the number of
1145 -- subscripts and the component type.
1147 -- P_BOUNDS is a pointer to a struct, the name of whose type is the
1148 -- unconstrained array name followed by "___XUB" and which has
1149 -- fields of the form
1151 -- LBn (n a decimal integer) lower bound of n'th dimension
1152 -- UBn (n a decimal integer) upper bound of n'th dimension
1154 -- The bounds may be any integral type. In the case of an
1155 -- enumeration type, Enum_Rep values are used.
1157 -- The debugging information will sometimes reference an anonymous
1158 -- fat pointer type. Such types are given the name xxx___XUP, where
1159 -- xxx is the name of the designated type. If the debugger is asked
1160 -- to output such a type name, the appropriate form is "access xxx".
1164 -- The value of a thin pointer is a pointer to the second field
1165 -- of a structure with two fields. The name of this structure's
1166 -- type is "arr___XUT", where "arr" is the name of the
1167 -- unconstrained array type. Even though it actually points into
1168 -- middle of this structure, the thin pointer's type in debugging
1169 -- information is pointer-to-arr___XUT.
1171 -- The first field of arr___XUT is named BOUNDS, and has a type
1172 -- named arr___XUB, with the structure described for such types
1173 -- in fat pointers, as described above.
1175 -- The second field of arr___XUT is named ARRAY, and contains
1176 -- the actual array. Because this array has a dynamic size,
1177 -- determined by the BOUNDS field that precedes it, all of the
1178 -- information about arr___XUT is encoded in a parallel type named
1179 -- arr___XUT___XVE, with fields BOUNDS and ARRAY___XVL. As for
1180 -- previously described ___XVE types, ARRAY___XVL has
1181 -- a pointer-to-array type. However, the array type in this case
1182 -- is named arr___XUA and only its element type is meaningful,
1183 -- just as described for fat pointers.
1185 --------------------------------------
1186 -- Tagged Types and Type Extensions --
1187 --------------------------------------
1189 -- A type C derived from a tagged type P has a field named "_parent"
1190 -- of type P that contains its inherited fields. The type of this
1191 -- field is usually P (encoded as usual if it has a dynamic size),
1192 -- but may be a more distant ancestor, if P is a null extension of
1195 -- The type tag of a tagged type is a field named _tag, of type void*.
1196 -- If the type is derived from another tagged type, its _tag field is
1197 -- found in its _parent field.
1199 -----------------------------
1200 -- Variant Record Encoding --
1201 -----------------------------
1203 -- The variant part of a variant record is encoded as a single field
1204 -- in the enclosing record, whose name is:
1208 -- where discrim is the unqualified name of the variant. This field name
1209 -- is built by gigi (not by code in this unit). In the case of an
1210 -- Unchecked_Union record, this discriminant will not appear in the
1211 -- record, and the debugger must proceed accordingly (basically it
1212 -- can treat this case as it would a C union).
1214 -- The type corresponding to this field has a name that is obtained
1215 -- by concatenating the type name with the above string and is similar
1216 -- to a C union, in which each member of the union corresponds to one
1217 -- variant. However, unlike a C union, the size of the type may be
1218 -- variable even if each of the components are fixed size, since it
1219 -- includes a computation of which variant is present. In that case,
1220 -- it will be encoded as above and a type with the suffix "___XVN___XVU"
1223 -- The name of the union member is encoded to indicate the choices, and
1224 -- is a string given by the following grammar:
1226 -- union_name ::= {choice} | others_choice
1227 -- choice ::= simple_choice | range_choice
1228 -- simple_choice ::= S number
1229 -- range_choice ::= R number T number
1230 -- number ::= {decimal_digit} [m]
1231 -- others_choice ::= O (upper case letter O)
1233 -- The m in a number indicates a negative value. As an example of this
1234 -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by
1238 -- In the case of enumeration values, the values used are the
1239 -- actual representation values in the case where an enumeration type
1240 -- has an enumeration representation spec (i.e. they are values that
1241 -- correspond to the use of the Enum_Rep attribute).
1243 -- The type of the inner record is given by the name of the union
1244 -- type (as above) concatenated with the above string. Since that
1245 -- type may itself be variable-sized, it may also be encoded as above
1246 -- with a new type with a further suffix of "___XVU".
1248 -- As an example, consider:
1250 -- type Var (Disc : Boolean := True) is record
1265 -- In this case, the type var is represented as a struct with three
1266 -- fields, the first two are "disc" and "m", representing the values
1267 -- of these record components.
1269 -- The third field is a union of two types, with field names S1 and O.
1270 -- S1 is a struct with fields "r" and "s", and O is a struct with
1273 ------------------------------------------------
1274 -- Subprograms for Handling Variant Encodings --
1275 ------------------------------------------------
1277 procedure Get_Variant_Encoding (V : Node_Id);
1278 -- This procedure is called by Gigi with V being the variant node.
1279 -- The corresponding encoding string is returned in Name_Buffer with
1280 -- the length of the string in Name_Len, and an ASCII.NUL character
1281 -- stored following the name.
1283 ---------------------------------
1284 -- Subtypes of Variant Records --
1285 ---------------------------------
1287 -- A subtype of a variant record is represented by a type in which the
1288 -- union field from the base type is replaced by one of the possible
1289 -- values. For example, if we have:
1291 -- type Var (Disc : Boolean := True) is record
1306 -- V3 : Var (False);
1308 -- Here V2 for example is represented with a subtype whose name is
1309 -- something like TvarS3b, which is a struct with three fields. The
1310 -- first two fields are "disc" and "m" as for the base type, and
1311 -- the third field is S1, which contains the fields "r" and "s".
1313 -- The debugger should simply ignore structs with names of the form
1314 -- corresponding to variants, and consider the fields inside as
1315 -- belonging to the containing record.
1317 -------------------------------------------
1318 -- Character literals in Character Types --
1319 -------------------------------------------
1321 -- Character types are enumeration types at least one of whose
1322 -- enumeration literals is a character literal. Enumeration literals
1323 -- are usually simply represented using their identifier names. In
1324 -- the case where an enumeration literal is a character literal, the
1325 -- name aencoded as described in the following paragraph.
1327 -- A name QUhh, where each 'h' is a lower-case hexadecimal digit,
1328 -- stands for a character whose Unicode encoding is hh, and
1329 -- QWhhhh likewise stands for a wide character whose encoding
1330 -- is hhhh. The representation values are encoded as for ordinary
1331 -- enumeration literals (and have no necessary relationship to the
1332 -- values encoded in the names).
1334 -- For example, given the type declaration
1336 -- type x is (A, 'C', B);
1338 -- the second enumeration literal would be named QU43 and the
1339 -- value assigned to it would be 1.
1341 ----------------------------
1342 -- Effect of Optimization --
1343 ----------------------------
1345 -- If the program is compiled with optimization on (e.g. -O1 switch
1346 -- specified), then there may be variations in the output from the
1347 -- above specification. In particular, objects may disappear from
1348 -- the output. This includes not only constants and variables that
1349 -- the program declares at the source level, but also the x___L and
1350 -- x___U constants created to describe the lower and upper bounds of
1351 -- subtypes with dynamic bounds. This means for example, that array
1352 -- bounds may disappear if optimization is turned on. The debugger
1353 -- is expected to recognize that these constants are missing and
1354 -- deal as best as it can with the limited information available.