1------------------------------------------------------------------------------ 2-- -- 3-- GNAT COMPILER COMPONENTS -- 4-- -- 5-- E X P _ D B U G -- 6-- -- 7-- S p e c -- 8-- -- 9-- Copyright (C) 1996-2021, Free Software Foundation, Inc. -- 10-- -- 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. -- 20-- -- 21-- GNAT was originally developed by the GNAT team at New York University. -- 22-- Extensive contributions were provided by Ada Core Technologies Inc. -- 23-- -- 24------------------------------------------------------------------------------ 25 26-- Expand routines for the generation of special declarations used by the 27-- debugger. In accordance with the DWARF specification, certain type names 28-- may also be encoded to provide additional information to the debugger, but 29-- this practice is being deprecated and some encodings described below are no 30-- longer generated by default (they are marked OBSOLETE). 31 32with Namet; use Namet; 33with Types; use Types; 34with Uintp; use Uintp; 35 36package Exp_Dbug is 37 38 ----------------------------------------------------- 39 -- Encoding and Qualification of Names of Entities -- 40 ----------------------------------------------------- 41 42 -- This section describes how the names of entities are encoded in the 43 -- generated debugging information. 44 45 -- An entity in Ada has a name of the form X.Y.Z ... E where X,Y,Z are the 46 -- enclosing scopes (not including Standard at the start). 47 48 -- The encoding of the name follows this basic qualified naming scheme, 49 -- where the encoding of individual entity names is as described in Namet 50 -- (i.e. in particular names present in the original source are folded to 51 -- all lower case, with upper half and wide characters encoded as described 52 -- in Namet). Upper case letters are used only for entities generated by 53 -- the compiler. 54 55 -- There are two cases, global entities, and local entities. In more formal 56 -- terms, local entities are those which have a dynamic enclosing scope, 57 -- and global entities are at the library level, except that we always 58 -- consider procedures to be global entities, even if they are nested 59 -- (that's because at the debugger level a procedure name refers to the 60 -- code, and the code is indeed a global entity, including the case of 61 -- nested procedures.) In addition, we also consider all types to be global 62 -- entities, even if they are defined within a procedure. 63 64 -- The reason for treating all type names as global entities is that a 65 -- number of our type encodings work by having related type names, and we 66 -- need the full qualification to keep this unique. 67 68 -- For global entities, the encoded name includes all components of the 69 -- fully expanded name (but omitting Standard at the start). For example, 70 -- if a library-level child package P.Q has an embedded package R, and 71 -- there is an entity in this embedded package whose name is S, the encoded 72 -- name will include the components p.q.r.s. 73 74 -- For local entities, the encoded name only includes the components up to 75 -- the enclosing dynamic scope (other than a block). At run time, such a 76 -- dynamic scope is a subprogram, and the debugging formats know about 77 -- local variables of procedures, so it is not necessary to have full 78 -- qualification for such entities. In particular this means that direct 79 -- local variables of a procedure are not qualified. 80 81 -- For Ghost entities, the encoding adds a prefix "___ghost_" to aid the 82 -- detection of leaks of Ignored Ghost entities in the "living" space. 83 -- Ignored Ghost entities and any code associated with them should be 84 -- removed by the compiler in a post-processing pass. As a result, 85 -- object files should not contain any occurrences of this prefix. 86 87 -- As an example of the local name convention, consider a procedure V.W 88 -- with a local variable X, and a nested block Y containing an entity Z. 89 -- The fully qualified names of the entities X and Z are: 90 91 -- V.W.X 92 -- V.W.Y.Z 93 94 -- but since V.W is a subprogram, the encoded names will end up 95 -- encoding only 96 97 -- x 98 -- y.z 99 100 -- The separating dots are translated into double underscores 101 102 ----------------------------- 103 -- Handling of Overloading -- 104 ----------------------------- 105 106 -- The above scheme is incomplete for overloaded subprograms, since 107 -- overloading can legitimately result in case of two entities with 108 -- exactly the same fully qualified names. To distinguish between 109 -- entries in a set of overloaded subprograms, the encoded names are 110 -- serialized by adding the suffix: 111 112 -- __nn (two underscores) 113 114 -- where nn is a serial number (2 for the second overloaded function, 115 -- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no 116 -- suffix implies the first instance). 117 118 -- These names are prefixed by the normal full qualification. So for 119 -- example, the third instance of the subprogram qrs in package yz 120 -- would have the name: 121 122 -- yz__qrs__3 123 124 -- A more subtle case arises with entities declared within overloaded 125 -- subprograms. If we have two overloaded subprograms, and both declare 126 -- an entity xyz, then the fully expanded name of the two xyz's is the 127 -- same. To distinguish these, we add the same __n suffix at the end of 128 -- the inner entity names. 129 130 -- In more complex cases, we can have multiple levels of overloading, 131 -- and we must make sure to distinguish which final declarative region 132 -- we are talking about. For this purpose, we use a more complex suffix 133 -- which has the form: 134 135 -- __nn_nn_nn ... 136 137 -- where the nn values are the homonym numbers as needed for any of the 138 -- qualifying entities, separated by a single underscore. If all the nn 139 -- values are 1, the suffix is omitted, Otherwise the suffix is present 140 -- (including any values of 1). The following example shows how this 141 -- suffixing works. 142 143 -- package body Yz is 144 -- procedure Qrs is -- Name is yz__qrs 145 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv 146 -- begin ... end Qrs; 147 148 -- procedure Qrs (X: Int) is -- Name is yz__qrs__2 149 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1 150 -- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2 151 -- begin ... end Tuv; 152 153 -- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3 154 -- type m is new float; -- Name is yz__qrs__tuv__m__2_3 155 -- begin ... end Tuv; 156 -- begin ... end Qrs; 157 -- end Yz; 158 159 -------------------- 160 -- Operator Names -- 161 -------------------- 162 163 -- The above rules applied to operator names would result in names with 164 -- quotation marks, which are not typically allowed by assemblers and 165 -- linkers, and even if allowed would be odd and hard to deal with. To 166 -- avoid this problem, operator names are encoded as follows: 167 168 -- Oabs abs 169 -- Oand and 170 -- Omod mod 171 -- Onot not 172 -- Oor or 173 -- Orem rem 174 -- Oxor xor 175 -- Oeq = 176 -- One /= 177 -- Olt < 178 -- Ole <= 179 -- Ogt > 180 -- Oge >= 181 -- Oadd + 182 -- Osubtract - 183 -- Oconcat & 184 -- Omultiply * 185 -- Odivide / 186 -- Oexpon ** 187 188 -- These names are prefixed by the normal full qualification, and 189 -- suffixed by the overloading identification. So for example, the 190 -- second operator "=" defined in package Extra.Messages would have 191 -- the name: 192 193 -- extra__messages__Oeq__2 194 195 ---------------------------------- 196 -- Resolving Other Name Clashes -- 197 ---------------------------------- 198 199 -- It might be thought that the above scheme is complete, but in Ada 95, 200 -- full qualification is insufficient to uniquely identify an entity in 201 -- the program, even if it is not an overloaded subprogram. There are 202 -- two possible confusions: 203 204 -- a.b 205 206 -- interpretation 1: entity b in body of package a 207 -- interpretation 2: child procedure b of package a 208 209 -- a.b.c 210 211 -- interpretation 1: entity c in child package a.b 212 -- interpretation 2: entity c in nested package b in body of a 213 214 -- It is perfectly legal in both cases for both interpretations to be 215 -- valid within a single program. This is a bit of a surprise since 216 -- certainly in Ada 83, full qualification was sufficient, but not in 217 -- Ada 95. The result is that the above scheme can result in duplicate 218 -- names. This would not be so bad if the effect were just restricted 219 -- to debugging information, but in fact in both the above cases, it 220 -- is possible for both symbols to be external names, and so we have 221 -- a real problem of name clashes. 222 223 -- To deal with this situation, we provide two additional encoding 224 -- rules for names: 225 226 -- First: all library subprogram names are preceded by the string 227 -- _ada_ (which causes no duplications, since normal Ada names can 228 -- never start with an underscore. This not only solves the first 229 -- case of duplication, but also solves another pragmatic problem 230 -- which is that otherwise Ada procedures can generate names that 231 -- clash with existing system function names. Most notably, we can 232 -- have clashes in the case of procedure Main with the C main that 233 -- in some systems is always present. 234 235 -- Second, for the case where nested packages declared in package 236 -- bodies can cause trouble, we add a suffix which shows which 237 -- entities in the list are body-nested packages, i.e. packages 238 -- whose spec is within a package body. The rules are as follows, 239 -- given a list of names in a qualified name name1.name2.... 240 241 -- If none are body-nested package entities, then there is no suffix 242 243 -- If at least one is a body-nested package entity, then the suffix 244 -- is X followed by a string of b's and n's (b = body-nested package 245 -- entity, n = not a body-nested package). 246 247 -- There is one element in this string for each entity in the encoded 248 -- expanded name except the first (the rules are such that the first 249 -- entity of the encoded expanded name can never be a body-nested' 250 -- package. Trailing n's are omitted, as is the last b (there must 251 -- be at least one b, or we would not be generating a suffix at all). 252 253 -- For example, suppose we have 254 255 -- package x is 256 -- pragma Elaborate_Body; 257 -- m1 : integer; -- #1 258 -- end x; 259 260 -- package body x is 261 -- package y is m2 : integer; end y; -- #2 262 -- package body y is 263 -- package z is r : integer; end z; -- #3 264 -- end; 265 -- m3 : integer; -- #4 266 -- end x; 267 268 -- package x.y is 269 -- pragma Elaborate_Body; 270 -- m2 : integer; -- #5 271 -- end x.y; 272 273 -- package body x.y is 274 -- m3 : integer; -- #6 275 -- procedure j is -- #7 276 -- package k is 277 -- z : integer; -- #8 278 -- end k; 279 -- begin 280 -- null; 281 -- end j; 282 -- end x.y; 283 284 -- procedure x.m3 is begin null; end; -- #9 285 286 -- Then the encodings would be: 287 288 -- #1. x__m1 (no BNPE's in sight) 289 -- #2. x__y__m2X (y is a BNPE) 290 -- #3. x__y__z__rXb (y is a BNPE, so is z) 291 -- #4. x__m3 (no BNPE's in sight) 292 -- #5. x__y__m2 (no BNPE's in sight) 293 -- #6. x__y__m3 (no BNPE's in signt) 294 -- #7. x__y__j (no BNPE's in sight) 295 -- #8. k__z (no BNPE's, only up to procedure) 296 -- #9 _ada_x__m3 (library-level subprogram) 297 298 -- Note that we have instances here of both kind of potential name 299 -- clashes, and the above examples show how the encodings avoid the 300 -- clash as follows: 301 302 -- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library 303 -- level subprogram, so it is preceded by the string _ada_ which acts 304 -- to distinguish it from the package body entity. 305 306 -- Lines #2 and #5 both refer to the entity x.y.m2, but the first 307 -- instance is inside the body-nested package y, so there is an X 308 -- suffix to distinguish it from the child library entity. 309 310 -- Note that enumeration literals never need Xb type suffixes, since 311 -- they are never referenced using global external names. 312 313 --------------------- 314 -- Interface Names -- 315 --------------------- 316 317 -- Note: if an interface name is present, then the external name is 318 -- taken from the specified interface name. Given current limitations of 319 -- the gcc backend, this means that the debugging name is also set to 320 -- the interface name, but conceptually, it would be possible (and 321 -- indeed desirable) to have the debugging information still use the Ada 322 -- name as qualified above, so we still fully qualify the name in the 323 -- front end. 324 325 ------------------------------------- 326 -- Encodings Related to Task Types -- 327 ------------------------------------- 328 329 -- Each task object defined by a single task declaration is associated 330 -- with a prefix that is used to qualify procedures defined in that 331 -- task. Given 332 -- 333 -- package body P is 334 -- task body TaskObj is 335 -- procedure F1 is ... end; 336 -- begin 337 -- B; 338 -- end TaskObj; 339 -- end P; 340 -- 341 -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1. 342 -- The body, B, is contained in a subprogram whose name is 343 -- p__taskobjTKB. 344 345 ------------------------------------------ 346 -- Encodings Related to Protected Types -- 347 ------------------------------------------ 348 349 -- Each protected type has an associated record type, that describes 350 -- the actual layout of the private data. In addition to the private 351 -- components of the type, the Corresponding_Record_Type includes one 352 -- component of type Protection, which is the actual lock structure. 353 -- The run-time size of the protected type is the size of the corres- 354 -- ponding record. 355 356 -- For a protected type prot, the Corresponding_Record_Type is encoded 357 -- as protV. 358 359 -- The operations of a protected type are encoded as follows: each 360 -- operation results in two subprograms, a locking one that is called 361 -- from outside of the object, and a non-locking one that is used for 362 -- calls from other operations on the same object. The locking operation 363 -- simply acquires the lock, and then calls the non-locking version. 364 -- The names of all of these have a prefix constructed from the name of 365 -- the type, and a suffix which is P or N, depending on whether this is 366 -- the protected/non-locking version of the operation. 367 368 -- Operations generated for protected entries follow the same encoding. 369 -- Each entry results in two subprograms: a procedure that holds the 370 -- entry body, and a function that holds the evaluation of the barrier. 371 -- The names of these subprograms include the prefix '_E' or '_B' res- 372 -- pectively. The names also include a numeric suffix to render them 373 -- unique in the presence of overloaded entries. 374 375 -- Given the declaration: 376 377 -- protected type Lock is 378 -- function Get return Integer; 379 -- procedure Set (X: Integer); 380 -- entry Update (Val : Integer); 381 -- private 382 -- Value : Integer := 0; 383 -- end Lock; 384 385 -- the following operations are created: 386 387 -- lock_getN 388 -- lock_getP, 389 390 -- lock_setN 391 -- lock_setP 392 393 -- lock_update_E1s 394 -- lock_udpate_B2s 395 396 -- If the protected type implements at least one interface, the 397 -- following additional operations are created: 398 399 -- lock_get 400 401 -- lock_set 402 403 -- These operations are used to ensure overriding of interface level 404 -- subprograms and proper dispatching on interface class-wide objects. 405 -- The bodies of these operations contain calls to their respective 406 -- protected versions: 407 408 -- function lock_get return Integer is 409 -- begin 410 -- return lock_getP; 411 -- end lock_get; 412 413 -- procedure lock_set (X : Integer) is 414 -- begin 415 -- lock_setP (X); 416 -- end lock_set; 417 418 ---------------------------------------------------- 419 -- Conversion between Entities and External Names -- 420 ---------------------------------------------------- 421 422 procedure Get_External_Name 423 (Entity : Entity_Id; 424 Has_Suffix : Boolean := False; 425 Suffix : String := ""); 426 -- Set Name_Buffer and Name_Len to the external name of the entity. The 427 -- external name is the Interface_Name, if specified, unless the entity 428 -- has an address clause or Has_Suffix is true. 429 -- 430 -- If the Interface is not present, or not used, the external name is the 431 -- concatenation of: 432 -- 433 -- - the string "_ada_", if the entity is a library subprogram, 434 -- - the names of any enclosing scopes, each followed by "__", 435 -- or "X_" if the next entity is a subunit) 436 -- - the name of the entity 437 -- - the string "$" (or "__" if target does not allow "$"), followed 438 -- by homonym suffix, if the entity is an overloaded subprogram 439 -- or is defined within an overloaded subprogram. 440 -- - the string "___" followed by Suffix if Has_Suffix is true. 441 -- 442 -- Note that a call to this procedure has no effect if we are not 443 -- generating code, since the necessary information for computing the 444 -- proper external name is not available in this case. 445 446 -- WARNING: There is a matching C declaration of this subprogram in fe.h 447 448 ------------------------------------- 449 -- Encoding for translation into C -- 450 ------------------------------------- 451 452 -- In Modify_Tree_For_C mode we must add encodings to dismabiguate cases 453 -- where Ada block structure cannot be directly translated. These cases 454 -- are as follows: 455 456 -- a) A loop variable may hide a homonym in an enclosing block 457 -- b) A block-local variable may hide a homonym in an enclosing block 458 459 -- In C these constructs are not scopes and we must distinguish the names 460 -- explicitly. In the first case we create a qualified name with the suffix 461 -- 'L', in the second case with a suffix 'B'. 462 463 -------------------------------------------- 464 -- Subprograms for Handling Qualification -- 465 -------------------------------------------- 466 467 procedure Qualify_Entity_Names (N : Node_Id); 468 -- Given a node N, that represents a block, subprogram body, or package 469 -- body or spec, or protected or task type, sets a fully qualified name 470 -- for the defining entity of given construct, and also sets fully 471 -- qualified names for all enclosed entities of the construct (using 472 -- First_Entity/Next_Entity). Note that the actual modifications of the 473 -- names is postponed till a subsequent call to Qualify_All_Entity_Names. 474 -- Note: this routine does not deal with prepending _ada_ to library 475 -- subprogram names. The reason for this is that we only prepend _ada_ 476 -- to the library entity itself, and not to names built from this name. 477 478 procedure Qualify_All_Entity_Names; 479 -- When Qualify_Entity_Names is called, no actual name changes are made, 480 -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call 481 -- is made to this procedure. The reason for this deferral is that when 482 -- names are changed semantic processing may be affected. By deferring 483 -- the changes till just before gigi is called, we avoid any concerns 484 -- about such effects. Gigi itself does not use the names except for 485 -- output of names for debugging purposes (which is why we are doing 486 -- the name changes in the first place). 487 488 -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are 489 -- useful to remove qualification from a name qualified by the call to 490 -- Qualify_All_Entity_Names. 491 492 -------------------------------- 493 -- Handling of Numeric Values -- 494 -------------------------------- 495 496 -- All numeric values here are encoded as strings of decimal digits. Only 497 -- integer values need to be encoded. A negative value is encoded as the 498 -- corresponding positive value followed by a lower case m for minus to 499 -- indicate that the value is negative (e.g. 2m for -2). 500 501 ------------------------ 502 -- Encapsulated Types -- 503 ------------------------ 504 505 -- In some cases, the compiler may encapsulate a type by wrapping it in a 506 -- record. For example, this is used when a size or alignment specification 507 -- requires a larger type. Consider: 508 509 -- type x is mod 2 ** 64; 510 -- for x'size use 256; 511 512 -- In this case, the compiler generates a record type x___PAD, which has 513 -- a single field whose name is F. This single field is 64-bit long and 514 -- contains the actual value. This kind of padding is used when the logical 515 -- value to be stored is shorter than the object in which it is allocated. 516 517 -- A similar encapsulation is done for some packed array types, in which 518 -- case the record type is x___JM and the field name is OBJECT. This is 519 -- used in the case of a packed array stored using modular representation 520 -- (see the section on representation of packed array objects). In this 521 -- case the wrapping is used to achieve correct positioning of the packed 522 -- array value (left/right justified in its field depending on endianness). 523 524 -- When the debugger sees an object of a type whose name has a suffix of 525 -- ___PAD or ___JM, the type will be a record containing a single field, 526 -- and the name of that field will be all upper case. In this case, it 527 -- should look inside to get the value of the inner field, and neither 528 -- the outer structure name, nor the field name should appear when the 529 -- value is printed. 530 531 -- Similarly, when the debugger sees a record named REP being the type of 532 -- a field inside another record type, it should treat the fields inside 533 -- REP as being part of the outer record (this REP field is only present 534 -- for code generation purposes). The REP record should not appear in the 535 -- values printed by the debugger. 536 537 -------------------- 538 -- Implicit Types -- 539 -------------------- 540 541 -- The compiler creates implicit type names in many situations where a 542 -- type is present semantically, but no specific name is present. For 543 -- example: 544 545 -- S : Integer range M .. N; 546 547 -- Here the subtype of S is not integer, but rather an anonymous subtype 548 -- of Integer. Where possible, the compiler generates names for such 549 -- anonymous types that are related to the type from which the subtype 550 -- is obtained as follows: 551 552 -- T name suffix 553 554 -- where name is the name from which the subtype is obtained, using 555 -- lower case letters and underscores, and suffix starts with an upper 556 -- case letter. For example the name for the above declaration might be: 557 558 -- TintegerS4b 559 560 -- If the debugger is asked to give the type of an entity and the type 561 -- has the form T name suffix, it is probably appropriate to just use 562 -- "name" in the response since this is what is meaningful to the 563 -- programmer. 564 565 ------------------- 566 -- Modular Types -- 567 ------------------- 568 569 -- A type declared 570 571 -- type x is mod N; 572 573 -- is encoded as a subrange of an unsigned base type with lower bound zero 574 -- and upper bound N - 1. Thus we give these types a somewhat nonstandard 575 -- interpretation: the standard interpretation would not, in general, imply 576 -- that arithmetic operations on type x are performed modulo N (especially 577 -- not when N is not a power of 2). 578 579 -------------------------------------- 580 -- Tagged Types and Type Extensions -- 581 -------------------------------------- 582 583 -- A type D derived from a tagged type P has a field named "_parent" of 584 -- type P that contains its inherited fields. The type of this field is 585 -- usually P, but may be a more distant ancestor, if P is a null extension 586 -- of that type. 587 588 -- The type tag of a tagged type is a field named "_tag" of a pointer type. 589 -- If the type is derived from another tagged type, its _tag field is found 590 -- in its _parent field. 591 592 ------------------------------------ 593 -- Type Name Encodings (OBSOLETE) -- 594 ------------------------------------ 595 596 -- In the following typ is the name of the type as normally encoded by the 597 -- debugger rules, i.e. a non-qualified name, all in lower case, with 598 -- standard encoding of upper half and wide characters. 599 600 ----------------------- 601 -- Fixed-Point Types -- 602 ----------------------- 603 604 -- Fixed-point types are encoded using a suffix that indicates the 605 -- delta and small values. The actual type itself is a normal integer 606 -- type. 607 608 -- typ___XF_nn_dd 609 -- typ___XF_nn_dd_nn_dd 610 611 -- The first form is used when small = delta. The value of delta (and 612 -- small) is given by the rational nn/dd, where nn and dd are decimal 613 -- integers. 614 -- 615 -- The second form is used if the small value is different from the 616 -- delta. In this case, the first nn/dd rational value is for delta, 617 -- and the second value is for small. 618 619 -------------------- 620 -- Discrete Types -- 621 -------------------- 622 623 -- Discrete types are coded with a suffix indicating the range in the 624 -- case where one or both of the bounds are discriminants or variable. 625 626 -- Note: at the current time, we also encode compile time known bounds 627 -- if they do not match the natural machine type bounds, but this may 628 -- be removed in the future, since it is redundant for most debugging 629 -- formats. However, we do not ever need XD encoding for enumeration 630 -- base types, since here it is always clear what the bounds are from 631 -- the total number of enumeration literals. 632 633 -- typ___XD 634 -- typ___XDL_lowerbound 635 -- typ___XDU_upperbound 636 -- typ___XDLU_lowerbound__upperbound 637 638 -- If a discrete type is a natural machine type (i.e. its bounds 639 -- correspond in a natural manner to its size), then it is left 640 -- unencoded. The above encoding forms are used when there is a 641 -- constrained range that does not correspond to the size or that 642 -- has discriminant references or other compile time known bounds. 643 644 -- The first form is used if both bounds are dynamic, in which case two 645 -- constant objects are present whose names are typ___L and typ___U in 646 -- the same scope as typ, and the values of these constants indicate 647 -- the bounds. As far as the debugger is concerned, these are simply 648 -- variables that can be accessed like any other variables. In the 649 -- enumeration case, these values correspond to the Enum_Rep values for 650 -- the lower and upper bounds. 651 652 -- The second form is used if the upper bound is dynamic, but the lower 653 -- bound is either constant or depends on a discriminant of the record 654 -- with which the type is associated. The upper bound is stored in a 655 -- constant object of name typ___U as previously described, but the 656 -- lower bound is encoded directly into the name as either a decimal 657 -- integer, or as the discriminant name. 658 659 -- The third form is similarly used if the lower bound is dynamic, but 660 -- the upper bound is compile time known or a discriminant reference, 661 -- in which case the lower bound is stored in a constant object of name 662 -- typ___L, and the upper bound is encoded directly into the name as 663 -- either a decimal integer, or as the discriminant name. 664 665 -- The fourth form is used if both bounds are discriminant references 666 -- or compile time known values, with the encoding first for the lower 667 -- bound, then for the upper bound, as previously described. 668 669 ------------------ 670 -- Biased Types -- 671 ------------------ 672 673 -- Only discrete types can be biased, and the fact that they are biased 674 -- is indicated by a suffix of the form: 675 676 -- typ___XB_lowerbound__upperbound 677 678 -- Here lowerbound and upperbound are decimal integers, with the usual 679 -- (postfix "m") encoding for negative numbers. Biased types are only 680 -- possible where the bounds are compile time known, and the values are 681 -- represented as unsigned offsets from the lower bound given. For 682 -- example: 683 684 -- type Q is range 10 .. 15; 685 -- for Q'size use 3; 686 687 -- The size clause will force values of type Q in memory to be stored 688 -- in biased form (e.g. 11 will be represented by the bit pattern 001). 689 690 ---------------------------------------------- 691 -- Record Types with Variable-Length Fields -- 692 ---------------------------------------------- 693 694 -- The debugging formats do not fully support these types, and indeed 695 -- some formats simply generate no useful information at all for such 696 -- types. In order to provide information for the debugger, gigi creates 697 -- a parallel type in the same scope with one of the names 698 699 -- type___XVE 700 -- type___XVU 701 702 -- The former name is used for a record and the latter for the union 703 -- that is made for a variant record (see below) if that record or union 704 -- has a field of variable size or if the record or union itself has a 705 -- variable size. These encodings suffix any other encodings that that 706 -- might be suffixed to the type name. 707 708 -- The idea here is to provide all the needed information to interpret 709 -- objects of the original type in the form of a "fixed up" type, which 710 -- is representable using the normal debugging information. 711 712 -- There are three cases to be dealt with. First, some fields may have 713 -- variable positions because they appear after variable-length fields. 714 -- To deal with this, we encode *all* the field bit positions of the 715 -- special ___XV type in a non-standard manner. 716 717 -- The idea is to encode not the position, but rather information that 718 -- allows computing the position of a field from the position of the 719 -- previous field. The algorithm for computing the actual positions of 720 -- all fields and the length of the record is as follows. In this 721 -- description, let P represent the current bit position in the record. 722 723 -- 1. Initialize P to 0 724 725 -- 2. For each field in the record: 726 727 -- 2a. If an alignment is given (see below), then round P up, if 728 -- needed, to the next multiple of that alignment. 729 730 -- 2b. If a bit position is given, then increment P by that amount 731 -- (that is, treat it as an offset from the end of the preceding 732 -- record). 733 734 -- 2c. Assign P as the actual position of the field 735 736 -- 2d. Compute the length, L, of the represented field (see below) 737 -- and compute P'=P+L. Unless the field represents a variant part 738 -- (see below and also Variant Record Encoding), set P to P'. 739 740 -- The alignment, if present, is encoded in the field name of the 741 -- record, which has a suffix: 742 743 -- fieldname___XVAnn 744 745 -- where the nn after the XVA indicates the alignment value in storage 746 -- units. This encoding is present only if an alignment is present. 747 748 -- The size of the record described by an XVE-encoded type (in bits) is 749 -- generally the maximum value attained by P' in step 2d above, rounded 750 -- up according to the record's alignment. 751 752 -- Second, the variable-length fields themselves are represented by 753 -- replacing the type by a special access type. The designated type of 754 -- this access type is the original variable-length type, and the fact 755 -- that this field has been transformed in this way is signalled by 756 -- encoding the field name as: 757 758 -- field___XVL 759 760 -- where field is the original field name. If a field is both 761 -- variable-length and also needs an alignment encoding, then the 762 -- encodings are combined using: 763 764 -- field___XVLnn 765 766 -- Note: the reason that we change the type is so that the resulting 767 -- type has no variable-length fields. At least some of the formats used 768 -- for debugging information simply cannot tolerate variable- length 769 -- fields, so the encoded information would get lost. 770 771 -- Third, in the case of a variant record, the special union that 772 -- contains the variants is replaced by a normal C union. In this case, 773 -- the positions are all zero. 774 775 -- Discriminants appear before any variable-length fields that depend on 776 -- them, with one exception. In some cases, a discriminant governing the 777 -- choice of a variant clause may appear in the list of fields of an XVE 778 -- type after the entry for the variant clause itself (this can happen 779 -- in the presence of a representation clause for the record type in the 780 -- source program). However, when this happens, the discriminant's 781 -- position may be determined by first applying the rules described in 782 -- this section, ignoring the variant clause. As a result, discriminants 783 -- can always be located independently of the variable-length fields 784 -- that depend on them. 785 786 -- The size of the ___XVE or ___XVU record or union is set to the 787 -- alignment (in bytes) of the original object so that the debugger 788 -- can calculate the size of the original type. 789 790 -- As an example of this encoding, consider the declarations: 791 792 -- type Q is array (1 .. V1) of Float; -- alignment 4 793 -- type R is array (1 .. V2) of Long_Float; -- alignment 8 794 795 -- type X is record 796 -- A : Character; 797 -- B : Float; 798 -- C : String (1 .. V3); 799 -- D : Float; 800 -- E : Q; 801 -- F : R; 802 -- G : Float; 803 -- end record; 804 805 -- The encoded type looks like: 806 807 -- type anonymousQ is access Q; 808 -- type anonymousR is access R; 809 810 -- type X___XVE is record 811 -- A : Character; -- position contains 0 812 -- B : Float; -- position contains 24 813 -- C___XVL : access String (1 .. V3); -- position contains 0 814 -- D___XVA4 : Float; -- position contains 0 815 -- E___XVL4 : anonymousQ; -- position contains 0 816 -- F___XVL8 : anonymousR; -- position contains 0 817 -- G : Float; -- position contains 0 818 -- end record; 819 820 -- Any bit sizes recorded for fields other than dynamic fields and 821 -- variants are honored as for ordinary records. 822 823 -- Notes: 824 825 -- 1) The B field could also have been encoded by using a position of 826 -- zero and an alignment of 4, but in such a case the coding by position 827 -- is preferred (since it takes up less space). We have used the 828 -- (illegal) notation access xxx as field types in the example above. 829 830 -- 2) The E field does not actually need the alignment indication but 831 -- this may not be detected in this case by the conversion routines. 832 833 -- 3) Our conventions do not cover all XVE-encoded records in which 834 -- some, but not all, fields have representation clauses. Such records 835 -- may, therefore, be displayed incorrectly by debuggers. This situation 836 -- is not common. 837 838 ----------------------- 839 -- Base Record Types -- 840 ----------------------- 841 842 -- Under certain circumstances, debuggers need two descriptions of a 843 -- record type, one that gives the actual details of the base type's 844 -- structure (as described elsewhere in these comments) and one that may 845 -- be used to obtain information about the particular subtype and the 846 -- size of the objects being typed. In such cases the compiler will 847 -- substitute type whose name is typically compiler-generated and 848 -- irrelevant except as a key for obtaining the actual type. 849 850 -- Specifically, if this name is x, then we produce a record type named 851 -- x___XVS consisting of one field. The name of this field is that of 852 -- the actual type being encoded, which we'll call y. The type of this 853 -- single field can be either an arbitrary non-reference type, e.g. an 854 -- integer type, or a reference type; in the latter case, the referenced 855 -- type is also the actual type being encoded y. Both x and y may have 856 -- corresponding ___XVE types. 857 858 -- The size of the objects typed as x should be obtained from the 859 -- structure of x (and x___XVE, if applicable) as for ordinary types 860 -- unless there is a variable named x___XVZ, which, if present, will 861 -- hold the size (in bytes) of x. In this latter case, the size of the 862 -- x___XVS type will not be a constant but a reference to x___XVZ. 863 864 -- The type x will either be a subtype of y (see also Subtypes of 865 -- Variant Records, below) or will contain a single field of type y, 866 -- or no fields at all. The layout, types, and positions of these 867 -- fields will be accurate, if present. (Currently, however, the GDB 868 -- debugger makes no use of x except to determine its size). 869 870 -- Among other uses, XVS types are used to encode unconstrained types. 871 -- For example, given: 872 -- 873 -- subtype Int is INTEGER range 0..10; 874 -- type T1 (N: Int := 0) is record 875 -- F1: String (1 .. N); 876 -- end record; 877 -- type AT1 is array (INTEGER range <>) of T1; 878 -- 879 -- the element type for AT1 might have a type defined as if it had 880 -- been written: 881 -- 882 -- type at1___PAD is record F : T1; end record; 883 -- for at1___PAD'Size use 16 * 8; 884 -- 885 -- and there would also be: 886 -- 887 -- type at1___PAD___XVS is record t1: reft1; end record; 888 -- type t1 is ... 889 -- type reft1 is <reference to t1> 890 -- 891 -- Had the subtype Int been dynamic: 892 -- 893 -- subtype Int is INTEGER range 0 .. M; -- M a variable 894 -- 895 -- Then the compiler would also generate a declaration whose effect 896 -- would be 897 -- 898 -- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term; 899 -- 900 -- Not all unconstrained types are so encoded; the XVS convention may be 901 -- unnecessary for unconstrained types of fixed size. However, this 902 -- encoding is always necessary when a subcomponent type (array 903 -- element's type or record field's type) is an unconstrained record 904 -- type some of whose components depend on discriminant values. 905 906 ----------------- 907 -- Array Types -- 908 ----------------- 909 910 -- Since there is no way for the debugger to obtain the index subtypes 911 -- for an array type, we produce a type that has the name of the array 912 -- type followed by "___XA" and is a record type whose field types are 913 -- the respective types for the bounds (and whose field names are the 914 -- names of these types). 915 916 -- To conserve space, we do not produce this type unless one of the 917 -- index types is either an enumeration type, has a variable lower or 918 -- upper bound or is a biased type. 919 920 -- Given the full encoding of these types (see above description for 921 -- the encoding of discrete types), this means that all necessary 922 -- information for addressing arrays is available. In some debugging 923 -- formats, some or all of the bounds information may be available 924 -- redundantly, particularly in the fixed-point case, but this 925 -- information can in any case be ignored by the debugger. 926 927 ------------------------------------------------- 928 -- Subprograms for Handling Encoded Type Names -- 929 ------------------------------------------------- 930 931 procedure Get_Encoded_Name (E : Entity_Id); 932 -- If the entity is a typename, store the external name of the entity as in 933 -- Get_External_Name, followed by three underscores plus the type encoding 934 -- in Name_Buffer with the length in Name_Len, and an ASCII.NUL character 935 -- stored following the name. Otherwise set Name_Buffer and Name_Len to 936 -- hold the entity name. Note that a call to this procedure has no effect 937 -- if we are not generating code, since the necessary information for 938 -- computing the proper encoded name is not available in this case. 939 940 -- WARNING: There is a matching C declaration of this subprogram in fe.h 941 942 -------------- 943 -- Renaming -- 944 -------------- 945 946 -- Debugging information is generated for exception, object, package, and 947 -- subprogram renaming (generic renamings are not significant, since 948 -- generic templates are not relevant at debugging time). 949 950 -- Consider a renaming declaration of the form 951 952 -- x : typ renames y; 953 954 -- There is one case in which no special debugging information is required, 955 -- namely the case of an object renaming where the back end allocates a 956 -- reference for the renamed variable, and the entity x is this reference. 957 -- The debugger can handle this case without any special processing or 958 -- encoding (it won't know it was a renaming, but that does not matter). 959 960 -- All other cases of renaming generate a dummy variable for an entity 961 -- whose name is of the form: 962 963 -- x___XR_... for an object renaming 964 -- x___XRE_... for an exception renaming 965 -- x___XRP_... for a package renaming 966 967 -- and where the "..." represents a suffix that describes the structure of 968 -- the object name given in the renaming (see details below). 969 970 -- The name is fully qualified in the usual manner, i.e. qualified in the 971 -- same manner as the entity x would be. In the case of a package renaming 972 -- where x is a child unit, the qualification includes the name of the 973 -- parent unit, to disambiguate child units with the same simple name and 974 -- (of necessity) different parents. 975 976 -- Note: subprogram renamings are not encoded at the present time 977 978 -- The suffix of the variable name describing the renamed object is defined 979 -- to use the following encoding: 980 981 -- For the simple entity case, where y is just an entity name, the suffix 982 -- is of the form: 983 984 -- y___XE 985 986 -- i.e. the suffix has a single field, the first part matching the 987 -- name y, followed by a "___" separator, ending with sequence XE. 988 -- The entity name portion is fully qualified in the usual manner. 989 -- This same naming scheme is followed for all forms of encoded 990 -- renamings that rename a simple entity. 991 992 -- For the object renaming case where y is a selected component or an 993 -- indexed component, the variable name is suffixed by additional fields 994 -- that give details of the components. The name starts as above with a 995 -- y___XE name indicating the outer level object entity. Then a series of 996 -- selections and indexing operations can be specified as follows: 997 998 -- Indexed component 999 1000 -- A series of subscript values appear in sequence, the number 1001 -- corresponds to the number of dimensions of the array. The 1002 -- subscripts have one of the following two forms: 1003 1004 -- XSnnn 1005 1006 -- Here nnn is a constant value, encoded as a decimal integer 1007 -- (pos value for enumeration type case). Negative values have 1008 -- a trailing 'm' as usual. 1009 1010 -- XSe 1011 1012 -- Here e is the (unqualified) name of a constant entity in the 1013 -- same scope as the renaming which contains the subscript value. 1014 1015 -- Slice 1016 1017 -- For the slice case, we have two entries. The first is for the 1018 -- lower bound of the slice, and has the form: 1019 1020 -- XLnnn 1021 -- XLe 1022 1023 -- Specifies the lower bound, using exactly the same encoding as 1024 -- for an XS subscript as described above. 1025 1026 -- Then the upper bound appears in the usual XSnnn/XSe form 1027 1028 -- Selected component 1029 1030 -- For a selected component, we have a single entry 1031 1032 -- XRf 1033 1034 -- Here f is the field name for the selection 1035 1036 -- For an explicit dereference (.all), we have a single entry 1037 1038 -- XA 1039 1040 -- As an example, consider the declarations: 1041 1042 -- package p is 1043 -- type q is record 1044 -- m : string (2 .. 5); 1045 -- end record; 1046 -- 1047 -- type r is array (1 .. 10, 1 .. 20) of q; 1048 -- 1049 -- g : r; 1050 -- 1051 -- z : string renames g (1,5).m(2 ..3) 1052 -- end p; 1053 1054 -- The generated variable entity would appear as 1055 1056 -- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type; 1057 -- p__g___XE--------------------outer entity is g 1058 -- XS1-----------------first subscript for g 1059 -- XS5--------------second subscript for g 1060 -- XRm-----------select field m 1061 -- XL2--------lower bound of slice 1062 -- XS3-----upper bound of slice 1063 1064 -- Note that the type of the variable is a special internal type named 1065 -- _renaming_type. This type is an arbitrary type of zero size created 1066 -- in package Standard (see cstand.adb) and is ignored by the debugger. 1067 1068 function Debug_Renaming_Declaration (N : Node_Id) return Node_Id; 1069 -- The argument N is a renaming declaration. The result is a variable 1070 -- declaration as described in the above paragraphs. If N is not a special 1071 -- debug declaration, then Empty is returned. This function also takes care 1072 -- of setting Materialize_Entity on the renamed entity where required. 1073 1074 ------------------------------------------- 1075 -- Packed Array Representation in Memory -- 1076 ------------------------------------------- 1077 1078 -- Packed arrays are represented in tightly packed form, with no extra bits 1079 -- between components. This is true even when the component size is not a 1080 -- factor of the storage unit size, so that as a result it is possible for 1081 -- components to cross storage unit boundaries. 1082 1083 -- The layout in storage is identical, regardless of whether the 1084 -- implementation type is a modular type or an array-of-bytes type. See 1085 -- Exp_Pakd for details of how these implementation types are used, but for 1086 -- the purpose of the debugger, only the starting address of the object in 1087 -- memory is significant. 1088 1089 -- The following example should show clearly how the packing works in 1090 -- the little-endian and big-endian cases: 1091 1092 -- type B is range 0 .. 7; 1093 -- for B'Size use 3; 1094 1095 -- type BA is array (0 .. 5) of B; 1096 -- pragma Pack (BA); 1097 1098 -- BV : constant BA := (1,2,3,4,5,6); 1099 1100 -- Little endian case 1101 1102 -- BV'Address + 2 BV'Address + 1 BV'Address + 0 1103 -- +-----------------+-----------------+-----------------+ 1104 -- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 | 1105 -- +-----------------+-----------------+-----------------+ 1106 -- <---------> <-----> <---> <---> <-----> <---> <---> 1107 -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0) 1108 -- 1109 -- Big endian case 1110 -- 1111 -- BV'Address + 0 BV'Address + 1 BV'Address + 2 1112 -- +-----------------+-----------------+-----------------+ 1113 -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? | 1114 -- +-----------------+-----------------+-----------------+ 1115 -- <---> <---> <-----> <---> <---> <-----> <---------> 1116 -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits 1117 1118 -- Note that if a modular type is used to represent the array, the 1119 -- allocation in memory is not the same as a normal modular type. The 1120 -- difference occurs when the allocated object is larger than the size of 1121 -- the array. For a normal modular type, we extend the value on the left 1122 -- with zeroes. 1123 1124 -- For example, in the normal modular case, if we have a 6-bit modular 1125 -- type, declared as mod 2**6, and we allocate an 8-bit object for this 1126 -- type, then we extend the value with two bits on the most significant 1127 -- end, and in either the little-endian or big-endian case, the value 63 1128 -- is represented as 00111111 in binary in memory. 1129 1130 -- For a modular type used to represent a packed array, the rule is 1131 -- different. In this case, if we have to extend the value, then we do it 1132 -- with undefined bits (which are not initialized and whose value is 1133 -- irrelevant to any generated code). Furthermore these bits are on the 1134 -- right (least significant bits) in the big-endian case, and on the left 1135 -- (most significant bits) in the little-endian case. 1136 1137 -- For example, if we have a packed boolean array of 6 bits, all set to 1138 -- True, stored in an 8-bit object, then the value in memory in binary is 1139 -- ??111111 in the little-endian case, and 111111?? in the big-endian case. 1140 1141 -- This is done so that the representation of packed arrays does not 1142 -- depend on whether we use a modular representation or array of bytes 1143 -- as previously described. This ensures that we can pass such values by 1144 -- reference in the case where a subprogram has to be able to handle values 1145 -- stored in either form. 1146 1147 -- Note that when we extract the value of such a modular packed array, we 1148 -- expect to retrieve only the relevant bits, so in this same example, when 1149 -- we extract the value we get 111111 in both cases, and the code generated 1150 -- by the front end assumes this although it does not assume that any high 1151 -- order bits are defined. 1152 1153 -- There are opportunities for optimization based on the knowledge that the 1154 -- unused bits are irrelevant for these type of packed arrays. For example 1155 -- if we have two such 6-bit-in-8-bit values and we do an assignment: 1156 1157 -- a := b; 1158 1159 -- Then logically, we extract the 6 bits and store only 6 bits in the 1160 -- result, but the back end is free to simply assign the entire 8-bits in 1161 -- this case, since we don't actually care about the undefined bits. 1162 -- However, in the equality case, it is important to ensure that the 1163 -- undefined bits do not participate in an equality test. 1164 1165 -- If a modular packed array value is assigned to a register then logically 1166 -- it could always be held right justified, to avoid any need to shift, 1167 -- e.g. when doing comparisons. But probably this is a bad choice, as it 1168 -- would mean that an assignment such as a := above would require shifts 1169 -- when one value is in a register and the other value is in memory. 1170 1171 ------------------------------------------- 1172 -- Packed Array Name Encoding (OBSOLETE) -- 1173 ------------------------------------------- 1174 1175 -- For every constrained packed array, two types are created, and both 1176 -- appear in the debugging output: 1177 1178 -- The original declared array type is a perfectly normal array type, and 1179 -- its index bounds indicate the original bounds of the array. 1180 1181 -- The corresponding packed array type, which may be a modular type, or 1182 -- may be an array of bytes type (see Exp_Pakd for full details). This is 1183 -- the type that is actually used in the generated code and for debugging 1184 -- information for all objects of the packed type. 1185 1186 -- The name of the corresponding packed array type is: 1187 1188 -- ttt___XPnnn 1189 1190 -- where 1191 1192 -- ttt is the name of the original declared array 1193 -- nnn is the component size in bits (1-31) 1194 1195 -- Note that if the packed array is not bit-packed, the name will simply 1196 -- be tttP. 1197 1198 -- When the debugger sees that an object is of a type that is encoded in 1199 -- this manner, it can use the original type to determine the bounds and 1200 -- the component type, and the component size to determine the packing 1201 -- details. 1202 1203 -- For an unconstrained packed array, the corresponding packed array type 1204 -- is neither used in the generated code nor for debugging information, 1205 -- only the original type is used. In order to convey the packing in the 1206 -- debugging information, the compiler generates the associated fat- and 1207 -- thin-pointer types (see the Pointers to Unconstrained Array section 1208 -- below) using the name of the corresponding packed array type as the 1209 -- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively. 1210 1211 -- When the debugger sees that an object is of a type that is encoded in 1212 -- this manner, it can use the type of the fields to determine the bounds 1213 -- and the component type, and the component size to determine the packing 1214 -- details. 1215 1216 ------------------------------------------------------ 1217 -- Subprograms for Handling Packed Array Type Names -- 1218 ------------------------------------------------------ 1219 1220 function Make_Packed_Array_Impl_Type_Name 1221 (Typ : Entity_Id; 1222 Csize : Uint) return Name_Id; 1223 -- This function is used in Exp_Pakd to create the name that is encoded as 1224 -- described above. The entity Typ provides the name ttt, and the value 1225 -- Csize is the component size that provides the nnn value. 1226 1227 -------------------------------------- 1228 -- Pointers to Unconstrained Arrays -- 1229 -------------------------------------- 1230 1231 -- There are two kinds of pointer to unconstrained arrays. The debugger can 1232 -- tell which format is in use by the form of the type of the pointer. 1233 1234 -- Fat Pointers 1235 1236 -- Fat pointers are represented as a structure with two fields. This 1237 -- structure has two distinguished field names: 1238 1239 -- P_ARRAY is a pointer to the array type. The name of this type is 1240 -- the unconstrained type followed by "___XUA". The bounds of this 1241 -- array will be obtained through dereferences of P_BOUNDS below. 1242 1243 -- P_BOUNDS is a pointer to a structure. The name of this type is 1244 -- the unconstrained array name followed by "___XUB" and it has 1245 -- fields of the form: 1246 1247 -- LBn (n a decimal integer) lower bound of n'th dimension 1248 -- UBn (n a decimal integer) upper bound of n'th dimension 1249 1250 -- The bounds may be of any integral type. In the case of enumeration 1251 -- types, Enum_Rep values are used. 1252 1253 -- For a given unconstrained array type, the compiler will generate a 1254 -- fat pointer type whose name is the name of the array type, and use 1255 -- it to represent the array type itself in the debugging information. 1256 1257 -- This name was historically followed by "___XUP" (OBSOLETE). 1258 1259 -- For each pointer to this unconstrained array type, the compiler will 1260 -- generate a typedef that points to the above fat pointer type. As a 1261 -- consequence, when it comes to fat pointer types: 1262 1263 -- 1. The type name is given by the typedef, if any 1264 1265 -- 2. If the debugger is asked to output the type, the appropriate 1266 -- form is "access arr" if there is the typedef, otherwise it is 1267 -- the array definition. 1268 1269 -- Thin Pointers 1270 1271 -- The value of a thin pointer is a pointer to the second field of a 1272 -- structure with two fields. The first field of the structure is of 1273 -- the type ___XUB described for fat pointer types above. The second 1274 -- field of the structure contains the actual array. 1275 1276 -- Thin pointers are represented as a regular pointer to array in the 1277 -- debugging information. The bounds of this array will be the contents 1278 -- of the first field above obtained through (shifted) dereferences. 1279 1280 -- Thin Pointers (OBSOLETE) 1281 1282 -- The value of a thin pointer is a pointer to the second field of a 1283 -- structure with two fields. The name of this structure's type is 1284 -- "arr___XUT", where "arr" is the name of the unconstrained array 1285 -- type. Even though it points into the middle of this structure, 1286 -- the type in the debugging information is pointer to structure. 1287 1288 -- The first field of the structure is named BOUNDS and is of the type 1289 -- ___XUB described for fat pointer types above. 1290 1291 -- The second field of the structure is named ARRAY, and contains the 1292 -- actual array. Because this array has a dynamic size, determined by 1293 -- the BOUNDS field that precedes it, all of the information about 1294 -- arr___XUT is encoded in a parallel type named arr___XUT___XVE, with 1295 -- fields BOUNDS and ARRAY___XVL. As for previously described ___XVE 1296 -- types, ARRAY___XVL has a pointer-to-array type. However, the array 1297 -- type in this case is named arr___XUA and only its element type is 1298 -- meaningful, just as described for fat pointers. 1299 1300 ----------------------------- 1301 -- Variant Record Encoding -- 1302 ----------------------------- 1303 1304 -- The variant part of a variant record is encoded as a single field in the 1305 -- enclosing record, whose name is: 1306 1307 -- discrim___XVN 1308 1309 -- where discrim is the unqualified name of the variant. This field name is 1310 -- built by gigi (not by code in this unit). For Unchecked_Union record, 1311 -- this discriminant will not appear in the record (see Unchecked Unions, 1312 -- below). 1313 1314 -- The type corresponding to this field has a name that is obtained by 1315 -- concatenating the type name with the above string and is similar to a C 1316 -- union, in which each member of the union corresponds to one variant. 1317 -- However, unlike a C union, the size of the type may be variable even if 1318 -- each of the components are fixed size, since it includes a computation 1319 -- of which variant is present. 1320 1321 -- The name of the union member is encoded to indicate the choices, and 1322 -- is a string given by the following grammar: 1323 1324 -- member_name ::= {choice} | others_choice 1325 -- choice ::= simple_choice | range_choice 1326 -- simple_choice ::= S number 1327 -- range_choice ::= R number T number 1328 -- number ::= {decimal_digit} [m] 1329 -- others_choice ::= O (upper case letter O) 1330 1331 -- The m in a number indicates a negative value. As an example of this 1332 -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by 1333 1334 -- R1T4S7S10m 1335 1336 -- In the case of enumeration values, the values used are the actual 1337 -- representation values in the case where an enumeration type has an 1338 -- enumeration representation spec (i.e. they are values that correspond 1339 -- to the use of the Enum_Rep attribute). 1340 1341 -- The type of the inner record is given by the name of the union type (as 1342 -- above) concatenated with the above string. 1343 1344 -- As an example, consider: 1345 1346 -- type Var (Disc : Boolean := True) is record 1347 -- M : Integer; 1348 1349 -- case Disc is 1350 -- when True => 1351 -- R : Integer; 1352 -- S : Integer; 1353 1354 -- when False => 1355 -- T : Integer; 1356 -- end case; 1357 -- end record; 1358 1359 -- V1 : Var; 1360 1361 -- In this case, the type var is represented as a struct with three fields. 1362 -- The first two are "disc" and "m", representing the values of these 1363 -- record components. The third field is a union of two types, with field 1364 -- names S1 and O. S1 is a struct with fields "r" and "s", and O is a 1365 -- struct with field "t". 1366 1367 ---------------------- 1368 -- Unchecked Unions -- 1369 ---------------------- 1370 1371 -- The encoding for variant records changes somewhat under the influence 1372 -- of a "pragma Unchecked_Union" clause: 1373 1374 -- 1. The discriminant will not be present in the record, although its 1375 -- name is still used in the encodings. 1376 -- 2. Variants containing a single component named "x" of type "T" may 1377 -- be encoded, as in ordinary C unions, as a single field of the 1378 -- enclosing union type named "x" of type "T", dispensing with the 1379 -- enclosing struct. In this case, of course, the discriminant values 1380 -- corresponding to the variant are unavailable. 1381 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. 1387 1388 ------------------------------------------------ 1389 -- Subprograms for Handling Variant Encodings -- 1390 ------------------------------------------------ 1391 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 1396 -- the name. 1397 1398 -- WARNING: There is a matching C declaration of this subprogram in fe.h 1399 1400 --------------------------------- 1401 -- Subtypes of Variant Records -- 1402 --------------------------------- 1403 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: 1407 1408 -- type Var (Disc : Boolean := True) is record 1409 -- M : Integer; 1410 1411 -- case Disc is 1412 -- when True => 1413 -- R : Integer; 1414 -- S : Integer; 1415 1416 -- when False => 1417 -- T : Integer; 1418 -- end case; 1419 1420 -- end record; 1421 -- V1 : Var; 1422 -- V2 : Var (True); 1423 -- V3 : Var (False); 1424 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". 1429 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. 1433 1434 ----------------------------------------------- 1435 -- Extra renamings for subprogram instances -- 1436 ----------------------------------------------- 1437 1438 procedure Build_Subprogram_Instance_Renamings 1439 (N : Node_Id; 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. 1450 1451 ------------------------------------------- 1452 -- Character literals in Character Types -- 1453 ------------------------------------------- 1454 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 1459 -- paragraph. 1460 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). 1468 1469 -- For example, given the type declaration 1470 1471 -- type x is (A, 'C', 'b'); 1472 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. 1476 1477 ----------------------------------------------- 1478 -- Secondary Dispatch tables of tagged types -- 1479 ----------------------------------------------- 1480 1481 procedure Get_Secondary_DT_External_Name 1482 (Typ : Entity_Id; 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. 1490 -- 1491 -- Internal rule followed for the generation of the external name: 1492 -- 1493 -- Case 1. If the secondary dispatch has not been inherited from some 1494 -- ancestor of Typ then the external name is composed as 1495 -- follows: 1496 -- External_Name (Typ) + Suffix_Number + 'P' 1497 -- 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' 1502 -- 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. 1508 -- 1509 -- Examples: 1510 -- 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; 1515 -- 1516 -- with Pkg1, Pkg2, Pkg3; 1517 -- package Case_1 is 1518 -- type Typ is new Pkg1.Typ and Pkg2.Typ and Pkg3.Typ with ... 1519 -- end Case_1; 1520 -- 1521 -- with Case_1; 1522 -- package Case_2 is 1523 -- type Typ is new Case_1.Typ with ... 1524 -- end Case_2; 1525 -- 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) 1532 -- 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 1536 1537 ---------------------------- 1538 -- Effect of Optimization -- 1539 ---------------------------- 1540 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 1550 -- available. 1551 1552 ----------------------------------------- 1553 -- GNAT Extensions to DWARF (OBSOLETE) -- 1554 ----------------------------------------- 1555 1556 -- DW_AT_use_GNAT_descriptive_type, encoded with value 0x2301 1557 1558 -- This extension has never been implemented in the compiler. 1559 1560 -- DW_AT_GNAT_descriptive_type, encoded with value 0x2302 1561 1562 -- Any debugging information entry representing a type may have a 1563 -- DW_AT_GNAT_descriptive_type attribute whose value is a reference, 1564 -- pointing to a debugging information entry representing another type 1565 -- associated to the type. 1566 1567end Exp_Dbug; 1568