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) 1996-2019, 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 generation of special declarations used by the 27-- debugger. In accordance with the Dwarf 2.2 specification, certain 28-- type names are encoded to provide information to the debugger. 29 30with Namet; use Namet; 31with Types; use Types; 32with Uintp; use Uintp; 33 34package Exp_Dbug is 35 36 ----------------------------------------------------- 37 -- Encoding and Qualification of Names of Entities -- 38 ----------------------------------------------------- 39 40 -- This section describes how the names of entities are encoded in the 41 -- generated debugging information. 42 43 -- An entity in Ada has a name of the form X.Y.Z ... E where X,Y,Z are the 44 -- enclosing scopes (not including Standard at the start). 45 46 -- The encoding of the name follows this basic qualified naming scheme, 47 -- where the encoding of individual entity names is as described in Namet 48 -- (i.e. in particular names present in the original source are folded to 49 -- all lower case, with upper half and wide characters encoded as described 50 -- in Namet). Upper case letters are used only for entities generated by 51 -- the compiler. 52 53 -- There are two cases, global entities, and local entities. In more formal 54 -- terms, local entities are those which have a dynamic enclosing scope, 55 -- and global entities are at the library level, except that we always 56 -- consider procedures to be global entities, even if they are nested 57 -- (that's because at the debugger level a procedure name refers to the 58 -- code, and the code is indeed a global entity, including the case of 59 -- nested procedures.) In addition, we also consider all types to be global 60 -- entities, even if they are defined within a procedure. 61 62 -- The reason for treating all type names as global entities is that a 63 -- number of our type encodings work by having related type names, and we 64 -- need the full qualification to keep this unique. 65 66 -- For global entities, the encoded name includes all components of the 67 -- fully expanded name (but omitting Standard at the start). For example, 68 -- if a library-level child package P.Q has an embedded package R, and 69 -- there is an entity in this embedded package whose name is S, the encoded 70 -- name will include the components p.q.r.s. 71 72 -- For local entities, the encoded name only includes the components up to 73 -- the enclosing dynamic scope (other than a block). At run time, such a 74 -- dynamic scope is a subprogram, and the debugging formats know about 75 -- local variables of procedures, so it is not necessary to have full 76 -- qualification for such entities. In particular this means that direct 77 -- local variables of a procedure are not qualified. 78 79 -- For Ghost entities, the encoding adds a prefix "___ghost_" to aid the 80 -- detection of leaks of Ignored Ghost entities in the "living" space. 81 -- Ignored Ghost entities and any code associated with them should be 82 -- removed by the compiler in a post-processing pass. As a result, 83 -- object files should not contain any occurrences of this prefix. 84 85 -- As an example of the local name convention, consider a procedure V.W 86 -- with a local variable X, and a nested block Y containing an entity Z. 87 -- The fully qualified names of the entities X and Z are: 88 89 -- V.W.X 90 -- V.W.Y.Z 91 92 -- but since V.W is a subprogram, the encoded names will end up 93 -- encoding only 94 95 -- x 96 -- y.z 97 98 -- The separating dots are translated into double underscores 99 100 ----------------------------- 101 -- Handling of Overloading -- 102 ----------------------------- 103 104 -- The above scheme is incomplete for overloaded subprograms, since 105 -- overloading can legitimately result in case of two entities with 106 -- exactly the same fully qualified names. To distinguish between 107 -- entries in a set of overloaded subprograms, the encoded names are 108 -- serialized by adding the suffix: 109 110 -- __nn (two underscores) 111 112 -- where nn is a serial number (2 for the second overloaded function, 113 -- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no 114 -- suffix implies the first instance). 115 116 -- These names are prefixed by the normal full qualification. So for 117 -- example, the third instance of the subprogram qrs in package yz 118 -- would have the name: 119 120 -- yz__qrs__3 121 122 -- A more subtle case arises with entities declared within overloaded 123 -- subprograms. If we have two overloaded subprograms, and both declare 124 -- an entity xyz, then the fully expanded name of the two xyz's is the 125 -- same. To distinguish these, we add the same __n suffix at the end of 126 -- the inner entity names. 127 128 -- In more complex cases, we can have multiple levels of overloading, 129 -- and we must make sure to distinguish which final declarative region 130 -- we are talking about. For this purpose, we use a more complex suffix 131 -- which has the form: 132 133 -- __nn_nn_nn ... 134 135 -- where the nn values are the homonym numbers as needed for any of the 136 -- qualifying entities, separated by a single underscore. If all the nn 137 -- values are 1, the suffix is omitted, Otherwise the suffix is present 138 -- (including any values of 1). The following example shows how this 139 -- suffixing works. 140 141 -- package body Yz is 142 -- procedure Qrs is -- Name is yz__qrs 143 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv 144 -- begin ... end Qrs; 145 146 -- procedure Qrs (X: Int) is -- Name is yz__qrs__2 147 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1 148 -- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2 149 -- begin ... end Tuv; 150 151 -- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3 152 -- type m is new float; -- Name is yz__qrs__tuv__m__2_3 153 -- begin ... end Tuv; 154 -- begin ... end Qrs; 155 -- end Yz; 156 157 -------------------- 158 -- Operator Names -- 159 -------------------- 160 161 -- The above rules applied to operator names would result in names with 162 -- quotation marks, which are not typically allowed by assemblers and 163 -- linkers, and even if allowed would be odd and hard to deal with. To 164 -- avoid this problem, operator names are encoded as follows: 165 166 -- Oabs abs 167 -- Oand and 168 -- Omod mod 169 -- Onot not 170 -- Oor or 171 -- Orem rem 172 -- Oxor xor 173 -- Oeq = 174 -- One /= 175 -- Olt < 176 -- Ole <= 177 -- Ogt > 178 -- Oge >= 179 -- Oadd + 180 -- Osubtract - 181 -- Oconcat & 182 -- Omultiply * 183 -- Odivide / 184 -- Oexpon ** 185 186 -- These names are prefixed by the normal full qualification, and 187 -- suffixed by the overloading identification. So for example, the 188 -- second operator "=" defined in package Extra.Messages would have 189 -- the name: 190 191 -- extra__messages__Oeq__2 192 193 ---------------------------------- 194 -- Resolving Other Name Clashes -- 195 ---------------------------------- 196 197 -- It might be thought that the above scheme is complete, but in Ada 95, 198 -- full qualification is insufficient to uniquely identify an entity in 199 -- the program, even if it is not an overloaded subprogram. There are 200 -- two possible confusions: 201 202 -- a.b 203 204 -- interpretation 1: entity b in body of package a 205 -- interpretation 2: child procedure b of package a 206 207 -- a.b.c 208 209 -- interpretation 1: entity c in child package a.b 210 -- interpretation 2: entity c in nested package b in body of a 211 212 -- It is perfectly legal in both cases for both interpretations to be 213 -- valid within a single program. This is a bit of a surprise since 214 -- certainly in Ada 83, full qualification was sufficient, but not in 215 -- Ada 95. The result is that the above scheme can result in duplicate 216 -- names. This would not be so bad if the effect were just restricted 217 -- to debugging information, but in fact in both the above cases, it 218 -- is possible for both symbols to be external names, and so we have 219 -- a real problem of name clashes. 220 221 -- To deal with this situation, we provide two additional encoding 222 -- rules for names: 223 224 -- First: all library subprogram names are preceded by the string 225 -- _ada_ (which causes no duplications, since normal Ada names can 226 -- never start with an underscore. This not only solves the first 227 -- case of duplication, but also solves another pragmatic problem 228 -- which is that otherwise Ada procedures can generate names that 229 -- clash with existing system function names. Most notably, we can 230 -- have clashes in the case of procedure Main with the C main that 231 -- in some systems is always present. 232 233 -- Second, for the case where nested packages declared in package 234 -- bodies can cause trouble, we add a suffix which shows which 235 -- entities in the list are body-nested packages, i.e. packages 236 -- whose spec is within a package body. The rules are as follows, 237 -- given a list of names in a qualified name name1.name2.... 238 239 -- If none are body-nested package entities, then there is no suffix 240 241 -- If at least one is a body-nested package entity, then the suffix 242 -- is X followed by a string of b's and n's (b = body-nested package 243 -- entity, n = not a body-nested package). 244 245 -- There is one element in this string for each entity in the encoded 246 -- expanded name except the first (the rules are such that the first 247 -- entity of the encoded expanded name can never be a body-nested' 248 -- package. Trailing n's are omitted, as is the last b (there must 249 -- be at least one b, or we would not be generating a suffix at all). 250 251 -- For example, suppose we have 252 253 -- package x is 254 -- pragma Elaborate_Body; 255 -- m1 : integer; -- #1 256 -- end x; 257 258 -- package body x is 259 -- package y is m2 : integer; end y; -- #2 260 -- package body y is 261 -- package z is r : integer; end z; -- #3 262 -- end; 263 -- m3 : integer; -- #4 264 -- end x; 265 266 -- package x.y is 267 -- pragma Elaborate_Body; 268 -- m2 : integer; -- #5 269 -- end x.y; 270 271 -- package body x.y is 272 -- m3 : integer; -- #6 273 -- procedure j is -- #7 274 -- package k is 275 -- z : integer; -- #8 276 -- end k; 277 -- begin 278 -- null; 279 -- end j; 280 -- end x.y; 281 282 -- procedure x.m3 is begin null; end; -- #9 283 284 -- Then the encodings would be: 285 286 -- #1. x__m1 (no BNPE's in sight) 287 -- #2. x__y__m2X (y is a BNPE) 288 -- #3. x__y__z__rXb (y is a BNPE, so is z) 289 -- #4. x__m3 (no BNPE's in sight) 290 -- #5. x__y__m2 (no BNPE's in sight) 291 -- #6. x__y__m3 (no BNPE's in signt) 292 -- #7. x__y__j (no BNPE's in sight) 293 -- #8. k__z (no BNPE's, only up to procedure) 294 -- #9 _ada_x__m3 (library-level subprogram) 295 296 -- Note that we have instances here of both kind of potential name 297 -- clashes, and the above examples show how the encodings avoid the 298 -- clash as follows: 299 300 -- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library 301 -- level subprogram, so it is preceded by the string _ada_ which acts 302 -- to distinguish it from the package body entity. 303 304 -- Lines #2 and #5 both refer to the entity x.y.m2, but the first 305 -- instance is inside the body-nested package y, so there is an X 306 -- suffix to distinguish it from the child library entity. 307 308 -- Note that enumeration literals never need Xb type suffixes, since 309 -- they are never referenced using global external names. 310 311 --------------------- 312 -- Interface Names -- 313 --------------------- 314 315 -- Note: if an interface name is present, then the external name is 316 -- taken from the specified interface name. Given current limitations of 317 -- the gcc backend, this means that the debugging name is also set to 318 -- the interface name, but conceptually, it would be possible (and 319 -- indeed desirable) to have the debugging information still use the Ada 320 -- name as qualified above, so we still fully qualify the name in the 321 -- front end. 322 323 ------------------------------------- 324 -- Encodings Related to Task Types -- 325 ------------------------------------- 326 327 -- Each task object defined by a single task declaration is associated 328 -- with a prefix that is used to qualify procedures defined in that 329 -- task. Given 330 -- 331 -- package body P is 332 -- task body TaskObj is 333 -- procedure F1 is ... end; 334 -- begin 335 -- B; 336 -- end TaskObj; 337 -- end P; 338 -- 339 -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1. 340 -- The body, B, is contained in a subprogram whose name is 341 -- p__taskobjTKB. 342 343 ------------------------------------------ 344 -- Encodings Related to Protected Types -- 345 ------------------------------------------ 346 347 -- Each protected type has an associated record type, that describes 348 -- the actual layout of the private data. In addition to the private 349 -- components of the type, the Corresponding_Record_Type includes one 350 -- component of type Protection, which is the actual lock structure. 351 -- The run-time size of the protected type is the size of the corres- 352 -- ponding record. 353 354 -- For a protected type prot, the Corresponding_Record_Type is encoded 355 -- as protV. 356 357 -- The operations of a protected type are encoded as follows: each 358 -- operation results in two subprograms, a locking one that is called 359 -- from outside of the object, and a non-locking one that is used for 360 -- calls from other operations on the same object. The locking operation 361 -- simply acquires the lock, and then calls the non-locking version. 362 -- The names of all of these have a prefix constructed from the name of 363 -- the type, and a suffix which is P or N, depending on whether this is 364 -- the protected/non-locking version of the operation. 365 366 -- Operations generated for protected entries follow the same encoding. 367 -- Each entry results in two subprograms: a procedure that holds the 368 -- entry body, and a function that holds the evaluation of the barrier. 369 -- The names of these subprograms include the prefix '_E' or '_B' res- 370 -- pectively. The names also include a numeric suffix to render them 371 -- unique in the presence of overloaded entries. 372 373 -- Given the declaration: 374 375 -- protected type Lock is 376 -- function Get return Integer; 377 -- procedure Set (X: Integer); 378 -- entry Update (Val : Integer); 379 -- private 380 -- Value : Integer := 0; 381 -- end Lock; 382 383 -- the following operations are created: 384 385 -- lock_getN 386 -- lock_getP, 387 388 -- lock_setN 389 -- lock_setP 390 391 -- lock_update_E1s 392 -- lock_udpate_B2s 393 394 -- If the protected type implements at least one interface, the 395 -- following additional operations are created: 396 397 -- lock_get 398 399 -- lock_set 400 401 -- These operations are used to ensure overriding of interface level 402 -- subprograms and proper dispatching on interface class-wide objects. 403 -- The bodies of these operations contain calls to their respective 404 -- protected versions: 405 406 -- function lock_get return Integer is 407 -- begin 408 -- return lock_getP; 409 -- end lock_get; 410 411 -- procedure lock_set (X : Integer) is 412 -- begin 413 -- lock_setP (X); 414 -- end lock_set; 415 416 ---------------------------------------------------- 417 -- Conversion between Entities and External Names -- 418 ---------------------------------------------------- 419 420 procedure Get_External_Name 421 (Entity : Entity_Id; 422 Has_Suffix : Boolean := False; 423 Suffix : String := ""); 424 -- Set Name_Buffer and Name_Len to the external name of the entity. The 425 -- external name is the Interface_Name, if specified, unless the entity 426 -- has an address clause or Has_Suffix is true. 427 -- 428 -- If the Interface is not present, or not used, the external name is the 429 -- concatenation of: 430 -- 431 -- - the string "_ada_", if the entity is a library subprogram, 432 -- - the names of any enclosing scopes, each followed by "__", 433 -- or "X_" if the next entity is a subunit) 434 -- - the name of the entity 435 -- - the string "$" (or "__" if target does not allow "$"), followed 436 -- by homonym suffix, if the entity is an overloaded subprogram 437 -- or is defined within an overloaded subprogram. 438 -- - the string "___" followed by Suffix if Has_Suffix is true. 439 -- 440 -- Note that a call to this procedure has no effect if we are not 441 -- generating code, since the necessary information for computing the 442 -- proper external name is not available in this case. 443 444 ------------------------------------- 445 -- Encoding for translation into C -- 446 ------------------------------------- 447 448 -- In Modify_Tree_For_C mode we must add encodings to dismabiguate cases 449 -- where Ada block structure cannot be directly translated. These cases 450 -- are as follows: 451 452 -- a) A loop variable may hide a homonym in an enclosing block 453 -- b) A block-local variable may hide a homonym in an enclosing block 454 455 -- In C these constructs are not scopes and we must distinguish the names 456 -- explicitly. In the first case we create a qualified name with the suffix 457 -- 'L', in the second case with a suffix 'B'. 458 459 -------------------------------------------- 460 -- Subprograms for Handling Qualification -- 461 -------------------------------------------- 462 463 procedure Qualify_Entity_Names (N : Node_Id); 464 -- Given a node N, that represents a block, subprogram body, or package 465 -- body or spec, or protected or task type, sets a fully qualified name 466 -- for the defining entity of given construct, and also sets fully 467 -- qualified names for all enclosed entities of the construct (using 468 -- First_Entity/Next_Entity). Note that the actual modifications of the 469 -- names is postponed till a subsequent call to Qualify_All_Entity_Names. 470 -- Note: this routine does not deal with prepending _ada_ to library 471 -- subprogram names. The reason for this is that we only prepend _ada_ 472 -- to the library entity itself, and not to names built from this name. 473 474 procedure Qualify_All_Entity_Names; 475 -- When Qualify_Entity_Names is called, no actual name changes are made, 476 -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call 477 -- is made to this procedure. The reason for this deferral is that when 478 -- names are changed semantic processing may be affected. By deferring 479 -- the changes till just before gigi is called, we avoid any concerns 480 -- about such effects. Gigi itself does not use the names except for 481 -- output of names for debugging purposes (which is why we are doing 482 -- the name changes in the first place). 483 484 -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are 485 -- useful to remove qualification from a name qualified by the call to 486 -- Qualify_All_Entity_Names. 487 488 -------------------------------- 489 -- Handling of Numeric Values -- 490 -------------------------------- 491 492 -- All numeric values here are encoded as strings of decimal digits. Only 493 -- integer values need to be encoded. A negative value is encoded as the 494 -- corresponding positive value followed by a lower case m for minus to 495 -- indicate that the value is negative (e.g. 2m for -2). 496 497 ------------------------- 498 -- Type Name Encodings -- 499 ------------------------- 500 501 -- In the following typ is the name of the type as normally encoded by the 502 -- debugger rules, i.e. a non-qualified name, all in lower case, with 503 -- standard encoding of upper half and wide characters 504 505 ------------------------ 506 -- Encapsulated Types -- 507 ------------------------ 508 509 -- In some cases, the compiler encapsulates a type by wrapping it in a 510 -- structure. For example, this is used when a size or alignment 511 -- specification requires a larger type. Consider: 512 513 -- type y is mod 2 ** 64; 514 -- for y'size use 256; 515 516 -- In this case the compile generates a structure type y___PAD, which 517 -- has a single field whose name is F. This single field is 64 bits 518 -- long and contains the actual value. This kind of padding is used 519 -- when the logical value to be stored is shorter than the object in 520 -- which it is allocated. For example if a size clause is used to set 521 -- a size of 256 for a signed integer value, then a typical choice is 522 -- to wrap a 64-bit integer in a 256 bit PAD structure. 523 524 -- A similar encapsulation is done for some packed array types, in which 525 -- case the structure type is y___JM and the field name is OBJECT. 526 -- This is used in the case of a packed array stored using modular 527 -- representation (see section on representation of packed array 528 -- objects). In this case the JM wrapping is used to achieve correct 529 -- positioning of the packed array value (left or right justified in its 530 -- field depending on endianness. 531 532 -- When the debugger sees an object of a type whose name has a suffix of 533 -- ___PAD or ___JM, the type will be a record containing a single field, 534 -- and the name of that field will be all upper case. In this case, it 535 -- should look inside to get the value of the inner field, and neither 536 -- the outer structure name, nor the field name should appear when the 537 -- value is printed. 538 539 -- When the debugger sees a record named REP being a field inside 540 -- another record, it should treat the fields inside REP as being part 541 -- of the outer record (this REP field is only present for code 542 -- generation purposes). The REP record should not appear in the values 543 -- printed by the debugger. 544 545 ----------------------- 546 -- Fixed-Point Types -- 547 ----------------------- 548 549 -- Fixed-point types are encoded using a suffix that indicates the 550 -- delta and small values. The actual type itself is a normal integer 551 -- type. 552 553 -- typ___XF_nn_dd 554 -- typ___XF_nn_dd_nn_dd 555 556 -- The first form is used when small = delta. The value of delta (and 557 -- small) is given by the rational nn/dd, where nn and dd are decimal 558 -- integers. 559 -- 560 -- The second form is used if the small value is different from the 561 -- delta. In this case, the first nn/dd rational value is for delta, 562 -- and the second value is for small. 563 564 -------------------- 565 -- Discrete Types -- 566 -------------------- 567 568 -- Discrete types are coded with a suffix indicating the range in the 569 -- case where one or both of the bounds are discriminants or variable. 570 571 -- Note: at the current time, we also encode compile time known bounds 572 -- if they do not match the natural machine type bounds, but this may 573 -- be removed in the future, since it is redundant for most debugging 574 -- formats. However, we do not ever need XD encoding for enumeration 575 -- base types, since here it is always clear what the bounds are from 576 -- the total number of enumeration literals. 577 578 -- typ___XD 579 -- typ___XDL_lowerbound 580 -- typ___XDU_upperbound 581 -- typ___XDLU_lowerbound__upperbound 582 583 -- If a discrete type is a natural machine type (i.e. its bounds 584 -- correspond in a natural manner to its size), then it is left 585 -- unencoded. The above encoding forms are used when there is a 586 -- constrained range that does not correspond to the size or that 587 -- has discriminant references or other compile time known bounds. 588 589 -- The first form is used if both bounds are dynamic, in which case two 590 -- constant objects are present whose names are typ___L and typ___U in 591 -- the same scope as typ, and the values of these constants indicate 592 -- the bounds. As far as the debugger is concerned, these are simply 593 -- variables that can be accessed like any other variables. In the 594 -- enumeration case, these values correspond to the Enum_Rep values for 595 -- the lower and upper bounds. 596 597 -- The second form is used if the upper bound is dynamic, but the lower 598 -- bound is either constant or depends on a discriminant of the record 599 -- with which the type is associated. The upper bound is stored in a 600 -- constant object of name typ___U as previously described, but the 601 -- lower bound is encoded directly into the name as either a decimal 602 -- integer, or as the discriminant name. 603 604 -- The third form is similarly used if the lower bound is dynamic, but 605 -- the upper bound is compile time known or a discriminant reference, 606 -- in which case the lower bound is stored in a constant object of name 607 -- typ___L, and the upper bound is encoded directly into the name as 608 -- either a decimal integer, or as the discriminant name. 609 610 -- The fourth form is used if both bounds are discriminant references 611 -- or compile time known values, with the encoding first for the lower 612 -- bound, then for the upper bound, as previously described. 613 614 ------------------- 615 -- Modular Types -- 616 ------------------- 617 618 -- A type declared 619 620 -- type x is mod N; 621 622 -- Is encoded as a subrange of an unsigned base type with lower bound 623 -- zero and upper bound N. That is, there is no name encoding. We use 624 -- the standard encodings provided by the debugging format. Thus we 625 -- give these types a non-standard interpretation: the standard 626 -- interpretation of our encoding would not, in general, imply that 627 -- arithmetic on type x was to be performed modulo N (especially not 628 -- when N is not a power of 2). 629 630 ------------------ 631 -- Biased Types -- 632 ------------------ 633 634 -- Only discrete types can be biased, and the fact that they are biased 635 -- is indicated by a suffix of the form: 636 637 -- typ___XB_lowerbound__upperbound 638 639 -- Here lowerbound and upperbound are decimal integers, with the usual 640 -- (postfix "m") encoding for negative numbers. Biased types are only 641 -- possible where the bounds are compile time known, and the values are 642 -- represented as unsigned offsets from the lower bound given. For 643 -- example: 644 645 -- type Q is range 10 .. 15; 646 -- for Q'size use 3; 647 648 -- The size clause will force values of type Q in memory to be stored 649 -- in biased form (e.g. 11 will be represented by the bit pattern 001). 650 651 ---------------------------------------------- 652 -- Record Types with Variable-Length Fields -- 653 ---------------------------------------------- 654 655 -- The debugging formats do not fully support these types, and indeed 656 -- some formats simply generate no useful information at all for such 657 -- types. In order to provide information for the debugger, gigi creates 658 -- a parallel type in the same scope with one of the names 659 660 -- type___XVE 661 -- type___XVU 662 663 -- The former name is used for a record and the latter for the union 664 -- that is made for a variant record (see below) if that record or union 665 -- has a field of variable size or if the record or union itself has a 666 -- variable size. These encodings suffix any other encodings that that 667 -- might be suffixed to the type name. 668 669 -- The idea here is to provide all the needed information to interpret 670 -- objects of the original type in the form of a "fixed up" type, which 671 -- is representable using the normal debugging information. 672 673 -- There are three cases to be dealt with. First, some fields may have 674 -- variable positions because they appear after variable-length fields. 675 -- To deal with this, we encode *all* the field bit positions of the 676 -- special ___XV type in a non-standard manner. 677 678 -- The idea is to encode not the position, but rather information that 679 -- allows computing the position of a field from the position of the 680 -- previous field. The algorithm for computing the actual positions of 681 -- all fields and the length of the record is as follows. In this 682 -- description, let P represent the current bit position in the record. 683 684 -- 1. Initialize P to 0 685 686 -- 2. For each field in the record: 687 688 -- 2a. If an alignment is given (see below), then round P up, if 689 -- needed, to the next multiple of that alignment. 690 691 -- 2b. If a bit position is given, then increment P by that amount 692 -- (that is, treat it as an offset from the end of the preceding 693 -- record). 694 695 -- 2c. Assign P as the actual position of the field 696 697 -- 2d. Compute the length, L, of the represented field (see below) 698 -- and compute P'=P+L. Unless the field represents a variant part 699 -- (see below and also Variant Record Encoding), set P to P'. 700 701 -- The alignment, if present, is encoded in the field name of the 702 -- record, which has a suffix: 703 704 -- fieldname___XVAnn 705 706 -- where the nn after the XVA indicates the alignment value in storage 707 -- units. This encoding is present only if an alignment is present. 708 709 -- The size of the record described by an XVE-encoded type (in bits) is 710 -- generally the maximum value attained by P' in step 2d above, rounded 711 -- up according to the record's alignment. 712 713 -- Second, the variable-length fields themselves are represented by 714 -- replacing the type by a special access type. The designated type of 715 -- this access type is the original variable-length type, and the fact 716 -- that this field has been transformed in this way is signalled by 717 -- encoding the field name as: 718 719 -- field___XVL 720 721 -- where field is the original field name. If a field is both 722 -- variable-length and also needs an alignment encoding, then the 723 -- encodings are combined using: 724 725 -- field___XVLnn 726 727 -- Note: the reason that we change the type is so that the resulting 728 -- type has no variable-length fields. At least some of the formats used 729 -- for debugging information simply cannot tolerate variable- length 730 -- fields, so the encoded information would get lost. 731 732 -- Third, in the case of a variant record, the special union that 733 -- contains the variants is replaced by a normal C union. In this case, 734 -- the positions are all zero. 735 736 -- Discriminants appear before any variable-length fields that depend on 737 -- them, with one exception. In some cases, a discriminant governing the 738 -- choice of a variant clause may appear in the list of fields of an XVE 739 -- type after the entry for the variant clause itself (this can happen 740 -- in the presence of a representation clause for the record type in the 741 -- source program). However, when this happens, the discriminant's 742 -- position may be determined by first applying the rules described in 743 -- this section, ignoring the variant clause. As a result, discriminants 744 -- can always be located independently of the variable-length fields 745 -- that depend on them. 746 747 -- The size of the ___XVE or ___XVU record or union is set to the 748 -- alignment (in bytes) of the original object so that the debugger 749 -- can calculate the size of the original type. 750 751 -- As an example of this encoding, consider the declarations: 752 753 -- type Q is array (1 .. V1) of Float; -- alignment 4 754 -- type R is array (1 .. V2) of Long_Float; -- alignment 8 755 756 -- type X is record 757 -- A : Character; 758 -- B : Float; 759 -- C : String (1 .. V3); 760 -- D : Float; 761 -- E : Q; 762 -- F : R; 763 -- G : Float; 764 -- end record; 765 766 -- The encoded type looks like: 767 768 -- type anonymousQ is access Q; 769 -- type anonymousR is access R; 770 771 -- type X___XVE is record 772 -- A : Character; -- position contains 0 773 -- B : Float; -- position contains 24 774 -- C___XVL : access String (1 .. V3); -- position contains 0 775 -- D___XVA4 : Float; -- position contains 0 776 -- E___XVL4 : anonymousQ; -- position contains 0 777 -- F___XVL8 : anonymousR; -- position contains 0 778 -- G : Float; -- position contains 0 779 -- end record; 780 781 -- Any bit sizes recorded for fields other than dynamic fields and 782 -- variants are honored as for ordinary records. 783 784 -- Notes: 785 786 -- 1) The B field could also have been encoded by using a position of 787 -- zero and an alignment of 4, but in such a case the coding by position 788 -- is preferred (since it takes up less space). We have used the 789 -- (illegal) notation access xxx as field types in the example above. 790 791 -- 2) The E field does not actually need the alignment indication but 792 -- this may not be detected in this case by the conversion routines. 793 794 -- 3) Our conventions do not cover all XVE-encoded records in which 795 -- some, but not all, fields have representation clauses. Such records 796 -- may, therefore, be displayed incorrectly by debuggers. This situation 797 -- is not common. 798 799 ----------------------- 800 -- Base Record Types -- 801 ----------------------- 802 803 -- Under certain circumstances, debuggers need two descriptions of a 804 -- record type, one that gives the actual details of the base type's 805 -- structure (as described elsewhere in these comments) and one that may 806 -- be used to obtain information about the particular subtype and the 807 -- size of the objects being typed. In such cases the compiler will 808 -- substitute type whose name is typically compiler-generated and 809 -- irrelevant except as a key for obtaining the actual type. 810 811 -- Specifically, if this name is x, then we produce a record type named 812 -- x___XVS consisting of one field. The name of this field is that of 813 -- the actual type being encoded, which we'll call y. The type of this 814 -- single field can be either an arbitrary non-reference type, e.g. an 815 -- integer type, or a reference type; in the latter case, the referenced 816 -- type is also the actual type being encoded y. Both x and y may have 817 -- corresponding ___XVE types. 818 819 -- The size of the objects typed as x should be obtained from the 820 -- structure of x (and x___XVE, if applicable) as for ordinary types 821 -- unless there is a variable named x___XVZ, which, if present, will 822 -- hold the size (in bytes) of x. In this latter case, the size of the 823 -- x___XVS type will not be a constant but a reference to x___XVZ. 824 825 -- The type x will either be a subtype of y (see also Subtypes of 826 -- Variant Records, below) or will contain a single field of type y, 827 -- or no fields at all. The layout, types, and positions of these 828 -- fields will be accurate, if present. (Currently, however, the GDB 829 -- debugger makes no use of x except to determine its size). 830 831 -- Among other uses, XVS types are used to encode unconstrained types. 832 -- For example, given: 833 -- 834 -- subtype Int is INTEGER range 0..10; 835 -- type T1 (N: Int := 0) is record 836 -- F1: String (1 .. N); 837 -- end record; 838 -- type AT1 is array (INTEGER range <>) of T1; 839 -- 840 -- the element type for AT1 might have a type defined as if it had 841 -- been written: 842 -- 843 -- type at1___PAD is record F : T1; end record; 844 -- for at1___PAD'Size use 16 * 8; 845 -- 846 -- and there would also be: 847 -- 848 -- type at1___PAD___XVS is record t1: reft1; end record; 849 -- type t1 is ... 850 -- type reft1 is <reference to t1> 851 -- 852 -- Had the subtype Int been dynamic: 853 -- 854 -- subtype Int is INTEGER range 0 .. M; -- M a variable 855 -- 856 -- Then the compiler would also generate a declaration whose effect 857 -- would be 858 -- 859 -- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term; 860 -- 861 -- Not all unconstrained types are so encoded; the XVS convention may be 862 -- unnecessary for unconstrained types of fixed size. However, this 863 -- encoding is always necessary when a subcomponent type (array 864 -- element's type or record field's type) is an unconstrained record 865 -- type some of whose components depend on discriminant values. 866 867 ----------------- 868 -- Array Types -- 869 ----------------- 870 871 -- Since there is no way for the debugger to obtain the index subtypes 872 -- for an array type, we produce a type that has the name of the array 873 -- type followed by "___XA" and is a record type whose field types are 874 -- the respective types for the bounds (and whose field names are the 875 -- names of these types). 876 877 -- To conserve space, we do not produce this type unless one of the 878 -- index types is either an enumeration type, has a variable lower or 879 -- upper bound or is a biased type. 880 881 -- Given the full encoding of these types (see above description for 882 -- the encoding of discrete types), this means that all necessary 883 -- information for addressing arrays is available. In some debugging 884 -- formats, some or all of the bounds information may be available 885 -- redundantly, particularly in the fixed-point case, but this 886 -- information can in any case be ignored by the debugger. 887 888 ---------------------------- 889 -- Note on Implicit Types -- 890 ---------------------------- 891 892 -- The compiler creates implicit type names in many situations where a 893 -- type is present semantically, but no specific name is present. For 894 -- example: 895 896 -- S : Integer range M .. N; 897 898 -- Here the subtype of S is not integer, but rather an anonymous subtype 899 -- of Integer. Where possible, the compiler generates names for such 900 -- anonymous types that are related to the type from which the subtype 901 -- is obtained as follows: 902 903 -- T name suffix 904 905 -- where name is the name from which the subtype is obtained, using 906 -- lower case letters and underscores, and suffix starts with an upper 907 -- case letter. For example the name for the above declaration might be: 908 909 -- TintegerS4b 910 911 -- If the debugger is asked to give the type of an entity and the type 912 -- has the form T name suffix, it is probably appropriate to just use 913 -- "name" in the response since this is what is meaningful to the 914 -- programmer. 915 916 ------------------------------------------------- 917 -- Subprograms for Handling Encoded Type Names -- 918 ------------------------------------------------- 919 920 procedure Get_Encoded_Name (E : Entity_Id); 921 -- If the entity is a typename, store the external name of the entity as in 922 -- Get_External_Name, followed by three underscores plus the type encoding 923 -- in Name_Buffer with the length in Name_Len, and an ASCII.NUL character 924 -- stored following the name. Otherwise set Name_Buffer and Name_Len to 925 -- hold the entity name. Note that a call to this procedure has no effect 926 -- if we are not generating code, since the necessary information for 927 -- computing the proper encoded name is not available in this case. 928 929 -------------- 930 -- Renaming -- 931 -------------- 932 933 -- Debugging information is generated for exception, object, package, and 934 -- subprogram renaming (generic renamings are not significant, since 935 -- generic templates are not relevant at debugging time). 936 937 -- Consider a renaming declaration of the form 938 939 -- x : typ renames y; 940 941 -- There is one case in which no special debugging information is required, 942 -- namely the case of an object renaming where the back end allocates a 943 -- reference for the renamed variable, and the entity x is this reference. 944 -- The debugger can handle this case without any special processing or 945 -- encoding (it won't know it was a renaming, but that does not matter). 946 947 -- All other cases of renaming generate a dummy variable for an entity 948 -- whose name is of the form: 949 950 -- x___XR_... for an object renaming 951 -- x___XRE_... for an exception renaming 952 -- x___XRP_... for a package renaming 953 954 -- and where the "..." represents a suffix that describes the structure of 955 -- the object name given in the renaming (see details below). 956 957 -- The name is fully qualified in the usual manner, i.e. qualified in the 958 -- same manner as the entity x would be. In the case of a package renaming 959 -- where x is a child unit, the qualification includes the name of the 960 -- parent unit, to disambiguate child units with the same simple name and 961 -- (of necessity) different parents. 962 963 -- Note: subprogram renamings are not encoded at the present time 964 965 -- The suffix of the variable name describing the renamed object is defined 966 -- to use the following encoding: 967 968 -- For the simple entity case, where y is just an entity name, the suffix 969 -- is of the form: 970 971 -- y___XE 972 973 -- i.e. the suffix has a single field, the first part matching the 974 -- name y, followed by a "___" separator, ending with sequence XE. 975 -- The entity name portion is fully qualified in the usual manner. 976 -- This same naming scheme is followed for all forms of encoded 977 -- renamings that rename a simple entity. 978 979 -- For the object renaming case where y is a selected component or an 980 -- indexed component, the variable name is suffixed by additional fields 981 -- that give details of the components. The name starts as above with a 982 -- y___XE name indicating the outer level object entity. Then a series of 983 -- selections and indexing operations can be specified as follows: 984 985 -- Indexed component 986 987 -- A series of subscript values appear in sequence, the number 988 -- corresponds to the number of dimensions of the array. The 989 -- subscripts have one of the following two forms: 990 991 -- XSnnn 992 993 -- Here nnn is a constant value, encoded as a decimal integer 994 -- (pos value for enumeration type case). Negative values have 995 -- a trailing 'm' as usual. 996 997 -- XSe 998 999 -- Here e is the (unqualified) name of a constant entity in the 1000 -- same scope as the renaming which contains the subscript value. 1001 1002 -- Slice 1003 1004 -- For the slice case, we have two entries. The first is for the 1005 -- lower bound of the slice, and has the form: 1006 1007 -- XLnnn 1008 -- XLe 1009 1010 -- Specifies the lower bound, using exactly the same encoding as 1011 -- for an XS subscript as described above. 1012 1013 -- Then the upper bound appears in the usual XSnnn/XSe form 1014 1015 -- Selected component 1016 1017 -- For a selected component, we have a single entry 1018 1019 -- XRf 1020 1021 -- Here f is the field name for the selection 1022 1023 -- For an explicit dereference (.all), we have a single entry 1024 1025 -- XA 1026 1027 -- As an example, consider the declarations: 1028 1029 -- package p is 1030 -- type q is record 1031 -- m : string (2 .. 5); 1032 -- end record; 1033 -- 1034 -- type r is array (1 .. 10, 1 .. 20) of q; 1035 -- 1036 -- g : r; 1037 -- 1038 -- z : string renames g (1,5).m(2 ..3) 1039 -- end p; 1040 1041 -- The generated variable entity would appear as 1042 1043 -- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type; 1044 -- p__g___XE--------------------outer entity is g 1045 -- XS1-----------------first subscript for g 1046 -- XS5--------------second subscript for g 1047 -- XRm-----------select field m 1048 -- XL2--------lower bound of slice 1049 -- XS3-----upper bound of slice 1050 1051 -- Note that the type of the variable is a special internal type named 1052 -- _renaming_type. This type is an arbitrary type of zero size created 1053 -- in package Standard (see cstand.adb) and is ignored by the debugger. 1054 1055 function Debug_Renaming_Declaration (N : Node_Id) return Node_Id; 1056 -- The argument N is a renaming declaration. The result is a variable 1057 -- declaration as described in the above paragraphs. If N is not a special 1058 -- debug declaration, then Empty is returned. This function also takes care 1059 -- of setting Materialize_Entity on the renamed entity where required. 1060 1061 --------------------------- 1062 -- Packed Array Encoding -- 1063 --------------------------- 1064 1065 -- For every constrained packed array, two types are created, and both 1066 -- appear in the debugging output: 1067 1068 -- The original declared array type is a perfectly normal array type, and 1069 -- its index bounds indicate the original bounds of the array. 1070 1071 -- The corresponding packed array type, which may be a modular type, or 1072 -- may be an array of bytes type (see Exp_Pakd for full details). This is 1073 -- the type that is actually used in the generated code and for debugging 1074 -- information for all objects of the packed type. 1075 1076 -- The name of the corresponding packed array type is: 1077 1078 -- ttt___XPnnn 1079 1080 -- where 1081 1082 -- ttt is the name of the original declared array 1083 -- nnn is the component size in bits (1-31) 1084 1085 -- Note that if the packed array is not bit-packed, the name will simply 1086 -- be tttP. 1087 1088 -- When the debugger sees that an object is of a type that is encoded in 1089 -- this manner, it can use the original type to determine the bounds and 1090 -- the component type, and the component size to determine the packing 1091 -- details. 1092 1093 -- For an unconstrained packed array, the corresponding packed array type 1094 -- is neither used in the generated code nor for debugging information, 1095 -- only the original type is used. In order to convey the packing in the 1096 -- debugging information, the compiler generates the associated fat- and 1097 -- thin-pointer types (see the Pointers to Unconstrained Array section 1098 -- below) using the name of the corresponding packed array type as the 1099 -- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively. 1100 1101 -- When the debugger sees that an object is of a type that is encoded in 1102 -- this manner, it can use the type of the fields to determine the bounds 1103 -- and the component type, and the component size to determine the packing 1104 -- details. 1105 1106 ------------------------------------------- 1107 -- Packed Array Representation in Memory -- 1108 ------------------------------------------- 1109 1110 -- Packed arrays are represented in tightly packed form, with no extra bits 1111 -- between components. This is true even when the component size is not a 1112 -- factor of the storage unit size, so that as a result it is possible for 1113 -- components to cross storage unit boundaries. 1114 1115 -- The layout in storage is identical, regardless of whether the 1116 -- implementation type is a modular type or an array-of-bytes type. See 1117 -- Exp_Pakd for details of how these implementation types are used, but for 1118 -- the purpose of the debugger, only the starting address of the object in 1119 -- memory is significant. 1120 1121 -- The following example should show clearly how the packing works in 1122 -- the little-endian and big-endian cases: 1123 1124 -- type B is range 0 .. 7; 1125 -- for B'Size use 3; 1126 1127 -- type BA is array (0 .. 5) of B; 1128 -- pragma Pack (BA); 1129 1130 -- BV : constant BA := (1,2,3,4,5,6); 1131 1132 -- Little endian case 1133 1134 -- BV'Address + 2 BV'Address + 1 BV'Address + 0 1135 -- +-----------------+-----------------+-----------------+ 1136 -- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 | 1137 -- +-----------------+-----------------+-----------------+ 1138 -- <---------> <-----> <---> <---> <-----> <---> <---> 1139 -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0) 1140 -- 1141 -- Big endian case 1142 -- 1143 -- BV'Address + 0 BV'Address + 1 BV'Address + 2 1144 -- +-----------------+-----------------+-----------------+ 1145 -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? | 1146 -- +-----------------+-----------------+-----------------+ 1147 -- <---> <---> <-----> <---> <---> <-----> <---------> 1148 -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits 1149 1150 -- Note that if a modular type is used to represent the array, the 1151 -- allocation in memory is not the same as a normal modular type. The 1152 -- difference occurs when the allocated object is larger than the size of 1153 -- the array. For a normal modular type, we extend the value on the left 1154 -- with zeroes. 1155 1156 -- For example, in the normal modular case, if we have a 6-bit modular 1157 -- type, declared as mod 2**6, and we allocate an 8-bit object for this 1158 -- type, then we extend the value with two bits on the most significant 1159 -- end, and in either the little-endian or big-endian case, the value 63 1160 -- is represented as 00111111 in binary in memory. 1161 1162 -- For a modular type used to represent a packed array, the rule is 1163 -- different. In this case, if we have to extend the value, then we do it 1164 -- with undefined bits (which are not initialized and whose value is 1165 -- irrelevant to any generated code). Furthermore these bits are on the 1166 -- right (least significant bits) in the big-endian case, and on the left 1167 -- (most significant bits) in the little-endian case. 1168 1169 -- For example, if we have a packed boolean array of 6 bits, all set to 1170 -- True, stored in an 8-bit object, then the value in memory in binary is 1171 -- ??111111 in the little-endian case, and 111111?? in the big-endian case. 1172 1173 -- This is done so that the representation of packed arrays does not 1174 -- depend on whether we use a modular representation or array of bytes 1175 -- as previously described. This ensures that we can pass such values by 1176 -- reference in the case where a subprogram has to be able to handle values 1177 -- stored in either form. 1178 1179 -- Note that when we extract the value of such a modular packed array, we 1180 -- expect to retrieve only the relevant bits, so in this same example, when 1181 -- we extract the value we get 111111 in both cases, and the code generated 1182 -- by the front end assumes this although it does not assume that any high 1183 -- order bits are defined. 1184 1185 -- There are opportunities for optimization based on the knowledge that the 1186 -- unused bits are irrelevant for these type of packed arrays. For example 1187 -- if we have two such 6-bit-in-8-bit values and we do an assignment: 1188 1189 -- a := b; 1190 1191 -- Then logically, we extract the 6 bits and store only 6 bits in the 1192 -- result, but the back end is free to simply assign the entire 8-bits in 1193 -- this case, since we don't actually care about the undefined bits. 1194 -- However, in the equality case, it is important to ensure that the 1195 -- undefined bits do not participate in an equality test. 1196 1197 -- If a modular packed array value is assigned to a register then logically 1198 -- it could always be held right justified, to avoid any need to shift, 1199 -- e.g. when doing comparisons. But probably this is a bad choice, as it 1200 -- would mean that an assignment such as a := above would require shifts 1201 -- when one value is in a register and the other value is in memory. 1202 1203 ------------------------------------------------------ 1204 -- Subprograms for Handling Packed Array Type Names -- 1205 ------------------------------------------------------ 1206 1207 function Make_Packed_Array_Impl_Type_Name 1208 (Typ : Entity_Id; 1209 Csize : Uint) return Name_Id; 1210 -- This function is used in Exp_Pakd to create the name that is encoded as 1211 -- described above. The entity Typ provides the name ttt, and the value 1212 -- Csize is the component size that provides the nnn value. 1213 1214 -------------------------------------- 1215 -- Pointers to Unconstrained Arrays -- 1216 -------------------------------------- 1217 1218 -- There are two kinds of pointers to arrays. The debugger can tell which 1219 -- format is in use by the form of the type of the pointer. 1220 1221 -- Fat Pointers 1222 1223 -- Fat pointers are represented as a struct with two fields. This 1224 -- struct has two distinguished field names: 1225 1226 -- P_ARRAY is a pointer to the array type. The name of this type is 1227 -- the unconstrained type followed by "___XUA". This array will have 1228 -- bounds which are the discriminants, and hence are unparsable, but 1229 -- will give the number of subscripts and the component type. 1230 1231 -- P_BOUNDS is a pointer to a struct, the name of whose type is the 1232 -- unconstrained array name followed by "___XUB" and which has 1233 -- fields of the form 1234 1235 -- LBn (n a decimal integer) lower bound of n'th dimension 1236 -- UBn (n a decimal integer) upper bound of n'th dimension 1237 1238 -- The bounds may be any integral type. In the case of an enumeration 1239 -- type, Enum_Rep values are used. 1240 1241 -- For a given unconstrained array type, the compiler will generate one 1242 -- fat-pointer type whose name is "arr___XUP", where "arr" is the name 1243 -- of the array type, and use it to represent the array type itself in 1244 -- the debugging information. 1245 1246 -- For each pointer to this unconstrained array type, the compiler will 1247 -- generate a typedef that points to the above "arr___XUP" fat-pointer 1248 -- type. As a consequence, when it comes to fat-pointer types: 1249 1250 -- 1. The type name is given by the typedef 1251 1252 -- 2. If the debugger is asked to output the type, the appropriate 1253 -- form is "access arr", except if the type name is "arr___XUP" 1254 -- for which it is the array definition. 1255 1256 -- Thin Pointers 1257 1258 -- The value of a thin pointer is a pointer to the second field of a 1259 -- structure with two fields. The name of this structure's type is 1260 -- "arr___XUT", where "arr" is the name of the unconstrained array 1261 -- type. Even though it actually points into middle of this structure, 1262 -- the thin pointer's type in debugging information is 1263 -- pointer-to-arr___XUT. 1264 1265 -- The first field of arr___XUT is named BOUNDS, and has a type named 1266 -- arr___XUB, with the structure described for such types in fat 1267 -- pointers, as described above. 1268 1269 -- The second field of arr___XUT is named ARRAY, and contains the 1270 -- actual array. Because this array has a dynamic size, determined by 1271 -- the BOUNDS field that precedes it, all of the information about 1272 -- arr___XUT is encoded in a parallel type named arr___XUT___XVE, with 1273 -- fields BOUNDS and ARRAY___XVL. As for previously described ___XVE 1274 -- types, ARRAY___XVL has a pointer-to-array type. However, the array 1275 -- type in this case is named arr___XUA and only its element type is 1276 -- meaningful, just as described for fat pointers. 1277 1278 -------------------------------------- 1279 -- Tagged Types and Type Extensions -- 1280 -------------------------------------- 1281 1282 -- A type C derived from a tagged type P has a field named "_parent" of 1283 -- type P that contains its inherited fields. The type of this field is 1284 -- usually P (encoded as usual if it has a dynamic size), but may be a more 1285 -- distant ancestor, if P is a null extension of that type. 1286 1287 -- The type tag of a tagged type is a field named _tag, of type void*. If 1288 -- the type is derived from another tagged type, its _tag field is found in 1289 -- its _parent field. 1290 1291 ----------------------------- 1292 -- Variant Record Encoding -- 1293 ----------------------------- 1294 1295 -- The variant part of a variant record is encoded as a single field in the 1296 -- enclosing record, whose name is: 1297 1298 -- discrim___XVN 1299 1300 -- where discrim is the unqualified name of the variant. This field name is 1301 -- built by gigi (not by code in this unit). For Unchecked_Union record, 1302 -- this discriminant will not appear in the record (see Unchecked Unions, 1303 -- below). 1304 1305 -- The type corresponding to this field has a name that is obtained by 1306 -- concatenating the type name with the above string and is similar to a C 1307 -- union, in which each member of the union corresponds to one variant. 1308 -- However, unlike a C union, the size of the type may be variable even if 1309 -- each of the components are fixed size, since it includes a computation 1310 -- of which variant is present. In that case, it will be encoded as above 1311 -- and a type with the suffix "___XVN___XVU" will be present. 1312 1313 -- The name of the union member is encoded to indicate the choices, and 1314 -- is a string given by the following grammar: 1315 1316 -- member_name ::= {choice} | others_choice 1317 -- choice ::= simple_choice | range_choice 1318 -- simple_choice ::= S number 1319 -- range_choice ::= R number T number 1320 -- number ::= {decimal_digit} [m] 1321 -- others_choice ::= O (upper case letter O) 1322 1323 -- The m in a number indicates a negative value. As an example of this 1324 -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by 1325 1326 -- R1T4S7S10m 1327 1328 -- In the case of enumeration values, the values used are the actual 1329 -- representation values in the case where an enumeration type has an 1330 -- enumeration representation spec (i.e. they are values that correspond 1331 -- to the use of the Enum_Rep attribute). 1332 1333 -- The type of the inner record is given by the name of the union type (as 1334 -- above) concatenated with the above string. Since that type may itself be 1335 -- variable-sized, it may also be encoded as above with a new type with a 1336 -- further suffix of "___XVU". 1337 1338 -- As an example, consider: 1339 1340 -- type Var (Disc : Boolean := True) is record 1341 -- M : Integer; 1342 1343 -- case Disc is 1344 -- when True => 1345 -- R : Integer; 1346 -- S : Integer; 1347 1348 -- when False => 1349 -- T : Integer; 1350 -- end case; 1351 -- end record; 1352 1353 -- V1 : Var; 1354 1355 -- In this case, the type var is represented as a struct with three fields. 1356 -- The first two are "disc" and "m", representing the values of these 1357 -- record components. The third field is a union of two types, with field 1358 -- names S1 and O. S1 is a struct with fields "r" and "s", and O is a 1359 -- struct with field "t". 1360 1361 ---------------------- 1362 -- Unchecked Unions -- 1363 ---------------------- 1364 1365 -- The encoding for variant records changes somewhat under the influence 1366 -- of a "pragma Unchecked_Union" clause: 1367 1368 -- 1. The discriminant will not be present in the record, although its 1369 -- name is still used in the encodings. 1370 -- 2. Variants containing a single component named "x" of type "T" may 1371 -- be encoded, as in ordinary C unions, as a single field of the 1372 -- enclosing union type named "x" of type "T", dispensing with the 1373 -- enclosing struct. In this case, of course, the discriminant values 1374 -- corresponding to the variant are unavailable. As for normal 1375 -- variants, the field name "x" may be suffixed with ___XVL if it 1376 -- has dynamic size. 1377 1378 -- For example, the type Var in the preceding section, if followed by 1379 -- "pragma Unchecked_Union (Var);" may be encoded as a struct with two 1380 -- fields. The first is "m". The second field is a union of two types, 1381 -- with field names S1 and "t". As before, S1 is a struct with fields 1382 -- "r" and "s". "t" is a field of type Integer. 1383 1384 ------------------------------------------------ 1385 -- Subprograms for Handling Variant Encodings -- 1386 ------------------------------------------------ 1387 1388 procedure Get_Variant_Encoding (V : Node_Id); 1389 -- This procedure is called by Gigi with V being the variant node. The 1390 -- corresponding encoding string is returned in Name_Buffer with the length 1391 -- of the string in Name_Len, and an ASCII.NUL character stored following 1392 -- the name. 1393 1394 --------------------------------- 1395 -- Subtypes of Variant Records -- 1396 --------------------------------- 1397 1398 -- A subtype of a variant record is represented by a type in which the 1399 -- union field from the base type is replaced by one of the possible 1400 -- values. For example, if we have: 1401 1402 -- type Var (Disc : Boolean := True) is record 1403 -- M : Integer; 1404 1405 -- case Disc is 1406 -- when True => 1407 -- R : Integer; 1408 -- S : Integer; 1409 1410 -- when False => 1411 -- T : Integer; 1412 -- end case; 1413 1414 -- end record; 1415 -- V1 : Var; 1416 -- V2 : Var (True); 1417 -- V3 : Var (False); 1418 1419 -- Here V2, for example, is represented with a subtype whose name is 1420 -- something like TvarS3b, which is a struct with three fields. The first 1421 -- two fields are "disc" and "m" as for the base type, and the third field 1422 -- is S1, which contains the fields "r" and "s". 1423 1424 -- The debugger should simply ignore structs with names of the form 1425 -- corresponding to variants, and consider the fields inside as belonging 1426 -- to the containing record. 1427 1428 ----------------------------------------------- 1429 -- Extra renamings for subprogram instances -- 1430 ----------------------------------------------- 1431 1432 procedure Build_Subprogram_Instance_Renamings 1433 (N : Node_Id; 1434 Wrapper : Entity_Id); 1435 -- The debugger has difficulties in recovering the value of actuals of an 1436 -- elementary type, from within the body of a subprogram instantiation. 1437 -- This is because such actuals generate an object declaration that is 1438 -- placed within the wrapper package of the instance, and the entity in 1439 -- these declarations is encoded in a complex way that GDB does not handle 1440 -- well. These new renaming declarations appear within the body of the 1441 -- subprogram, and are redundant from a visibility point of view, but They 1442 -- should have no measurable performance impact, and require no special 1443 -- decoding in the debugger. 1444 1445 ------------------------------------------- 1446 -- Character literals in Character Types -- 1447 ------------------------------------------- 1448 1449 -- Character types are enumeration types at least one of whose enumeration 1450 -- literals is a character literal. Enumeration literals are usually simply 1451 -- represented using their identifier names. If the enumeration literal is 1452 -- a character literal, the name is encoded as described in the following 1453 -- paragraph. 1454 1455 -- A name QUhh, where each 'h' is a lower-case hexadecimal digit, stands 1456 -- for a character whose Unicode encoding is hh, and QWhhhh likewise stands 1457 -- for a wide character whose encoding is hhhh. The representation values 1458 -- are encoded as for ordinary enumeration literals (and have no necessary 1459 -- relationship to the values encoded in the names). 1460 1461 -- For example, given the type declaration 1462 1463 -- type x is (A, 'C', B); 1464 1465 -- the second enumeration literal would be named QU43 and the value 1466 -- assigned to it would be 1. 1467 1468 ----------------------------------------------- 1469 -- Secondary Dispatch tables of tagged types -- 1470 ----------------------------------------------- 1471 1472 procedure Get_Secondary_DT_External_Name 1473 (Typ : Entity_Id; 1474 Ancestor_Typ : Entity_Id; 1475 Suffix_Index : Int); 1476 -- Set Name_Buffer and Name_Len to the external name of one secondary 1477 -- dispatch table of Typ. If the interface has been inherited from some 1478 -- ancestor then Ancestor_Typ is such node (in this case the secondary DT 1479 -- is needed to handle overridden primitives); if there is no such ancestor 1480 -- then Ancestor_Typ is equal to Typ. 1481 -- 1482 -- Internal rule followed for the generation of the external name: 1483 -- 1484 -- Case 1. If the secondary dispatch has not been inherited from some 1485 -- ancestor of Typ then the external name is composed as 1486 -- follows: 1487 -- External_Name (Typ) + Suffix_Number + 'P' 1488 -- 1489 -- Case 2. if the secondary dispatch table has been inherited from some 1490 -- ancestor then the external name is composed as follows: 1491 -- External_Name (Typ) + '_' + External_Name (Ancestor_Typ) 1492 -- + Suffix_Number + 'P' 1493 -- 1494 -- Note: We have to use the external names (instead of simply their names) 1495 -- to protect the frontend against programs that give the same name to all 1496 -- the interfaces and use the expanded name to reference them. The 1497 -- Suffix_Number is used to differentiate all the secondary dispatch 1498 -- tables of a given type. 1499 -- 1500 -- Examples: 1501 -- 1502 -- package Pkg1 is | package Pkg2 is | package Pkg3 is 1503 -- type Typ is | type Typ is | type Typ is 1504 -- interface; | interface; | interface; 1505 -- end Pkg1; | end Pkg; | end Pkg3; 1506 -- 1507 -- with Pkg1, Pkg2, Pkg3; 1508 -- package Case_1 is 1509 -- type Typ is new Pkg1.Typ and Pkg2.Typ and Pkg3.Typ with ... 1510 -- end Case_1; 1511 -- 1512 -- with Case_1; 1513 -- package Case_2 is 1514 -- type Typ is new Case_1.Typ with ... 1515 -- end Case_2; 1516 -- 1517 -- These are the external names generated for Case_1.Typ (note that 1518 -- Pkg1.Typ is associated with the Primary Dispatch Table, because it 1519 -- is the parent of this type, and hence no external name is 1520 -- generated for it). 1521 -- case_1__typ0P (associated with Pkg2.Typ) 1522 -- case_1__typ1P (associated with Pkg3.Typ) 1523 -- 1524 -- These are the external names generated for Case_2.Typ: 1525 -- case_2__typ_case_1__typ0P 1526 -- case_2__typ_case_1__typ1P 1527 1528 ---------------------------- 1529 -- Effect of Optimization -- 1530 ---------------------------- 1531 1532 -- If the program is compiled with optimization on (e.g. -O1 switch 1533 -- specified), then there may be variations in the output from the above 1534 -- specification. In particular, objects may disappear from the output. 1535 -- This includes not only constants and variables that the program declares 1536 -- at the source level, but also the x___L and x___U constants created to 1537 -- describe the lower and upper bounds of subtypes with dynamic bounds. 1538 -- This means for example, that array bounds may disappear if optimization 1539 -- is turned on. The debugger is expected to recognize that these constants 1540 -- are missing and deal as best as it can with the limited information 1541 -- available. 1542 1543 --------------------------------- 1544 -- GNAT Extensions to DWARF2/3 -- 1545 --------------------------------- 1546 1547 -- If the compiler switch "-gdwarf+" is specified, GNAT Vendor extensions 1548 -- to DWARF2/3 are generated, with the following variations from the above 1549 -- specification. 1550 1551 -- Change in the contents of the DW_AT_name attribute 1552 1553 -- The operators are represented in their natural form. (for example, 1554 -- the addition operator is written as "+" instead of "Oadd"). The 1555 -- component separator is "." instead of "__" 1556 1557 -- Introduction of DW_AT_GNAT_encoding, encoded with value 0x2301 1558 1559 -- Any debugging information entry representing a program entity, named 1560 -- or implicit, may have a DW_AT_GNAT_encoding attribute. The value of 1561 -- this attribute is a string representing the suffix internally added 1562 -- by GNAT for various purposes, mainly for representing debug 1563 -- information compatible with other formats. In particular this is 1564 -- useful for IDEs which need to filter out information internal to 1565 -- GNAT from their graphical interfaces. 1566 1567 -- If a debugging information entry has multiple encodings, all of them 1568 -- will be listed in DW_AT_GNAT_encoding using the list separator ':'. 1569 1570 -- Introduction of DW_AT_GNAT_descriptive_type, encoded with value 0x2302 1571 1572 -- Any debugging information entry representing a type may have a 1573 -- DW_AT_GNAT_descriptive_type attribute whose value is a reference, 1574 -- pointing to a debugging information entry representing another type 1575 -- associated to the type. 1576 1577 -- Modification of the contents of the DW_AT_producer string 1578 1579 -- When emitting full GNAT Vendor extensions to DWARF2/3, "-gdwarf+" 1580 -- is appended to the DW_AT_producer string. 1581 -- 1582 -- When emitting only DW_AT_GNAT_descriptive_type, "-gdwarf+-" is 1583 -- appended to the DW_AT_producer string. 1584 1585end Exp_Dbug; 1586