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) 1996-2020, 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 -- WARNING: There is a matching C declaration of this subprogram in fe.h 445 446 ------------------------------------- 447 -- Encoding for translation into C -- 448 ------------------------------------- 449 450 -- In Modify_Tree_For_C mode we must add encodings to dismabiguate cases 451 -- where Ada block structure cannot be directly translated. These cases 452 -- are as follows: 453 454 -- a) A loop variable may hide a homonym in an enclosing block 455 -- b) A block-local variable may hide a homonym in an enclosing block 456 457 -- In C these constructs are not scopes and we must distinguish the names 458 -- explicitly. In the first case we create a qualified name with the suffix 459 -- 'L', in the second case with a suffix 'B'. 460 461 -------------------------------------------- 462 -- Subprograms for Handling Qualification -- 463 -------------------------------------------- 464 465 procedure Qualify_Entity_Names (N : Node_Id); 466 -- Given a node N, that represents a block, subprogram body, or package 467 -- body or spec, or protected or task type, sets a fully qualified name 468 -- for the defining entity of given construct, and also sets fully 469 -- qualified names for all enclosed entities of the construct (using 470 -- First_Entity/Next_Entity). Note that the actual modifications of the 471 -- names is postponed till a subsequent call to Qualify_All_Entity_Names. 472 -- Note: this routine does not deal with prepending _ada_ to library 473 -- subprogram names. The reason for this is that we only prepend _ada_ 474 -- to the library entity itself, and not to names built from this name. 475 476 procedure Qualify_All_Entity_Names; 477 -- When Qualify_Entity_Names is called, no actual name changes are made, 478 -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call 479 -- is made to this procedure. The reason for this deferral is that when 480 -- names are changed semantic processing may be affected. By deferring 481 -- the changes till just before gigi is called, we avoid any concerns 482 -- about such effects. Gigi itself does not use the names except for 483 -- output of names for debugging purposes (which is why we are doing 484 -- the name changes in the first place). 485 486 -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are 487 -- useful to remove qualification from a name qualified by the call to 488 -- Qualify_All_Entity_Names. 489 490 -------------------------------- 491 -- Handling of Numeric Values -- 492 -------------------------------- 493 494 -- All numeric values here are encoded as strings of decimal digits. Only 495 -- integer values need to be encoded. A negative value is encoded as the 496 -- corresponding positive value followed by a lower case m for minus to 497 -- indicate that the value is negative (e.g. 2m for -2). 498 499 ------------------------- 500 -- Type Name Encodings -- 501 ------------------------- 502 503 -- In the following typ is the name of the type as normally encoded by the 504 -- debugger rules, i.e. a non-qualified name, all in lower case, with 505 -- standard encoding of upper half and wide characters 506 507 ------------------------ 508 -- Encapsulated Types -- 509 ------------------------ 510 511 -- In some cases, the compiler encapsulates a type by wrapping it in a 512 -- structure. For example, this is used when a size or alignment 513 -- specification requires a larger type. Consider: 514 515 -- type y is mod 2 ** 64; 516 -- for y'size use 256; 517 518 -- In this case the compile generates a structure type y___PAD, which 519 -- has a single field whose name is F. This single field is 64 bits 520 -- long and contains the actual value. This kind of padding is used 521 -- when the logical value to be stored is shorter than the object in 522 -- which it is allocated. For example if a size clause is used to set 523 -- a size of 256 for a signed integer value, then a typical choice is 524 -- to wrap a 64-bit integer in a 256 bit PAD structure. 525 526 -- A similar encapsulation is done for some packed array types, in which 527 -- case the structure type is y___JM and the field name is OBJECT. 528 -- This is used in the case of a packed array stored using modular 529 -- representation (see section on representation of packed array 530 -- objects). In this case the JM wrapping is used to achieve correct 531 -- positioning of the packed array value (left or right justified in its 532 -- field depending on endianness. 533 534 -- When the debugger sees an object of a type whose name has a suffix of 535 -- ___PAD or ___JM, the type will be a record containing a single field, 536 -- and the name of that field will be all upper case. In this case, it 537 -- should look inside to get the value of the inner field, and neither 538 -- the outer structure name, nor the field name should appear when the 539 -- value is printed. 540 541 -- When the debugger sees a record named REP being a field inside 542 -- another record, it should treat the fields inside REP as being part 543 -- of the outer record (this REP field is only present for code 544 -- generation purposes). The REP record should not appear in the values 545 -- printed by the debugger. 546 547 ----------------------- 548 -- Fixed-Point Types -- 549 ----------------------- 550 551 -- Fixed-point types are encoded using a suffix that indicates the 552 -- delta and small values. The actual type itself is a normal integer 553 -- type. 554 555 -- typ___XF_nn_dd 556 -- typ___XF_nn_dd_nn_dd 557 558 -- The first form is used when small = delta. The value of delta (and 559 -- small) is given by the rational nn/dd, where nn and dd are decimal 560 -- integers. 561 -- 562 -- The second form is used if the small value is different from the 563 -- delta. In this case, the first nn/dd rational value is for delta, 564 -- and the second value is for small. 565 566 -------------------- 567 -- Discrete Types -- 568 -------------------- 569 570 -- Discrete types are coded with a suffix indicating the range in the 571 -- case where one or both of the bounds are discriminants or variable. 572 573 -- Note: at the current time, we also encode compile time known bounds 574 -- if they do not match the natural machine type bounds, but this may 575 -- be removed in the future, since it is redundant for most debugging 576 -- formats. However, we do not ever need XD encoding for enumeration 577 -- base types, since here it is always clear what the bounds are from 578 -- the total number of enumeration literals. 579 580 -- typ___XD 581 -- typ___XDL_lowerbound 582 -- typ___XDU_upperbound 583 -- typ___XDLU_lowerbound__upperbound 584 585 -- If a discrete type is a natural machine type (i.e. its bounds 586 -- correspond in a natural manner to its size), then it is left 587 -- unencoded. The above encoding forms are used when there is a 588 -- constrained range that does not correspond to the size or that 589 -- has discriminant references or other compile time known bounds. 590 591 -- The first form is used if both bounds are dynamic, in which case two 592 -- constant objects are present whose names are typ___L and typ___U in 593 -- the same scope as typ, and the values of these constants indicate 594 -- the bounds. As far as the debugger is concerned, these are simply 595 -- variables that can be accessed like any other variables. In the 596 -- enumeration case, these values correspond to the Enum_Rep values for 597 -- the lower and upper bounds. 598 599 -- The second form is used if the upper bound is dynamic, but the lower 600 -- bound is either constant or depends on a discriminant of the record 601 -- with which the type is associated. The upper bound is stored in a 602 -- constant object of name typ___U as previously described, but the 603 -- lower bound is encoded directly into the name as either a decimal 604 -- integer, or as the discriminant name. 605 606 -- The third form is similarly used if the lower bound is dynamic, but 607 -- the upper bound is compile time known or a discriminant reference, 608 -- in which case the lower bound is stored in a constant object of name 609 -- typ___L, and the upper bound is encoded directly into the name as 610 -- either a decimal integer, or as the discriminant name. 611 612 -- The fourth form is used if both bounds are discriminant references 613 -- or compile time known values, with the encoding first for the lower 614 -- bound, then for the upper bound, as previously described. 615 616 ------------------- 617 -- Modular Types -- 618 ------------------- 619 620 -- A type declared 621 622 -- type x is mod N; 623 624 -- Is encoded as a subrange of an unsigned base type with lower bound 625 -- zero and upper bound N. That is, there is no name encoding. We use 626 -- the standard encodings provided by the debugging format. Thus we 627 -- give these types a non-standard interpretation: the standard 628 -- interpretation of our encoding would not, in general, imply that 629 -- arithmetic on type x was to be performed modulo N (especially not 630 -- when N is not a power of 2). 631 632 ------------------ 633 -- Biased Types -- 634 ------------------ 635 636 -- Only discrete types can be biased, and the fact that they are biased 637 -- is indicated by a suffix of the form: 638 639 -- typ___XB_lowerbound__upperbound 640 641 -- Here lowerbound and upperbound are decimal integers, with the usual 642 -- (postfix "m") encoding for negative numbers. Biased types are only 643 -- possible where the bounds are compile time known, and the values are 644 -- represented as unsigned offsets from the lower bound given. For 645 -- example: 646 647 -- type Q is range 10 .. 15; 648 -- for Q'size use 3; 649 650 -- The size clause will force values of type Q in memory to be stored 651 -- in biased form (e.g. 11 will be represented by the bit pattern 001). 652 653 ---------------------------------------------- 654 -- Record Types with Variable-Length Fields -- 655 ---------------------------------------------- 656 657 -- The debugging formats do not fully support these types, and indeed 658 -- some formats simply generate no useful information at all for such 659 -- types. In order to provide information for the debugger, gigi creates 660 -- a parallel type in the same scope with one of the names 661 662 -- type___XVE 663 -- type___XVU 664 665 -- The former name is used for a record and the latter for the union 666 -- that is made for a variant record (see below) if that record or union 667 -- has a field of variable size or if the record or union itself has a 668 -- variable size. These encodings suffix any other encodings that that 669 -- might be suffixed to the type name. 670 671 -- The idea here is to provide all the needed information to interpret 672 -- objects of the original type in the form of a "fixed up" type, which 673 -- is representable using the normal debugging information. 674 675 -- There are three cases to be dealt with. First, some fields may have 676 -- variable positions because they appear after variable-length fields. 677 -- To deal with this, we encode *all* the field bit positions of the 678 -- special ___XV type in a non-standard manner. 679 680 -- The idea is to encode not the position, but rather information that 681 -- allows computing the position of a field from the position of the 682 -- previous field. The algorithm for computing the actual positions of 683 -- all fields and the length of the record is as follows. In this 684 -- description, let P represent the current bit position in the record. 685 686 -- 1. Initialize P to 0 687 688 -- 2. For each field in the record: 689 690 -- 2a. If an alignment is given (see below), then round P up, if 691 -- needed, to the next multiple of that alignment. 692 693 -- 2b. If a bit position is given, then increment P by that amount 694 -- (that is, treat it as an offset from the end of the preceding 695 -- record). 696 697 -- 2c. Assign P as the actual position of the field 698 699 -- 2d. Compute the length, L, of the represented field (see below) 700 -- and compute P'=P+L. Unless the field represents a variant part 701 -- (see below and also Variant Record Encoding), set P to P'. 702 703 -- The alignment, if present, is encoded in the field name of the 704 -- record, which has a suffix: 705 706 -- fieldname___XVAnn 707 708 -- where the nn after the XVA indicates the alignment value in storage 709 -- units. This encoding is present only if an alignment is present. 710 711 -- The size of the record described by an XVE-encoded type (in bits) is 712 -- generally the maximum value attained by P' in step 2d above, rounded 713 -- up according to the record's alignment. 714 715 -- Second, the variable-length fields themselves are represented by 716 -- replacing the type by a special access type. The designated type of 717 -- this access type is the original variable-length type, and the fact 718 -- that this field has been transformed in this way is signalled by 719 -- encoding the field name as: 720 721 -- field___XVL 722 723 -- where field is the original field name. If a field is both 724 -- variable-length and also needs an alignment encoding, then the 725 -- encodings are combined using: 726 727 -- field___XVLnn 728 729 -- Note: the reason that we change the type is so that the resulting 730 -- type has no variable-length fields. At least some of the formats used 731 -- for debugging information simply cannot tolerate variable- length 732 -- fields, so the encoded information would get lost. 733 734 -- Third, in the case of a variant record, the special union that 735 -- contains the variants is replaced by a normal C union. In this case, 736 -- the positions are all zero. 737 738 -- Discriminants appear before any variable-length fields that depend on 739 -- them, with one exception. In some cases, a discriminant governing the 740 -- choice of a variant clause may appear in the list of fields of an XVE 741 -- type after the entry for the variant clause itself (this can happen 742 -- in the presence of a representation clause for the record type in the 743 -- source program). However, when this happens, the discriminant's 744 -- position may be determined by first applying the rules described in 745 -- this section, ignoring the variant clause. As a result, discriminants 746 -- can always be located independently of the variable-length fields 747 -- that depend on them. 748 749 -- The size of the ___XVE or ___XVU record or union is set to the 750 -- alignment (in bytes) of the original object so that the debugger 751 -- can calculate the size of the original type. 752 753 -- As an example of this encoding, consider the declarations: 754 755 -- type Q is array (1 .. V1) of Float; -- alignment 4 756 -- type R is array (1 .. V2) of Long_Float; -- alignment 8 757 758 -- type X is record 759 -- A : Character; 760 -- B : Float; 761 -- C : String (1 .. V3); 762 -- D : Float; 763 -- E : Q; 764 -- F : R; 765 -- G : Float; 766 -- end record; 767 768 -- The encoded type looks like: 769 770 -- type anonymousQ is access Q; 771 -- type anonymousR is access R; 772 773 -- type X___XVE is record 774 -- A : Character; -- position contains 0 775 -- B : Float; -- position contains 24 776 -- C___XVL : access String (1 .. V3); -- position contains 0 777 -- D___XVA4 : Float; -- position contains 0 778 -- E___XVL4 : anonymousQ; -- position contains 0 779 -- F___XVL8 : anonymousR; -- position contains 0 780 -- G : Float; -- position contains 0 781 -- end record; 782 783 -- Any bit sizes recorded for fields other than dynamic fields and 784 -- variants are honored as for ordinary records. 785 786 -- Notes: 787 788 -- 1) The B field could also have been encoded by using a position of 789 -- zero and an alignment of 4, but in such a case the coding by position 790 -- is preferred (since it takes up less space). We have used the 791 -- (illegal) notation access xxx as field types in the example above. 792 793 -- 2) The E field does not actually need the alignment indication but 794 -- this may not be detected in this case by the conversion routines. 795 796 -- 3) Our conventions do not cover all XVE-encoded records in which 797 -- some, but not all, fields have representation clauses. Such records 798 -- may, therefore, be displayed incorrectly by debuggers. This situation 799 -- is not common. 800 801 ----------------------- 802 -- Base Record Types -- 803 ----------------------- 804 805 -- Under certain circumstances, debuggers need two descriptions of a 806 -- record type, one that gives the actual details of the base type's 807 -- structure (as described elsewhere in these comments) and one that may 808 -- be used to obtain information about the particular subtype and the 809 -- size of the objects being typed. In such cases the compiler will 810 -- substitute type whose name is typically compiler-generated and 811 -- irrelevant except as a key for obtaining the actual type. 812 813 -- Specifically, if this name is x, then we produce a record type named 814 -- x___XVS consisting of one field. The name of this field is that of 815 -- the actual type being encoded, which we'll call y. The type of this 816 -- single field can be either an arbitrary non-reference type, e.g. an 817 -- integer type, or a reference type; in the latter case, the referenced 818 -- type is also the actual type being encoded y. Both x and y may have 819 -- corresponding ___XVE types. 820 821 -- The size of the objects typed as x should be obtained from the 822 -- structure of x (and x___XVE, if applicable) as for ordinary types 823 -- unless there is a variable named x___XVZ, which, if present, will 824 -- hold the size (in bytes) of x. In this latter case, the size of the 825 -- x___XVS type will not be a constant but a reference to x___XVZ. 826 827 -- The type x will either be a subtype of y (see also Subtypes of 828 -- Variant Records, below) or will contain a single field of type y, 829 -- or no fields at all. The layout, types, and positions of these 830 -- fields will be accurate, if present. (Currently, however, the GDB 831 -- debugger makes no use of x except to determine its size). 832 833 -- Among other uses, XVS types are used to encode unconstrained types. 834 -- For example, given: 835 -- 836 -- subtype Int is INTEGER range 0..10; 837 -- type T1 (N: Int := 0) is record 838 -- F1: String (1 .. N); 839 -- end record; 840 -- type AT1 is array (INTEGER range <>) of T1; 841 -- 842 -- the element type for AT1 might have a type defined as if it had 843 -- been written: 844 -- 845 -- type at1___PAD is record F : T1; end record; 846 -- for at1___PAD'Size use 16 * 8; 847 -- 848 -- and there would also be: 849 -- 850 -- type at1___PAD___XVS is record t1: reft1; end record; 851 -- type t1 is ... 852 -- type reft1 is <reference to t1> 853 -- 854 -- Had the subtype Int been dynamic: 855 -- 856 -- subtype Int is INTEGER range 0 .. M; -- M a variable 857 -- 858 -- Then the compiler would also generate a declaration whose effect 859 -- would be 860 -- 861 -- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term; 862 -- 863 -- Not all unconstrained types are so encoded; the XVS convention may be 864 -- unnecessary for unconstrained types of fixed size. However, this 865 -- encoding is always necessary when a subcomponent type (array 866 -- element's type or record field's type) is an unconstrained record 867 -- type some of whose components depend on discriminant values. 868 869 ----------------- 870 -- Array Types -- 871 ----------------- 872 873 -- Since there is no way for the debugger to obtain the index subtypes 874 -- for an array type, we produce a type that has the name of the array 875 -- type followed by "___XA" and is a record type whose field types are 876 -- the respective types for the bounds (and whose field names are the 877 -- names of these types). 878 879 -- To conserve space, we do not produce this type unless one of the 880 -- index types is either an enumeration type, has a variable lower or 881 -- upper bound or is a biased type. 882 883 -- Given the full encoding of these types (see above description for 884 -- the encoding of discrete types), this means that all necessary 885 -- information for addressing arrays is available. In some debugging 886 -- formats, some or all of the bounds information may be available 887 -- redundantly, particularly in the fixed-point case, but this 888 -- information can in any case be ignored by the debugger. 889 890 ---------------------------- 891 -- Note on Implicit Types -- 892 ---------------------------- 893 894 -- The compiler creates implicit type names in many situations where a 895 -- type is present semantically, but no specific name is present. For 896 -- example: 897 898 -- S : Integer range M .. N; 899 900 -- Here the subtype of S is not integer, but rather an anonymous subtype 901 -- of Integer. Where possible, the compiler generates names for such 902 -- anonymous types that are related to the type from which the subtype 903 -- is obtained as follows: 904 905 -- T name suffix 906 907 -- where name is the name from which the subtype is obtained, using 908 -- lower case letters and underscores, and suffix starts with an upper 909 -- case letter. For example the name for the above declaration might be: 910 911 -- TintegerS4b 912 913 -- If the debugger is asked to give the type of an entity and the type 914 -- has the form T name suffix, it is probably appropriate to just use 915 -- "name" in the response since this is what is meaningful to the 916 -- programmer. 917 918 ------------------------------------------------- 919 -- Subprograms for Handling Encoded Type Names -- 920 ------------------------------------------------- 921 922 procedure Get_Encoded_Name (E : Entity_Id); 923 -- If the entity is a typename, store the external name of the entity as in 924 -- Get_External_Name, followed by three underscores plus the type encoding 925 -- in Name_Buffer with the length in Name_Len, and an ASCII.NUL character 926 -- stored following the name. Otherwise set Name_Buffer and Name_Len to 927 -- hold the entity name. Note that a call to this procedure has no effect 928 -- if we are not generating code, since the necessary information for 929 -- computing the proper encoded name is not available in this case. 930 931 -- WARNING: There is a matching C declaration of this subprogram in fe.h 932 933 -------------- 934 -- Renaming -- 935 -------------- 936 937 -- Debugging information is generated for exception, object, package, and 938 -- subprogram renaming (generic renamings are not significant, since 939 -- generic templates are not relevant at debugging time). 940 941 -- Consider a renaming declaration of the form 942 943 -- x : typ renames y; 944 945 -- There is one case in which no special debugging information is required, 946 -- namely the case of an object renaming where the back end allocates a 947 -- reference for the renamed variable, and the entity x is this reference. 948 -- The debugger can handle this case without any special processing or 949 -- encoding (it won't know it was a renaming, but that does not matter). 950 951 -- All other cases of renaming generate a dummy variable for an entity 952 -- whose name is of the form: 953 954 -- x___XR_... for an object renaming 955 -- x___XRE_... for an exception renaming 956 -- x___XRP_... for a package renaming 957 958 -- and where the "..." represents a suffix that describes the structure of 959 -- the object name given in the renaming (see details below). 960 961 -- The name is fully qualified in the usual manner, i.e. qualified in the 962 -- same manner as the entity x would be. In the case of a package renaming 963 -- where x is a child unit, the qualification includes the name of the 964 -- parent unit, to disambiguate child units with the same simple name and 965 -- (of necessity) different parents. 966 967 -- Note: subprogram renamings are not encoded at the present time 968 969 -- The suffix of the variable name describing the renamed object is defined 970 -- to use the following encoding: 971 972 -- For the simple entity case, where y is just an entity name, the suffix 973 -- is of the form: 974 975 -- y___XE 976 977 -- i.e. the suffix has a single field, the first part matching the 978 -- name y, followed by a "___" separator, ending with sequence XE. 979 -- The entity name portion is fully qualified in the usual manner. 980 -- This same naming scheme is followed for all forms of encoded 981 -- renamings that rename a simple entity. 982 983 -- For the object renaming case where y is a selected component or an 984 -- indexed component, the variable name is suffixed by additional fields 985 -- that give details of the components. The name starts as above with a 986 -- y___XE name indicating the outer level object entity. Then a series of 987 -- selections and indexing operations can be specified as follows: 988 989 -- Indexed component 990 991 -- A series of subscript values appear in sequence, the number 992 -- corresponds to the number of dimensions of the array. The 993 -- subscripts have one of the following two forms: 994 995 -- XSnnn 996 997 -- Here nnn is a constant value, encoded as a decimal integer 998 -- (pos value for enumeration type case). Negative values have 999 -- a trailing 'm' as usual. 1000 1001 -- XSe 1002 1003 -- Here e is the (unqualified) name of a constant entity in the 1004 -- same scope as the renaming which contains the subscript value. 1005 1006 -- Slice 1007 1008 -- For the slice case, we have two entries. The first is for the 1009 -- lower bound of the slice, and has the form: 1010 1011 -- XLnnn 1012 -- XLe 1013 1014 -- Specifies the lower bound, using exactly the same encoding as 1015 -- for an XS subscript as described above. 1016 1017 -- Then the upper bound appears in the usual XSnnn/XSe form 1018 1019 -- Selected component 1020 1021 -- For a selected component, we have a single entry 1022 1023 -- XRf 1024 1025 -- Here f is the field name for the selection 1026 1027 -- For an explicit dereference (.all), we have a single entry 1028 1029 -- XA 1030 1031 -- As an example, consider the declarations: 1032 1033 -- package p is 1034 -- type q is record 1035 -- m : string (2 .. 5); 1036 -- end record; 1037 -- 1038 -- type r is array (1 .. 10, 1 .. 20) of q; 1039 -- 1040 -- g : r; 1041 -- 1042 -- z : string renames g (1,5).m(2 ..3) 1043 -- end p; 1044 1045 -- The generated variable entity would appear as 1046 1047 -- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type; 1048 -- p__g___XE--------------------outer entity is g 1049 -- XS1-----------------first subscript for g 1050 -- XS5--------------second subscript for g 1051 -- XRm-----------select field m 1052 -- XL2--------lower bound of slice 1053 -- XS3-----upper bound of slice 1054 1055 -- Note that the type of the variable is a special internal type named 1056 -- _renaming_type. This type is an arbitrary type of zero size created 1057 -- in package Standard (see cstand.adb) and is ignored by the debugger. 1058 1059 function Debug_Renaming_Declaration (N : Node_Id) return Node_Id; 1060 -- The argument N is a renaming declaration. The result is a variable 1061 -- declaration as described in the above paragraphs. If N is not a special 1062 -- debug declaration, then Empty is returned. This function also takes care 1063 -- of setting Materialize_Entity on the renamed entity where required. 1064 1065 --------------------------- 1066 -- Packed Array Encoding -- 1067 --------------------------- 1068 1069 -- For every constrained packed array, two types are created, and both 1070 -- appear in the debugging output: 1071 1072 -- The original declared array type is a perfectly normal array type, and 1073 -- its index bounds indicate the original bounds of the array. 1074 1075 -- The corresponding packed array type, which may be a modular type, or 1076 -- may be an array of bytes type (see Exp_Pakd for full details). This is 1077 -- the type that is actually used in the generated code and for debugging 1078 -- information for all objects of the packed type. 1079 1080 -- The name of the corresponding packed array type is: 1081 1082 -- ttt___XPnnn 1083 1084 -- where 1085 1086 -- ttt is the name of the original declared array 1087 -- nnn is the component size in bits (1-31) 1088 1089 -- Note that if the packed array is not bit-packed, the name will simply 1090 -- be tttP. 1091 1092 -- When the debugger sees that an object is of a type that is encoded in 1093 -- this manner, it can use the original type to determine the bounds and 1094 -- the component type, and the component size to determine the packing 1095 -- details. 1096 1097 -- For an unconstrained packed array, the corresponding packed array type 1098 -- is neither used in the generated code nor for debugging information, 1099 -- only the original type is used. In order to convey the packing in the 1100 -- debugging information, the compiler generates the associated fat- and 1101 -- thin-pointer types (see the Pointers to Unconstrained Array section 1102 -- below) using the name of the corresponding packed array type as the 1103 -- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively. 1104 1105 -- When the debugger sees that an object is of a type that is encoded in 1106 -- this manner, it can use the type of the fields to determine the bounds 1107 -- and the component type, and the component size to determine the packing 1108 -- details. 1109 1110 ------------------------------------------- 1111 -- Packed Array Representation in Memory -- 1112 ------------------------------------------- 1113 1114 -- Packed arrays are represented in tightly packed form, with no extra bits 1115 -- between components. This is true even when the component size is not a 1116 -- factor of the storage unit size, so that as a result it is possible for 1117 -- components to cross storage unit boundaries. 1118 1119 -- The layout in storage is identical, regardless of whether the 1120 -- implementation type is a modular type or an array-of-bytes type. See 1121 -- Exp_Pakd for details of how these implementation types are used, but for 1122 -- the purpose of the debugger, only the starting address of the object in 1123 -- memory is significant. 1124 1125 -- The following example should show clearly how the packing works in 1126 -- the little-endian and big-endian cases: 1127 1128 -- type B is range 0 .. 7; 1129 -- for B'Size use 3; 1130 1131 -- type BA is array (0 .. 5) of B; 1132 -- pragma Pack (BA); 1133 1134 -- BV : constant BA := (1,2,3,4,5,6); 1135 1136 -- Little endian case 1137 1138 -- BV'Address + 2 BV'Address + 1 BV'Address + 0 1139 -- +-----------------+-----------------+-----------------+ 1140 -- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 | 1141 -- +-----------------+-----------------+-----------------+ 1142 -- <---------> <-----> <---> <---> <-----> <---> <---> 1143 -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0) 1144 -- 1145 -- Big endian case 1146 -- 1147 -- BV'Address + 0 BV'Address + 1 BV'Address + 2 1148 -- +-----------------+-----------------+-----------------+ 1149 -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? | 1150 -- +-----------------+-----------------+-----------------+ 1151 -- <---> <---> <-----> <---> <---> <-----> <---------> 1152 -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits 1153 1154 -- Note that if a modular type is used to represent the array, the 1155 -- allocation in memory is not the same as a normal modular type. The 1156 -- difference occurs when the allocated object is larger than the size of 1157 -- the array. For a normal modular type, we extend the value on the left 1158 -- with zeroes. 1159 1160 -- For example, in the normal modular case, if we have a 6-bit modular 1161 -- type, declared as mod 2**6, and we allocate an 8-bit object for this 1162 -- type, then we extend the value with two bits on the most significant 1163 -- end, and in either the little-endian or big-endian case, the value 63 1164 -- is represented as 00111111 in binary in memory. 1165 1166 -- For a modular type used to represent a packed array, the rule is 1167 -- different. In this case, if we have to extend the value, then we do it 1168 -- with undefined bits (which are not initialized and whose value is 1169 -- irrelevant to any generated code). Furthermore these bits are on the 1170 -- right (least significant bits) in the big-endian case, and on the left 1171 -- (most significant bits) in the little-endian case. 1172 1173 -- For example, if we have a packed boolean array of 6 bits, all set to 1174 -- True, stored in an 8-bit object, then the value in memory in binary is 1175 -- ??111111 in the little-endian case, and 111111?? in the big-endian case. 1176 1177 -- This is done so that the representation of packed arrays does not 1178 -- depend on whether we use a modular representation or array of bytes 1179 -- as previously described. This ensures that we can pass such values by 1180 -- reference in the case where a subprogram has to be able to handle values 1181 -- stored in either form. 1182 1183 -- Note that when we extract the value of such a modular packed array, we 1184 -- expect to retrieve only the relevant bits, so in this same example, when 1185 -- we extract the value we get 111111 in both cases, and the code generated 1186 -- by the front end assumes this although it does not assume that any high 1187 -- order bits are defined. 1188 1189 -- There are opportunities for optimization based on the knowledge that the 1190 -- unused bits are irrelevant for these type of packed arrays. For example 1191 -- if we have two such 6-bit-in-8-bit values and we do an assignment: 1192 1193 -- a := b; 1194 1195 -- Then logically, we extract the 6 bits and store only 6 bits in the 1196 -- result, but the back end is free to simply assign the entire 8-bits in 1197 -- this case, since we don't actually care about the undefined bits. 1198 -- However, in the equality case, it is important to ensure that the 1199 -- undefined bits do not participate in an equality test. 1200 1201 -- If a modular packed array value is assigned to a register then logically 1202 -- it could always be held right justified, to avoid any need to shift, 1203 -- e.g. when doing comparisons. But probably this is a bad choice, as it 1204 -- would mean that an assignment such as a := above would require shifts 1205 -- when one value is in a register and the other value is in memory. 1206 1207 ------------------------------------------------------ 1208 -- Subprograms for Handling Packed Array Type Names -- 1209 ------------------------------------------------------ 1210 1211 function Make_Packed_Array_Impl_Type_Name 1212 (Typ : Entity_Id; 1213 Csize : Uint) return Name_Id; 1214 -- This function is used in Exp_Pakd to create the name that is encoded as 1215 -- described above. The entity Typ provides the name ttt, and the value 1216 -- Csize is the component size that provides the nnn value. 1217 1218 -------------------------------------- 1219 -- Pointers to Unconstrained Arrays -- 1220 -------------------------------------- 1221 1222 -- There are two kinds of pointers to arrays. The debugger can tell which 1223 -- format is in use by the form of the type of the pointer. 1224 1225 -- Fat Pointers 1226 1227 -- Fat pointers are represented as a struct with two fields. This 1228 -- struct has two distinguished field names: 1229 1230 -- P_ARRAY is a pointer to the array type. The name of this type is 1231 -- the unconstrained type followed by "___XUA". This array will have 1232 -- bounds which are the discriminants, and hence are unparsable, but 1233 -- will give the number of subscripts and the component type. 1234 1235 -- P_BOUNDS is a pointer to a struct, the name of whose type is the 1236 -- unconstrained array name followed by "___XUB" and which has 1237 -- fields of the form 1238 1239 -- LBn (n a decimal integer) lower bound of n'th dimension 1240 -- UBn (n a decimal integer) upper bound of n'th dimension 1241 1242 -- The bounds may be any integral type. In the case of an enumeration 1243 -- type, Enum_Rep values are used. 1244 1245 -- For a given unconstrained array type, the compiler will generate one 1246 -- fat-pointer type whose name is "arr___XUP", where "arr" is the name 1247 -- of the array type, and use it to represent the array type itself in 1248 -- the debugging information. 1249 1250 -- For each pointer to this unconstrained array type, the compiler will 1251 -- generate a typedef that points to the above "arr___XUP" fat-pointer 1252 -- type. As a consequence, when it comes to fat-pointer types: 1253 1254 -- 1. The type name is given by the typedef 1255 1256 -- 2. If the debugger is asked to output the type, the appropriate 1257 -- form is "access arr", except if the type name is "arr___XUP" 1258 -- for which it is the array definition. 1259 1260 -- Thin Pointers 1261 1262 -- The value of a thin pointer is a pointer to the second field of a 1263 -- structure with two fields. The name of this structure's type is 1264 -- "arr___XUT", where "arr" is the name of the unconstrained array 1265 -- type. Even though it actually points into middle of this structure, 1266 -- the thin pointer's type in debugging information is 1267 -- pointer-to-arr___XUT. 1268 1269 -- The first field of arr___XUT is named BOUNDS, and has a type named 1270 -- arr___XUB, with the structure described for such types in fat 1271 -- pointers, as described above. 1272 1273 -- The second field of arr___XUT is named ARRAY, and contains the 1274 -- actual array. Because this array has a dynamic size, determined by 1275 -- the BOUNDS field that precedes it, all of the information about 1276 -- arr___XUT is encoded in a parallel type named arr___XUT___XVE, with 1277 -- fields BOUNDS and ARRAY___XVL. As for previously described ___XVE 1278 -- types, ARRAY___XVL has a pointer-to-array type. However, the array 1279 -- type in this case is named arr___XUA and only its element type is 1280 -- meaningful, just as described for fat pointers. 1281 1282 -------------------------------------- 1283 -- Tagged Types and Type Extensions -- 1284 -------------------------------------- 1285 1286 -- A type C derived from a tagged type P has a field named "_parent" of 1287 -- type P that contains its inherited fields. The type of this field is 1288 -- usually P (encoded as usual if it has a dynamic size), but may be a more 1289 -- distant ancestor, if P is a null extension of that type. 1290 1291 -- The type tag of a tagged type is a field named _tag, of type void*. If 1292 -- the type is derived from another tagged type, its _tag field is found in 1293 -- its _parent field. 1294 1295 ----------------------------- 1296 -- Variant Record Encoding -- 1297 ----------------------------- 1298 1299 -- The variant part of a variant record is encoded as a single field in the 1300 -- enclosing record, whose name is: 1301 1302 -- discrim___XVN 1303 1304 -- where discrim is the unqualified name of the variant. This field name is 1305 -- built by gigi (not by code in this unit). For Unchecked_Union record, 1306 -- this discriminant will not appear in the record (see Unchecked Unions, 1307 -- below). 1308 1309 -- The type corresponding to this field has a name that is obtained by 1310 -- concatenating the type name with the above string and is similar to a C 1311 -- union, in which each member of the union corresponds to one variant. 1312 -- However, unlike a C union, the size of the type may be variable even if 1313 -- each of the components are fixed size, since it includes a computation 1314 -- of which variant is present. In that case, it will be encoded as above 1315 -- and a type with the suffix "___XVN___XVU" will be present. 1316 1317 -- The name of the union member is encoded to indicate the choices, and 1318 -- is a string given by the following grammar: 1319 1320 -- member_name ::= {choice} | others_choice 1321 -- choice ::= simple_choice | range_choice 1322 -- simple_choice ::= S number 1323 -- range_choice ::= R number T number 1324 -- number ::= {decimal_digit} [m] 1325 -- others_choice ::= O (upper case letter O) 1326 1327 -- The m in a number indicates a negative value. As an example of this 1328 -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by 1329 1330 -- R1T4S7S10m 1331 1332 -- In the case of enumeration values, the values used are the actual 1333 -- representation values in the case where an enumeration type has an 1334 -- enumeration representation spec (i.e. they are values that correspond 1335 -- to the use of the Enum_Rep attribute). 1336 1337 -- The type of the inner record is given by the name of the union type (as 1338 -- above) concatenated with the above string. Since that type may itself be 1339 -- variable-sized, it may also be encoded as above with a new type with a 1340 -- further suffix of "___XVU". 1341 1342 -- As an example, consider: 1343 1344 -- type Var (Disc : Boolean := True) is record 1345 -- M : Integer; 1346 1347 -- case Disc is 1348 -- when True => 1349 -- R : Integer; 1350 -- S : Integer; 1351 1352 -- when False => 1353 -- T : Integer; 1354 -- end case; 1355 -- end record; 1356 1357 -- V1 : Var; 1358 1359 -- In this case, the type var is represented as a struct with three fields. 1360 -- The first two are "disc" and "m", representing the values of these 1361 -- record components. The third field is a union of two types, with field 1362 -- names S1 and O. S1 is a struct with fields "r" and "s", and O is a 1363 -- struct with field "t". 1364 1365 ---------------------- 1366 -- Unchecked Unions -- 1367 ---------------------- 1368 1369 -- The encoding for variant records changes somewhat under the influence 1370 -- of a "pragma Unchecked_Union" clause: 1371 1372 -- 1. The discriminant will not be present in the record, although its 1373 -- name is still used in the encodings. 1374 -- 2. Variants containing a single component named "x" of type "T" may 1375 -- be encoded, as in ordinary C unions, as a single field of the 1376 -- enclosing union type named "x" of type "T", dispensing with the 1377 -- enclosing struct. In this case, of course, the discriminant values 1378 -- corresponding to the variant are unavailable. As for normal 1379 -- variants, the field name "x" may be suffixed with ___XVL if it 1380 -- has dynamic size. 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 DWARF2/3 -- 1554 --------------------------------- 1555 1556 -- If the compiler switch "-gdwarf+" is specified, GNAT Vendor extensions 1557 -- to DWARF2/3 are generated, with the following variations from the above 1558 -- specification. 1559 1560 -- Change in the contents of the DW_AT_name attribute 1561 1562 -- The operators are represented in their natural form. (for example, 1563 -- the addition operator is written as "+" instead of "Oadd"). The 1564 -- component separator is "." instead of "__" 1565 1566 -- Introduction of DW_AT_GNAT_encoding, encoded with value 0x2301 1567 1568 -- Any debugging information entry representing a program entity, named 1569 -- or implicit, may have a DW_AT_GNAT_encoding attribute. The value of 1570 -- this attribute is a string representing the suffix internally added 1571 -- by GNAT for various purposes, mainly for representing debug 1572 -- information compatible with other formats. In particular this is 1573 -- useful for IDEs which need to filter out information internal to 1574 -- GNAT from their graphical interfaces. 1575 1576 -- If a debugging information entry has multiple encodings, all of them 1577 -- will be listed in DW_AT_GNAT_encoding using the list separator ':'. 1578 1579 -- Introduction of DW_AT_GNAT_descriptive_type, encoded with value 0x2302 1580 1581 -- Any debugging information entry representing a type may have a 1582 -- DW_AT_GNAT_descriptive_type attribute whose value is a reference, 1583 -- pointing to a debugging information entry representing another type 1584 -- associated to the type. 1585 1586 -- Modification of the contents of the DW_AT_producer string 1587 1588 -- When emitting full GNAT Vendor extensions to DWARF2/3, "-gdwarf+" 1589 -- is appended to the DW_AT_producer string. 1590 -- 1591 -- When emitting only DW_AT_GNAT_descriptive_type, "-gdwarf+-" is 1592 -- appended to the DW_AT_producer string. 1593 1594end Exp_Dbug; 1595