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) 1996-2012, 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 -- As an example of the local name convention, consider a procedure V.W 80 -- with a local variable X, and a nested block Y containing an entity Z. 81 -- The fully qualified names of the entities X and Z are: 82 83 -- V.W.X 84 -- V.W.Y.Z 85 86 -- but since V.W is a subprogram, the encoded names will end up 87 -- encoding only 88 89 -- x 90 -- y.z 91 92 -- The separating dots are translated into double underscores 93 94 ----------------------------- 95 -- Handling of Overloading -- 96 ----------------------------- 97 98 -- The above scheme is incomplete for overloaded subprograms, since 99 -- overloading can legitimately result in case of two entities with 100 -- exactly the same fully qualified names. To distinguish between 101 -- entries in a set of overloaded subprograms, the encoded names are 102 -- serialized by adding the suffix: 103 104 -- __nn (two underscores) 105 106 -- where nn is a serial number (2 for the second overloaded function, 107 -- 3 for the third, etc.). A suffix of __1 is always omitted (i.e. no 108 -- suffix implies the first instance). 109 110 -- These names are prefixed by the normal full qualification. So for 111 -- example, the third instance of the subprogram qrs in package yz 112 -- would have the name: 113 114 -- yz__qrs__3 115 116 -- A more subtle case arises with entities declared within overloaded 117 -- subprograms. If we have two overloaded subprograms, and both declare 118 -- an entity xyz, then the fully expanded name of the two xyz's is the 119 -- same. To distinguish these, we add the same __n suffix at the end of 120 -- the inner entity names. 121 122 -- In more complex cases, we can have multiple levels of overloading, 123 -- and we must make sure to distinguish which final declarative region 124 -- we are talking about. For this purpose, we use a more complex suffix 125 -- which has the form: 126 127 -- __nn_nn_nn ... 128 129 -- where the nn values are the homonym numbers as needed for any of the 130 -- qualifying entities, separated by a single underscore. If all the nn 131 -- values are 1, the suffix is omitted, Otherwise the suffix is present 132 -- (including any values of 1). The following example shows how this 133 -- suffixing works. 134 135 -- package body Yz is 136 -- procedure Qrs is -- Name is yz__qrs 137 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv 138 -- begin ... end Qrs; 139 140 -- procedure Qrs (X: Int) is -- Name is yz__qrs__2 141 -- procedure Tuv is ... end; -- Name is yz__qrs__tuv__2_1 142 -- procedure Tuv (X: Int) is -- Name is yz__qrs__tuv__2_2 143 -- begin ... end Tuv; 144 145 -- procedure Tuv (X: Float) is -- Name is yz__qrs__tuv__2_3 146 -- type m is new float; -- Name is yz__qrs__tuv__m__2_3 147 -- begin ... end Tuv; 148 -- begin ... end Qrs; 149 -- end Yz; 150 151 -------------------- 152 -- Operator Names -- 153 -------------------- 154 155 -- The above rules applied to operator names would result in names with 156 -- quotation marks, which are not typically allowed by assemblers and 157 -- linkers, and even if allowed would be odd and hard to deal with. To 158 -- avoid this problem, operator names are encoded as follows: 159 160 -- Oabs abs 161 -- Oand and 162 -- Omod mod 163 -- Onot not 164 -- Oor or 165 -- Orem rem 166 -- Oxor xor 167 -- Oeq = 168 -- One /= 169 -- Olt < 170 -- Ole <= 171 -- Ogt > 172 -- Oge >= 173 -- Oadd + 174 -- Osubtract - 175 -- Oconcat & 176 -- Omultiply * 177 -- Odivide / 178 -- Oexpon ** 179 180 -- These names are prefixed by the normal full qualification, and 181 -- suffixed by the overloading identification. So for example, the 182 -- second operator "=" defined in package Extra.Messages would have 183 -- the name: 184 185 -- extra__messages__Oeq__2 186 187 ---------------------------------- 188 -- Resolving Other Name Clashes -- 189 ---------------------------------- 190 191 -- It might be thought that the above scheme is complete, but in Ada 95, 192 -- full qualification is insufficient to uniquely identify an entity in 193 -- the program, even if it is not an overloaded subprogram. There are 194 -- two possible confusions: 195 196 -- a.b 197 198 -- interpretation 1: entity b in body of package a 199 -- interpretation 2: child procedure b of package a 200 201 -- a.b.c 202 203 -- interpretation 1: entity c in child package a.b 204 -- interpretation 2: entity c in nested package b in body of a 205 206 -- It is perfectly legal in both cases for both interpretations to be 207 -- valid within a single program. This is a bit of a surprise since 208 -- certainly in Ada 83, full qualification was sufficient, but not in 209 -- Ada 95. The result is that the above scheme can result in duplicate 210 -- names. This would not be so bad if the effect were just restricted 211 -- to debugging information, but in fact in both the above cases, it 212 -- is possible for both symbols to be external names, and so we have 213 -- a real problem of name clashes. 214 215 -- To deal with this situation, we provide two additional encoding 216 -- rules for names: 217 218 -- First: all library subprogram names are preceded by the string 219 -- _ada_ (which causes no duplications, since normal Ada names can 220 -- never start with an underscore. This not only solves the first 221 -- case of duplication, but also solves another pragmatic problem 222 -- which is that otherwise Ada procedures can generate names that 223 -- clash with existing system function names. Most notably, we can 224 -- have clashes in the case of procedure Main with the C main that 225 -- in some systems is always present. 226 227 -- Second, for the case where nested packages declared in package 228 -- bodies can cause trouble, we add a suffix which shows which 229 -- entities in the list are body-nested packages, i.e. packages 230 -- whose spec is within a package body. The rules are as follows, 231 -- given a list of names in a qualified name name1.name2.... 232 233 -- If none are body-nested package entities, then there is no suffix 234 235 -- If at least one is a body-nested package entity, then the suffix 236 -- is X followed by a string of b's and n's (b = body-nested package 237 -- entity, n = not a body-nested package). 238 239 -- There is one element in this string for each entity in the encoded 240 -- expanded name except the first (the rules are such that the first 241 -- entity of the encoded expanded name can never be a body-nested' 242 -- package. Trailing n's are omitted, as is the last b (there must 243 -- be at least one b, or we would not be generating a suffix at all). 244 245 -- For example, suppose we have 246 247 -- package x is 248 -- pragma Elaborate_Body; 249 -- m1 : integer; -- #1 250 -- end x; 251 252 -- package body x is 253 -- package y is m2 : integer; end y; -- #2 254 -- package body y is 255 -- package z is r : integer; end z; -- #3 256 -- end; 257 -- m3 : integer; -- #4 258 -- end x; 259 260 -- package x.y is 261 -- pragma Elaborate_Body; 262 -- m2 : integer; -- #5 263 -- end x.y; 264 265 -- package body x.y is 266 -- m3 : integer; -- #6 267 -- procedure j is -- #7 268 -- package k is 269 -- z : integer; -- #8 270 -- end k; 271 -- begin 272 -- null; 273 -- end j; 274 -- end x.y; 275 276 -- procedure x.m3 is begin null; end; -- #9 277 278 -- Then the encodings would be: 279 280 -- #1. x__m1 (no BNPE's in sight) 281 -- #2. x__y__m2X (y is a BNPE) 282 -- #3. x__y__z__rXb (y is a BNPE, so is z) 283 -- #4. x__m3 (no BNPE's in sight) 284 -- #5. x__y__m2 (no BNPE's in sight) 285 -- #6. x__y__m3 (no BNPE's in signt) 286 -- #7. x__y__j (no BNPE's in sight) 287 -- #8. k__z (no BNPE's, only up to procedure) 288 -- #9 _ada_x__m3 (library level subprogram) 289 290 -- Note that we have instances here of both kind of potential name 291 -- clashes, and the above examples show how the encodings avoid the 292 -- clash as follows: 293 294 -- Lines #4 and #9 both refer to the entity x.m3, but #9 is a library 295 -- level subprogram, so it is preceded by the string _ada_ which acts 296 -- to distinguish it from the package body entity. 297 298 -- Lines #2 and #5 both refer to the entity x.y.m2, but the first 299 -- instance is inside the body-nested package y, so there is an X 300 -- suffix to distinguish it from the child library entity. 301 302 -- Note that enumeration literals never need Xb type suffixes, since 303 -- they are never referenced using global external names. 304 305 --------------------- 306 -- Interface Names -- 307 --------------------- 308 309 -- Note: if an interface name is present, then the external name is 310 -- taken from the specified interface name. Given current limitations of 311 -- the gcc backend, this means that the debugging name is also set to 312 -- the interface name, but conceptually, it would be possible (and 313 -- indeed desirable) to have the debugging information still use the Ada 314 -- name as qualified above, so we still fully qualify the name in the 315 -- front end. 316 317 ------------------------------------- 318 -- Encodings Related to Task Types -- 319 ------------------------------------- 320 321 -- Each task object defined by a single task declaration is associated 322 -- with a prefix that is used to qualify procedures defined in that 323 -- task. Given 324 -- 325 -- package body P is 326 -- task body TaskObj is 327 -- procedure F1 is ... end; 328 -- begin 329 -- B; 330 -- end TaskObj; 331 -- end P; 332 -- 333 -- The name of subprogram TaskObj.F1 is encoded as p__taskobjTK__f1. 334 -- The body, B, is contained in a subprogram whose name is 335 -- p__taskobjTKB. 336 337 ------------------------------------------ 338 -- Encodings Related to Protected Types -- 339 ------------------------------------------ 340 341 -- Each protected type has an associated record type, that describes 342 -- the actual layout of the private data. In addition to the private 343 -- components of the type, the Corresponding_Record_Type includes one 344 -- component of type Protection, which is the actual lock structure. 345 -- The run-time size of the protected type is the size of the corres- 346 -- ponding record. 347 348 -- For a protected type prot, the Corresponding_Record_Type is encoded 349 -- as protV. 350 351 -- The operations of a protected type are encoded as follows: each 352 -- operation results in two subprograms, a locking one that is called 353 -- from outside of the object, and a non-locking one that is used for 354 -- calls from other operations on the same object. The locking operation 355 -- simply acquires the lock, and then calls the non-locking version. 356 -- The names of all of these have a prefix constructed from the name of 357 -- the type, and a suffix which is P or N, depending on whether this is 358 -- the protected/non-locking version of the operation. 359 360 -- Operations generated for protected entries follow the same encoding. 361 -- Each entry results in two subprograms: a procedure that holds the 362 -- entry body, and a function that holds the evaluation of the barrier. 363 -- The names of these subprograms include the prefix '_E' or '_B' res- 364 -- pectively. The names also include a numeric suffix to render them 365 -- unique in the presence of overloaded entries. 366 367 -- Given the declaration: 368 369 -- protected type Lock is 370 -- function Get return Integer; 371 -- procedure Set (X: Integer); 372 -- entry Update (Val : Integer); 373 -- private 374 -- Value : Integer := 0; 375 -- end Lock; 376 377 -- the following operations are created: 378 379 -- lock_getN 380 -- lock_getP, 381 382 -- lock_setN 383 -- lock_setP 384 385 -- lock_update_E1s 386 -- lock_udpate_B2s 387 388 -- If the protected type implements at least one interface, the 389 -- following additional operations are created: 390 391 -- lock_get 392 393 -- lock_set 394 395 -- These operations are used to ensure overriding of interface level 396 -- subprograms and proper dispatching on interface class-wide objects. 397 -- The bodies of these operations contain calls to their respective 398 -- protected versions: 399 400 -- function lock_get return Integer is 401 -- begin 402 -- return lock_getP; 403 -- end lock_get; 404 405 -- procedure lock_set (X : Integer) is 406 -- begin 407 -- lock_setP (X); 408 -- end lock_set; 409 410 ---------------------------------------------------- 411 -- Conversion between Entities and External Names -- 412 ---------------------------------------------------- 413 414 procedure Get_External_Name 415 (Entity : Entity_Id; 416 Has_Suffix : Boolean := False; 417 Suffix : String := ""); 418 -- Set Name_Buffer and Name_Len to the external name of the entity. The 419 -- external name is the Interface_Name, if specified, unless the entity 420 -- has an address clause or Has_Suffix is true. 421 -- 422 -- If the Interface is not present, or not used, the external name is the 423 -- concatenation of: 424 -- 425 -- - the string "_ada_", if the entity is a library subprogram, 426 -- - the names of any enclosing scopes, each followed by "__", 427 -- or "X_" if the next entity is a subunit) 428 -- - the name of the entity 429 -- - the string "$" (or "__" if target does not allow "$"), followed 430 -- by homonym suffix, if the entity is an overloaded subprogram 431 -- or is defined within an overloaded subprogram. 432 -- - the string "___" followed by Suffix if Has_Suffix is true. 433 -- 434 -- Note that a call to this procedure has no effect if we are not 435 -- generating code, since the necessary information for computing the 436 -- proper external name is not available in this case. 437 438 -------------------------------------------- 439 -- Subprograms for Handling Qualification -- 440 -------------------------------------------- 441 442 procedure Qualify_Entity_Names (N : Node_Id); 443 -- Given a node N, that represents a block, subprogram body, or package 444 -- body or spec, or protected or task type, sets a fully qualified name 445 -- for the defining entity of given construct, and also sets fully 446 -- qualified names for all enclosed entities of the construct (using 447 -- First_Entity/Next_Entity). Note that the actual modifications of the 448 -- names is postponed till a subsequent call to Qualify_All_Entity_Names. 449 -- Note: this routine does not deal with prepending _ada_ to library 450 -- subprogram names. The reason for this is that we only prepend _ada_ 451 -- to the library entity itself, and not to names built from this name. 452 453 procedure Qualify_All_Entity_Names; 454 -- When Qualify_Entity_Names is called, no actual name changes are made, 455 -- i.e. the actual calls to Qualify_Entity_Name are deferred until a call 456 -- is made to this procedure. The reason for this deferral is that when 457 -- names are changed semantic processing may be affected. By deferring 458 -- the changes till just before gigi is called, we avoid any concerns 459 -- about such effects. Gigi itself does not use the names except for 460 -- output of names for debugging purposes (which is why we are doing 461 -- the name changes in the first place. 462 463 -- Note: the routines Get_Unqualified_[Decoded]_Name_String in Namet are 464 -- useful to remove qualification from a name qualified by the call to 465 -- Qualify_All_Entity_Names. 466 467 -------------------------------- 468 -- Handling of Numeric Values -- 469 -------------------------------- 470 471 -- All numeric values here are encoded as strings of decimal digits. Only 472 -- integer values need to be encoded. A negative value is encoded as the 473 -- corresponding positive value followed by a lower case m for minus to 474 -- indicate that the value is negative (e.g. 2m for -2). 475 476 ------------------------- 477 -- Type Name Encodings -- 478 ------------------------- 479 480 -- In the following typ is the name of the type as normally encoded by the 481 -- debugger rules, i.e. a non-qualified name, all in lower case, with 482 -- standard encoding of upper half and wide characters 483 484 ------------------------ 485 -- Encapsulated Types -- 486 ------------------------ 487 488 -- In some cases, the compiler encapsulates a type by wrapping it in a 489 -- structure. For example, this is used when a size or alignment 490 -- specification requires a larger type. Consider: 491 492 -- type y is mod 2 ** 64; 493 -- for y'size use 256; 494 495 -- In this case the compile generates a structure type y___PAD, which 496 -- has a single field whose name is F. This single field is 64 bits 497 -- long and contains the actual value. This kind of padding is used 498 -- when the logical value to be stored is shorter than the object in 499 -- which it is allocated. For example if a size clause is used to set 500 -- a size of 256 for a signed integer value, then a typical choice is 501 -- to wrap a 64-bit integer in a 256 bit PAD structure. 502 503 -- A similar encapsulation is done for some packed array types, in which 504 -- case the structure type is y___JM and the field name is OBJECT. 505 -- This is used in the case of a packed array stored using modular 506 -- representation (see section on representation of packed array 507 -- objects). In this case the JM wrapping is used to achieve correct 508 -- positioning of the packed array value (left or right justified in its 509 -- field depending on endianness. 510 511 -- When the debugger sees an object of a type whose name has a suffix of 512 -- ___PAD or ___JM, the type will be a record containing a single field, 513 -- and the name of that field will be all upper case. In this case, it 514 -- should look inside to get the value of the inner field, and neither 515 -- the outer structure name, nor the field name should appear when the 516 -- value is printed. 517 518 -- When the debugger sees a record named REP being a field inside 519 -- another record, it should treat the fields inside REP as being part 520 -- of the outer record (this REP field is only present for code 521 -- generation purposes). The REP record should not appear in the values 522 -- printed by the debugger. 523 524 ----------------------- 525 -- Fixed-Point Types -- 526 ----------------------- 527 528 -- Fixed-point types are encoded using a suffix that indicates the 529 -- delta and small values. The actual type itself is a normal integer 530 -- type. 531 532 -- typ___XF_nn_dd 533 -- typ___XF_nn_dd_nn_dd 534 535 -- The first form is used when small = delta. The value of delta (and 536 -- small) is given by the rational nn/dd, where nn and dd are decimal 537 -- integers. 538 -- 539 -- The second form is used if the small value is different from the 540 -- delta. In this case, the first nn/dd rational value is for delta, 541 -- and the second value is for small. 542 543 ------------------------------ 544 -- VAX Floating-Point Types -- 545 ------------------------------ 546 547 -- Vax floating-point types are represented at run time as integer 548 -- types, which are treated specially by the code generator. Their 549 -- type names are encoded with the following suffix: 550 551 -- typ___XFF 552 -- typ___XFD 553 -- typ___XFG 554 555 -- representing the Vax F Float, D Float, and G Float types. The 556 -- debugger must treat these specially. In particular, printing these 557 -- values can be achieved using the debug procedures that are provided 558 -- in package System.Vax_Float_Operations: 559 560 -- procedure Debug_Output_D (Arg : D); 561 -- procedure Debug_Output_F (Arg : F); 562 -- procedure Debug_Output_G (Arg : G); 563 564 -- These three procedures take a Vax floating-point argument, and 565 -- output a corresponding decimal representation to standard output 566 -- with no terminating line return. 567 568 -------------------- 569 -- Discrete Types -- 570 -------------------- 571 572 -- Discrete types are coded with a suffix indicating the range in the 573 -- case where one or both of the bounds are discriminants or variable. 574 575 -- Note: at the current time, we also encode compile time known bounds 576 -- if they do not match the natural machine type bounds, but this may 577 -- be removed in the future, since it is redundant for most debugging 578 -- formats. However, we do not ever need XD encoding for enumeration 579 -- base types, since here it is always clear what the bounds are from 580 -- the total number of enumeration literals. 581 582 -- typ___XD 583 -- typ___XDL_lowerbound 584 -- typ___XDU_upperbound 585 -- typ___XDLU_lowerbound__upperbound 586 587 -- If a discrete type is a natural machine type (i.e. its bounds 588 -- correspond in a natural manner to its size), then it is left 589 -- unencoded. The above encoding forms are used when there is a 590 -- constrained range that does not correspond to the size or that 591 -- has discriminant references or other compile time known bounds. 592 593 -- The first form is used if both bounds are dynamic, in which case two 594 -- constant objects are present whose names are typ___L and typ___U in 595 -- the same scope as typ, and the values of these constants indicate 596 -- the bounds. As far as the debugger is concerned, these are simply 597 -- variables that can be accessed like any other variables. In the 598 -- enumeration case, these values correspond to the Enum_Rep values for 599 -- the lower and upper bounds. 600 601 -- The second form is used if the upper bound is dynamic, but the lower 602 -- bound is either constant or depends on a discriminant of the record 603 -- with which the type is associated. The upper bound is stored in a 604 -- constant object of name typ___U as previously described, but the 605 -- lower bound is encoded directly into the name as either a decimal 606 -- integer, or as the discriminant name. 607 608 -- The third form is similarly used if the lower bound is dynamic, but 609 -- the upper bound is compile time known or a discriminant reference, 610 -- in which case the lower bound is stored in a constant object of name 611 -- typ___L, and the upper bound is encoded directly into the name as 612 -- either a decimal integer, or as the discriminant name. 613 614 -- The fourth form is used if both bounds are discriminant references 615 -- or compile time known values, with the encoding first for the lower 616 -- bound, then for the upper bound, as previously described. 617 618 ------------------- 619 -- Modular Types -- 620 ------------------- 621 622 -- A type declared 623 624 -- type x is mod N; 625 626 -- Is encoded as a subrange of an unsigned base type with lower bound 627 -- zero and upper bound N. That is, there is no name encoding. We use 628 -- the standard encodings provided by the debugging format. Thus we 629 -- give these types a non-standard interpretation: the standard 630 -- interpretation of our encoding would not, in general, imply that 631 -- arithmetic on type x was to be performed modulo N (especially not 632 -- when N is not a power of 2). 633 634 ------------------ 635 -- Biased Types -- 636 ------------------ 637 638 -- Only discrete types can be biased, and the fact that they are biased 639 -- is indicated by a suffix of the form: 640 641 -- typ___XB_lowerbound__upperbound 642 643 -- Here lowerbound and upperbound are decimal integers, with the usual 644 -- (postfix "m") encoding for negative numbers. Biased types are only 645 -- possible where the bounds are compile time known, and the values are 646 -- represented as unsigned offsets from the lower bound given. For 647 -- example: 648 649 -- type Q is range 10 .. 15; 650 -- for Q'size use 3; 651 652 -- The size clause will force values of type Q in memory to be stored 653 -- in biased form (e.g. 11 will be represented by the bit pattern 001). 654 655 ---------------------------------------------- 656 -- Record Types with Variable-Length Fields -- 657 ---------------------------------------------- 658 659 -- The debugging formats do not fully support these types, and indeed 660 -- some formats simply generate no useful information at all for such 661 -- types. In order to provide information for the debugger, gigi creates 662 -- a parallel type in the same scope with one of the names 663 664 -- type___XVE 665 -- type___XVU 666 667 -- The former name is used for a record and the latter for the union 668 -- that is made for a variant record (see below) if that record or union 669 -- has a field of variable size or if the record or union itself has a 670 -- variable size. These encodings suffix any other encodings that that 671 -- might be suffixed to the type name. 672 673 -- The idea here is to provide all the needed information to interpret 674 -- objects of the original type in the form of a "fixed up" type, which 675 -- is representable using the normal debugging information. 676 677 -- There are three cases to be dealt with. First, some fields may have 678 -- variable positions because they appear after variable-length fields. 679 -- To deal with this, we encode *all* the field bit positions of the 680 -- special ___XV type in a non-standard manner. 681 682 -- The idea is to encode not the position, but rather information that 683 -- allows computing the position of a field from the position of the 684 -- previous field. The algorithm for computing the actual positions of 685 -- all fields and the length of the record is as follows. In this 686 -- description, let P represent the current bit position in the record. 687 688 -- 1. Initialize P to 0 689 690 -- 2. For each field in the record: 691 692 -- 2a. If an alignment is given (see below), then round P up, if 693 -- needed, to the next multiple of that alignment. 694 695 -- 2b. If a bit position is given, then increment P by that amount 696 -- (that is, treat it as an offset from the end of the preceding 697 -- record). 698 699 -- 2c. Assign P as the actual position of the field 700 701 -- 2d. Compute the length, L, of the represented field (see below) 702 -- and compute P'=P+L. Unless the field represents a variant part 703 -- (see below and also Variant Record Encoding), set P to P'. 704 705 -- The alignment, if present, is encoded in the field name of the 706 -- record, which has a suffix: 707 708 -- fieldname___XVAnn 709 710 -- where the nn after the XVA indicates the alignment value in storage 711 -- units. This encoding is present only if an alignment is present. 712 713 -- The size of the record described by an XVE-encoded type (in bits) is 714 -- generally the maximum value attained by P' in step 2d above, rounded 715 -- up according to the record's alignment. 716 717 -- Second, the variable-length fields themselves are represented by 718 -- replacing the type by a special access type. The designated type of 719 -- this access type is the original variable-length type, and the fact 720 -- that this field has been transformed in this way is signalled by 721 -- encoding the field name as: 722 723 -- field___XVL 724 725 -- where field is the original field name. If a field is both 726 -- variable-length and also needs an alignment encoding, then the 727 -- encodings are combined using: 728 729 -- field___XVLnn 730 731 -- Note: the reason that we change the type is so that the resulting 732 -- type has no variable-length fields. At least some of the formats used 733 -- for debugging information simply cannot tolerate variable- length 734 -- fields, so the encoded information would get lost. 735 736 -- Third, in the case of a variant record, the special union that 737 -- contains the variants is replaced by a normal C union. In this case, 738 -- the positions are all zero. 739 740 -- Discriminants appear before any variable-length fields that depend on 741 -- them, with one exception. In some cases, a discriminant governing the 742 -- choice of a variant clause may appear in the list of fields of an XVE 743 -- type after the entry for the variant clause itself (this can happen 744 -- in the presence of a representation clause for the record type in the 745 -- source program). However, when this happens, the discriminant's 746 -- position may be determined by first applying the rules described in 747 -- this section, ignoring the variant clause. As a result, discriminants 748 -- can always be located independently of the variable-length fields 749 -- that depend on them. 750 751 -- The size of the ___XVE or ___XVU record or union is set to the 752 -- alignment (in bytes) of the original object so that the debugger 753 -- can calculate the size of the original type. 754 755 -- As an example of this encoding, consider the declarations: 756 757 -- type Q is array (1 .. V1) of Float; -- alignment 4 758 -- type R is array (1 .. V2) of Long_Float; -- alignment 8 759 760 -- type X is record 761 -- A : Character; 762 -- B : Float; 763 -- C : String (1 .. V3); 764 -- D : Float; 765 -- E : Q; 766 -- F : R; 767 -- G : Float; 768 -- end record; 769 770 -- The encoded type looks like: 771 772 -- type anonymousQ is access Q; 773 -- type anonymousR is access R; 774 775 -- type X___XVE is record 776 -- A : Character; -- position contains 0 777 -- B : Float; -- position contains 24 778 -- C___XVL : access String (1 .. V3); -- position contains 0 779 -- D___XVA4 : Float; -- position contains 0 780 -- E___XVL4 : anonymousQ; -- position contains 0 781 -- F___XVL8 : anonymousR; -- position contains 0 782 -- G : Float; -- position contains 0 783 -- end record; 784 785 -- Any bit sizes recorded for fields other than dynamic fields and 786 -- variants are honored as for ordinary records. 787 788 -- Notes: 789 790 -- 1) The B field could also have been encoded by using a position of 791 -- zero and an alignment of 4, but in such a case the coding by position 792 -- is preferred (since it takes up less space). We have used the 793 -- (illegal) notation access xxx as field types in the example above. 794 795 -- 2) The E field does not actually need the alignment indication but 796 -- this may not be detected in this case by the conversion routines. 797 798 -- 3) Our conventions do not cover all XVE-encoded records in which 799 -- some, but not all, fields have representation clauses. Such records 800 -- may, therefore, be displayed incorrectly by debuggers. This situation 801 -- is not common. 802 803 ----------------------- 804 -- Base Record Types -- 805 ----------------------- 806 807 -- Under certain circumstances, debuggers need two descriptions of a 808 -- record type, one that gives the actual details of the base type's 809 -- structure (as described elsewhere in these comments) and one that may 810 -- be used to obtain information about the particular subtype and the 811 -- size of the objects being typed. In such cases the compiler will 812 -- substitute type whose name is typically compiler-generated and 813 -- irrelevant except as a key for obtaining the actual type. 814 815 -- Specifically, if this name is x, then we produce a record type named 816 -- x___XVS consisting of one field. The name of this field is that of 817 -- the actual type being encoded, which we'll call y. The type of this 818 -- single field can be either an arbitrary non-reference type, e.g. an 819 -- integer type, or a reference type; in the latter case, the referenced 820 -- type is also the actual type being encoded y. Both x and y may have 821 -- corresponding ___XVE types. 822 823 -- The size of the objects typed as x should be obtained from the 824 -- structure of x (and x___XVE, if applicable) as for ordinary types 825 -- unless there is a variable named x___XVZ, which, if present, will 826 -- hold the size (in bytes) of x. In this latter case, the size of the 827 -- x___XVS type will not be a constant but a reference to x___XVZ. 828 829 -- The type x will either be a subtype of y (see also Subtypes of 830 -- Variant Records, below) or will contain a single field of type y, 831 -- or no fields at all. The layout, types, and positions of these 832 -- fields will be accurate, if present. (Currently, however, the GDB 833 -- debugger makes no use of x except to determine its size). 834 835 -- Among other uses, XVS types are used to encode unconstrained types. 836 -- For example, given: 837 -- 838 -- subtype Int is INTEGER range 0..10; 839 -- type T1 (N: Int := 0) is record 840 -- F1: String (1 .. N); 841 -- end record; 842 -- type AT1 is array (INTEGER range <>) of T1; 843 -- 844 -- the element type for AT1 might have a type defined as if it had 845 -- been written: 846 -- 847 -- type at1___PAD is record F : T1; end record; 848 -- for at1___PAD'Size use 16 * 8; 849 -- 850 -- and there would also be: 851 -- 852 -- type at1___PAD___XVS is record t1: reft1; end record; 853 -- type t1 is ... 854 -- type reft1 is <reference to t1> 855 -- 856 -- Had the subtype Int been dynamic: 857 -- 858 -- subtype Int is INTEGER range 0 .. M; -- M a variable 859 -- 860 -- Then the compiler would also generate a declaration whose effect 861 -- would be 862 -- 863 -- at1___PAD___XVZ: constant Integer := 32 + M * 8 + padding term; 864 -- 865 -- Not all unconstrained types are so encoded; the XVS convention may be 866 -- unnecessary for unconstrained types of fixed size. However, this 867 -- encoding is always necessary when a subcomponent type (array 868 -- element's type or record field's type) is an unconstrained record 869 -- type some of whose components depend on discriminant values. 870 871 ----------------- 872 -- Array Types -- 873 ----------------- 874 875 -- Since there is no way for the debugger to obtain the index subtypes 876 -- for an array type, we produce a type that has the name of the array 877 -- type followed by "___XA" and is a record type whose field types are 878 -- the respective types for the bounds (and whose field names are the 879 -- names of these types). 880 881 -- To conserve space, we do not produce this type unless one of the 882 -- index types is either an enumeration type, has a variable upper 883 -- bound, has a lower bound different from the constant 1, is a biased 884 -- type, or is wider than "sizetype". 885 886 -- Given the full encoding of these types (see above description for 887 -- the encoding of discrete types), this means that all necessary 888 -- information for addressing arrays is available. In some debugging 889 -- formats, some or all of the bounds information may be available 890 -- redundantly, particularly in the fixed-point case, but this 891 -- information can in any case be ignored by the debugger. 892 893 ---------------------------- 894 -- Note on Implicit Types -- 895 ---------------------------- 896 897 -- The compiler creates implicit type names in many situations where a 898 -- type is present semantically, but no specific name is present. For 899 -- example: 900 901 -- S : Integer range M .. N; 902 903 -- Here the subtype of S is not integer, but rather an anonymous subtype 904 -- of Integer. Where possible, the compiler generates names for such 905 -- anonymous types that are related to the type from which the subtype 906 -- is obtained as follows: 907 908 -- T name suffix 909 910 -- where name is the name from which the subtype is obtained, using 911 -- lower case letters and underscores, and suffix starts with an upper 912 -- case letter. For example the name for the above declaration might be: 913 914 -- TintegerS4b 915 916 -- If the debugger is asked to give the type of an entity and the type 917 -- has the form T name suffix, it is probably appropriate to just use 918 -- "name" in the response since this is what is meaningful to the 919 -- programmer. 920 921 ------------------------------------------------- 922 -- Subprograms for Handling Encoded Type Names -- 923 ------------------------------------------------- 924 925 procedure Get_Encoded_Name (E : Entity_Id); 926 -- If the entity is a typename, store the external name of the entity as in 927 -- Get_External_Name, followed by three underscores plus the type encoding 928 -- in Name_Buffer with the length in Name_Len, and an ASCII.NUL character 929 -- stored following the name. Otherwise set Name_Buffer and Name_Len to 930 -- hold the entity name. Note that a call to this procedure has no effect 931 -- if we are not generating code, since the necessary information for 932 -- computing the proper encoded name is not available in this case. 933 934 -------------- 935 -- Renaming -- 936 -------------- 937 938 -- Debugging information is generated for exception, object, package, and 939 -- subprogram renaming (generic renamings are not significant, since 940 -- generic templates are not relevant at debugging time). 941 942 -- Consider a renaming declaration of the form 943 944 -- x : typ renames y; 945 946 -- There is one case in which no special debugging information is required, 947 -- namely the case of an object renaming where the back end allocates a 948 -- reference for the renamed variable, and the entity x is this reference. 949 -- The debugger can handle this case without any special processing or 950 -- encoding (it won't know it was a renaming, but that does not matter). 951 952 -- All other cases of renaming generate a dummy variable for an entity 953 -- whose name is of the form: 954 955 -- x___XR_... for an object renaming 956 -- x___XRE_... for an exception renaming 957 -- x___XRP_... for a package renaming 958 959 -- and where the "..." represents a suffix that describes the structure of 960 -- the object name given in the renaming (see details below). 961 962 -- The name is fully qualified in the usual manner, i.e. qualified in the 963 -- same manner as the entity x would be. In the case of a package renaming 964 -- where x is a child unit, the qualification includes the name of the 965 -- parent unit, to disambiguate child units with the same simple name and 966 -- (of necessity) different parents. 967 968 -- Note: subprogram renamings are not encoded at the present time 969 970 -- The suffix of the variable name describing the renamed object is defined 971 -- to use the following encoding: 972 973 -- For the simple entity case, where y is just an entity name, the suffix 974 -- is of the form: 975 976 -- y___XE 977 978 -- i.e. the suffix has a single field, the first part matching the 979 -- name y, followed by a "___" separator, ending with sequence XE. 980 -- The entity name portion is fully qualified in the usual manner. 981 -- This same naming scheme is followed for all forms of encoded 982 -- renamings that rename a simple entity. 983 984 -- For the object renaming case where y is a selected component or an 985 -- indexed component, the variable name is suffixed by additional fields 986 -- that give details of the components. The name starts as above with a 987 -- y___XE name indicating the outer level object entity. Then a series of 988 -- selections and indexing operations can be specified as follows: 989 990 -- Indexed component 991 992 -- A series of subscript values appear in sequence, the number 993 -- corresponds to the number of dimensions of the array. The 994 -- subscripts have one of the following two forms: 995 996 -- XSnnn 997 998 -- Here nnn is a constant value, encoded as a decimal integer 999 -- (pos value for enumeration type case). Negative values have 1000 -- a trailing 'm' as usual. 1001 1002 -- XSe 1003 1004 -- Here e is the (unqualified) name of a constant entity in the 1005 -- same scope as the renaming which contains the subscript value. 1006 1007 -- Slice 1008 1009 -- For the slice case, we have two entries. The first is for the 1010 -- lower bound of the slice, and has the form: 1011 1012 -- XLnnn 1013 -- XLe 1014 1015 -- Specifies the lower bound, using exactly the same encoding as 1016 -- for an XS subscript as described above. 1017 1018 -- Then the upper bound appears in the usual XSnnn/XSe form 1019 1020 -- Selected component 1021 1022 -- For a selected component, we have a single entry 1023 1024 -- XRf 1025 1026 -- Here f is the field name for the selection 1027 1028 -- For an explicit dereference (.all), we have a single entry 1029 1030 -- XA 1031 1032 -- As an example, consider the declarations: 1033 1034 -- package p is 1035 -- type q is record 1036 -- m : string (2 .. 5); 1037 -- end record; 1038 -- 1039 -- type r is array (1 .. 10, 1 .. 20) of q; 1040 -- 1041 -- g : r; 1042 -- 1043 -- z : string renames g (1,5).m(2 ..3) 1044 -- end p; 1045 1046 -- The generated variable entity would appear as 1047 1048 -- p__z___XR_p__g___XEXS1XS5XRmXL2XS3 : _renaming_type; 1049 -- p__g___XE--------------------outer entity is g 1050 -- XS1-----------------first subscript for g 1051 -- XS5--------------second subscript for g 1052 -- XRm-----------select field m 1053 -- XL2--------lower bound of slice 1054 -- XS3-----upper bound of slice 1055 1056 -- Note that the type of the variable is a special internal type named 1057 -- _renaming_type. This type is an arbitrary type of zero size created 1058 -- in package Standard (see cstand.adb) and is ignored by the debugger. 1059 1060 function Debug_Renaming_Declaration (N : Node_Id) return Node_Id; 1061 -- The argument N is a renaming declaration. The result is a variable 1062 -- declaration as described in the above paragraphs. If N is not a special 1063 -- debug declaration, then Empty is returned. This function also takes care 1064 -- of setting Materialize_Entity on the renamed entity where required. 1065 1066 --------------------------- 1067 -- Packed Array Encoding -- 1068 --------------------------- 1069 1070 -- For every constrained packed array, two types are created, and both 1071 -- appear in the debugging output: 1072 1073 -- The original declared array type is a perfectly normal array type, and 1074 -- its index bounds indicate the original bounds of the array. 1075 1076 -- The corresponding packed array type, which may be a modular type, or 1077 -- may be an array of bytes type (see Exp_Pakd for full details). This is 1078 -- the type that is actually used in the generated code and for debugging 1079 -- information for all objects of the packed type. 1080 1081 -- The name of the corresponding packed array type is: 1082 1083 -- ttt___XPnnn 1084 1085 -- where 1086 1087 -- ttt is the name of the original declared array 1088 -- nnn is the component size in bits (1-31) 1089 1090 -- When the debugger sees that an object is of a type that is encoded in 1091 -- this manner, it can use the original type to determine the bounds and 1092 -- the component type, and the component size to determine the packing 1093 -- details. 1094 1095 -- For an unconstrained packed array, the corresponding packed array type 1096 -- is neither used in the generated code nor for debugging information, 1097 -- only the original type is used. In order to convey the packing in the 1098 -- debugging information, the compiler generates the associated fat- and 1099 -- thin-pointer types (see the Pointers to Unconstrained Array section 1100 -- below) using the name of the corresponding packed array type as the 1101 -- base name, i.e. ttt___XPnnn___XUP and ttt___XPnnn___XUT respectively. 1102 1103 -- When the debugger sees that an object is of a type that is encoded in 1104 -- this manner, it can use the type of the fields to determine the bounds 1105 -- and the component type, and the component size to determine the packing 1106 -- details. 1107 1108 ------------------------------------------- 1109 -- Packed Array Representation in Memory -- 1110 ------------------------------------------- 1111 1112 -- Packed arrays are represented in tightly packed form, with no extra bits 1113 -- between components. This is true even when the component size is not a 1114 -- factor of the storage unit size, so that as a result it is possible for 1115 -- components to cross storage unit boundaries. 1116 1117 -- The layout in storage is identical, regardless of whether the 1118 -- implementation type is a modular type or an array-of-bytes type. See 1119 -- Exp_Pakd for details of how these implementation types are used, but for 1120 -- the purpose of the debugger, only the starting address of the object in 1121 -- memory is significant. 1122 1123 -- The following example should show clearly how the packing works in 1124 -- the little-endian and big-endian cases: 1125 1126 -- type B is range 0 .. 7; 1127 -- for B'Size use 3; 1128 1129 -- type BA is array (0 .. 5) of B; 1130 -- pragma Pack (BA); 1131 1132 -- BV : constant BA := (1,2,3,4,5,6); 1133 1134 -- Little endian case 1135 1136 -- BV'Address + 2 BV'Address + 1 BV'Address + 0 1137 -- +-----------------+-----------------+-----------------+ 1138 -- | ? ? ? ? ? ? 1 1 | 0 1 0 1 1 0 0 0 | 1 1 0 1 0 0 0 1 | 1139 -- +-----------------+-----------------+-----------------+ 1140 -- <---------> <-----> <---> <---> <-----> <---> <---> 1141 -- unused bits BV(5) BV(4) BV(3) BV(2) BV(1) BV(0) 1142 -- 1143 -- Big endian case 1144 -- 1145 -- BV'Address + 0 BV'Address + 1 BV'Address + 2 1146 -- +-----------------+-----------------+-----------------+ 1147 -- | 0 0 1 0 1 0 0 1 | 1 1 0 0 1 0 1 1 | 1 0 ? ? ? ? ? ? | 1148 -- +-----------------+-----------------+-----------------+ 1149 -- <---> <---> <-----> <---> <---> <-----> <---------> 1150 -- BV(0) BV(1) BV(2) BV(3) BV(4) BV(5) unused bits 1151 1152 -- Note that if a modular type is used to represent the array, the 1153 -- allocation in memory is not the same as a normal modular type. The 1154 -- difference occurs when the allocated object is larger than the size of 1155 -- the array. For a normal modular type, we extend the value on the left 1156 -- with zeroes. 1157 1158 -- For example, in the normal modular case, if we have a 6-bit modular 1159 -- type, declared as mod 2**6, and we allocate an 8-bit object for this 1160 -- type, then we extend the value with two bits on the most significant 1161 -- end, and in either the little-endian or big-endian case, the value 63 1162 -- is represented as 00111111 in binary in memory. 1163 1164 -- For a modular type used to represent a packed array, the rule is 1165 -- different. In this case, if we have to extend the value, then we do it 1166 -- with undefined bits (which are not initialized and whose value is 1167 -- irrelevant to any generated code). Furthermore these bits are on the 1168 -- right (least significant bits) in the big-endian case, and on the left 1169 -- (most significant bits) in the little-endian case. 1170 1171 -- For example, if we have a packed boolean array of 6 bits, all set to 1172 -- True, stored in an 8-bit object, then the value in memory in binary is 1173 -- ??111111 in the little-endian case, and 111111?? in the big-endian case. 1174 1175 -- This is done so that the representation of packed arrays does not 1176 -- depend on whether we use a modular representation or array of bytes 1177 -- as previously described. This ensures that we can pass such values by 1178 -- reference in the case where a subprogram has to be able to handle values 1179 -- stored in either form. 1180 1181 -- Note that when we extract the value of such a modular packed array, we 1182 -- expect to retrieve only the relevant bits, so in this same example, when 1183 -- we extract the value we get 111111 in both cases, and the code generated 1184 -- by the front end assumes this although it does not assume that any high 1185 -- order bits are defined. 1186 1187 -- There are opportunities for optimization based on the knowledge that the 1188 -- unused bits are irrelevant for these type of packed arrays. For example 1189 -- if we have two such 6-bit-in-8-bit values and we do an assignment: 1190 1191 -- a := b; 1192 1193 -- Then logically, we extract the 6 bits and store only 6 bits in the 1194 -- result, but the back end is free to simply assign the entire 8-bits in 1195 -- this case, since we don't actually care about the undefined bits. 1196 -- However, in the equality case, it is important to ensure that the 1197 -- undefined bits do not participate in an equality test. 1198 1199 -- If a modular packed array value is assigned to a register then logically 1200 -- it could always be held right justified, to avoid any need to shift, 1201 -- e.g. when doing comparisons. But probably this is a bad choice, as it 1202 -- would mean that an assignment such as a := above would require shifts 1203 -- when one value is in a register and the other value is in memory. 1204 1205 ------------------------------------------------------ 1206 -- Subprograms for Handling Packed Array Type Names -- 1207 ------------------------------------------------------ 1208 1209 function Make_Packed_Array_Type_Name 1210 (Typ : Entity_Id; 1211 Csize : Uint) 1212 return Name_Id; 1213 -- This function is used in Exp_Pakd to create the name that is encoded as 1214 -- described above. The entity Typ provides the name ttt, and the value 1215 -- Csize is the component size that provides the nnn value. 1216 1217 -------------------------------------- 1218 -- Pointers to Unconstrained Arrays -- 1219 -------------------------------------- 1220 1221 -- There are two kinds of pointers to arrays. The debugger can tell which 1222 -- format is in use by the form of the type of the pointer. 1223 1224 -- Fat Pointers 1225 1226 -- Fat pointers are represented as a struct with two fields. This 1227 -- struct has two distinguished field names: 1228 1229 -- P_ARRAY is a pointer to the array type. The name of this type is 1230 -- the unconstrained type followed by "___XUA". This array will have 1231 -- bounds which are the discriminants, and hence are unparsable, but 1232 -- will give the number of subscripts and the component type. 1233 1234 -- P_BOUNDS is a pointer to a struct, the name of whose type is the 1235 -- unconstrained array name followed by "___XUB" and which has 1236 -- fields of the form 1237 1238 -- LBn (n a decimal integer) lower bound of n'th dimension 1239 -- UBn (n a decimal integer) upper bound of n'th dimension 1240 1241 -- The bounds may be any integral type. In the case of an enumeration 1242 -- type, Enum_Rep values are used. 1243 1244 -- For a given unconstrained array type, the compiler will generate one 1245 -- fat-pointer type whose name is "arr___XUP", where "arr" is the name 1246 -- of the array type, and use it to represent the array type itself in 1247 -- the debugging information. 1248 1249 -- For each pointer to this unconstrained array type, the compiler will 1250 -- generate a typedef that points to the above "arr___XUP" fat-pointer 1251 -- type. As a consequence, when it comes to fat-pointer types: 1252 1253 -- 1. The type name is given by the typedef 1254 1255 -- 2. If the debugger is asked to output the type, the appropriate 1256 -- form is "access arr", except if the type name is "arr___XUP" 1257 -- for which it is the array definition. 1258 1259 -- Thin Pointers 1260 1261 -- The value of a thin pointer is a pointer to the second field of a 1262 -- structure with two fields. The name of this structure's type is 1263 -- "arr___XUT", where "arr" is the name of the unconstrained array 1264 -- type. Even though it actually points into middle of this structure, 1265 -- the thin pointer's type in debugging information is 1266 -- pointer-to-arr___XUT. 1267 1268 -- The first field of arr___XUT is named BOUNDS, and has a type named 1269 -- arr___XUB, with the structure described for such types in fat 1270 -- pointers, as described above. 1271 1272 -- The second field of arr___XUT is named ARRAY, and contains the 1273 -- actual array. Because this array has a dynamic size, determined by 1274 -- the BOUNDS field that precedes it, all of the information about 1275 -- arr___XUT is encoded in a parallel type named arr___XUT___XVE, with 1276 -- fields BOUNDS and ARRAY___XVL. As for previously described ___XVE 1277 -- types, ARRAY___XVL has a pointer-to-array type. However, the array 1278 -- type in this case is named arr___XUA and only its element type is 1279 -- meaningful, just as described for fat pointers. 1280 1281 -------------------------------------- 1282 -- Tagged Types and Type Extensions -- 1283 -------------------------------------- 1284 1285 -- A type C derived from a tagged type P has a field named "_parent" of 1286 -- type P that contains its inherited fields. The type of this field is 1287 -- usually P (encoded as usual if it has a dynamic size), but may be a more 1288 -- distant ancestor, if P is a null extension of that type. 1289 1290 -- The type tag of a tagged type is a field named _tag, of type void*. If 1291 -- the type is derived from another tagged type, its _tag field is found in 1292 -- its _parent field. 1293 1294 ----------------------------- 1295 -- Variant Record Encoding -- 1296 ----------------------------- 1297 1298 -- The variant part of a variant record is encoded as a single field in the 1299 -- enclosing record, whose name is: 1300 1301 -- discrim___XVN 1302 1303 -- where discrim is the unqualified name of the variant. This field name is 1304 -- built by gigi (not by code in this unit). For Unchecked_Union record, 1305 -- this discriminant will not appear in the record (see Unchecked Unions, 1306 -- below). 1307 1308 -- The type corresponding to this field has a name that is obtained by 1309 -- concatenating the type name with the above string and is similar to a C 1310 -- union, in which each member of the union corresponds to one variant. 1311 -- However, unlike a C union, the size of the type may be variable even if 1312 -- each of the components are fixed size, since it includes a computation 1313 -- of which variant is present. In that case, it will be encoded as above 1314 -- and a type with the suffix "___XVN___XVU" will be present. 1315 1316 -- The name of the union member is encoded to indicate the choices, and 1317 -- is a string given by the following grammar: 1318 1319 -- member_name ::= {choice} | others_choice 1320 -- choice ::= simple_choice | range_choice 1321 -- simple_choice ::= S number 1322 -- range_choice ::= R number T number 1323 -- number ::= {decimal_digit} [m] 1324 -- others_choice ::= O (upper case letter O) 1325 1326 -- The m in a number indicates a negative value. As an example of this 1327 -- encoding scheme, the choice 1 .. 4 | 7 | -10 would be represented by 1328 1329 -- R1T4S7S10m 1330 1331 -- In the case of enumeration values, the values used are the actual 1332 -- representation values in the case where an enumeration type has an 1333 -- enumeration representation spec (i.e. they are values that correspond 1334 -- to the use of the Enum_Rep attribute). 1335 1336 -- The type of the inner record is given by the name of the union type (as 1337 -- above) concatenated with the above string. Since that type may itself be 1338 -- variable-sized, it may also be encoded as above with a new type with a 1339 -- further suffix of "___XVU". 1340 1341 -- As an example, consider: 1342 1343 -- type Var (Disc : Boolean := True) is record 1344 -- M : Integer; 1345 1346 -- case Disc is 1347 -- when True => 1348 -- R : Integer; 1349 -- S : Integer; 1350 1351 -- when False => 1352 -- T : Integer; 1353 -- end case; 1354 -- end record; 1355 1356 -- V1 : Var; 1357 1358 -- In this case, the type var is represented as a struct with three fields. 1359 -- The first two are "disc" and "m", representing the values of these 1360 -- record components. The third field is a union of two types, with field 1361 -- names S1 and O. S1 is a struct with fields "r" and "s", and O is a 1362 -- struct with field "t". 1363 1364 ---------------------- 1365 -- Unchecked Unions -- 1366 ---------------------- 1367 1368 -- The encoding for variant records changes somewhat under the influence 1369 -- of a "pragma Unchecked_Union" clause: 1370 1371 -- 1. The discriminant will not be present in the record, although its 1372 -- name is still used in the encodings. 1373 -- 2. Variants containing a single component named "x" of type "T" may 1374 -- be encoded, as in ordinary C unions, as a single field of the 1375 -- enclosing union type named "x" of type "T", dispensing with the 1376 -- enclosing struct. In this case, of course, the discriminant values 1377 -- corresponding to the variant are unavailable. As for normal 1378 -- variants, the field name "x" may be suffixed with ___XVL if it 1379 -- has dynamic size. 1380 1381 -- For example, the type Var in the preceding section, if followed by 1382 -- "pragma Unchecked_Union (Var);" may be encoded as a struct with two 1383 -- fields. The first is "m". The second field is a union of two types, 1384 -- with field names S1 and "t". As before, S1 is a struct with fields 1385 -- "r" and "s". "t" is a field of type Integer. 1386 1387 ------------------------------------------------ 1388 -- Subprograms for Handling Variant Encodings -- 1389 ------------------------------------------------ 1390 1391 procedure Get_Variant_Encoding (V : Node_Id); 1392 -- This procedure is called by Gigi with V being the variant node. The 1393 -- corresponding encoding string is returned in Name_Buffer with the length 1394 -- of the string in Name_Len, and an ASCII.NUL character stored following 1395 -- the name. 1396 1397 --------------------------------- 1398 -- Subtypes of Variant Records -- 1399 --------------------------------- 1400 1401 -- A subtype of a variant record is represented by a type in which the 1402 -- union field from the base type is replaced by one of the possible 1403 -- values. For example, if we have: 1404 1405 -- type Var (Disc : Boolean := True) is record 1406 -- M : Integer; 1407 1408 -- case Disc is 1409 -- when True => 1410 -- R : Integer; 1411 -- S : Integer; 1412 1413 -- when False => 1414 -- T : Integer; 1415 -- end case; 1416 1417 -- end record; 1418 -- V1 : Var; 1419 -- V2 : Var (True); 1420 -- V3 : Var (False); 1421 1422 -- Here V2, for example, is represented with a subtype whose name is 1423 -- something like TvarS3b, which is a struct with three fields. The first 1424 -- two fields are "disc" and "m" as for the base type, and the third field 1425 -- is S1, which contains the fields "r" and "s". 1426 1427 -- The debugger should simply ignore structs with names of the form 1428 -- corresponding to variants, and consider the fields inside as belonging 1429 -- to the containing record. 1430 1431 ----------------------------------------------- 1432 -- Extra renamings for subprogram instances -- 1433 ----------------------------------------------- 1434 1435 procedure Build_Subprogram_Instance_Renamings 1436 (N : Node_Id; 1437 Wrapper : Entity_Id); 1438 -- The debugger has difficulties in recovering the value of actuals of an 1439 -- elementary type, from within the body of a subprogram instantiation. 1440 -- This is because such actuals generate an object declaration that is 1441 -- placed within the wrapper package of the instance, and the entity in 1442 -- these declarations is encoded in a complex way that GDB does not handle 1443 -- well. These new renaming declarations appear within the body of the 1444 -- subprogram, and are redundant from a visibility point of view, but They 1445 -- should have no measurable performance impact, and require no special 1446 -- decoding in the debugger. 1447 1448 ------------------------------------------- 1449 -- Character literals in Character Types -- 1450 ------------------------------------------- 1451 1452 -- Character types are enumeration types at least one of whose enumeration 1453 -- literals is a character literal. Enumeration literals are usually simply 1454 -- represented using their identifier names. If the enumeration literal is 1455 -- a character literal, the name is encoded as described in the following 1456 -- paragraph. 1457 1458 -- A name QUhh, where each 'h' is a lower-case hexadecimal digit, stands 1459 -- for a character whose Unicode encoding is hh, and QWhhhh likewise stands 1460 -- for a wide character whose encoding is hhhh. The representation values 1461 -- are encoded as for ordinary enumeration literals (and have no necessary 1462 -- relationship to the values encoded in the names). 1463 1464 -- For example, given the type declaration 1465 1466 -- type x is (A, 'C', B); 1467 1468 -- the second enumeration literal would be named QU43 and the value 1469 -- assigned to it would be 1. 1470 1471 ----------------------------------------------- 1472 -- Secondary Dispatch tables of tagged types -- 1473 ----------------------------------------------- 1474 1475 procedure Get_Secondary_DT_External_Name 1476 (Typ : Entity_Id; 1477 Ancestor_Typ : Entity_Id; 1478 Suffix_Index : Int); 1479 -- Set Name_Buffer and Name_Len to the external name of one secondary 1480 -- dispatch table of Typ. If the interface has been inherited from some 1481 -- ancestor then Ancestor_Typ is such node (in this case the secondary DT 1482 -- is needed to handle overridden primitives); if there is no such ancestor 1483 -- then Ancestor_Typ is equal to Typ. 1484 -- 1485 -- Internal rule followed for the generation of the external name: 1486 -- 1487 -- Case 1. If the secondary dispatch has not been inherited from some 1488 -- ancestor of Typ then the external name is composed as 1489 -- follows: 1490 -- External_Name (Typ) + Suffix_Number + 'P' 1491 -- 1492 -- Case 2. if the secondary dispatch table has been inherited from some 1493 -- ancestor then the external name is composed as follows: 1494 -- External_Name (Typ) + '_' + External_Name (Ancestor_Typ) 1495 -- + Suffix_Number + 'P' 1496 -- 1497 -- Note: We have to use the external names (instead of simply their names) 1498 -- to protect the frontend against programs that give the same name to all 1499 -- the interfaces and use the expanded name to reference them. The 1500 -- Suffix_Number is used to differentiate all the secondary dispatch 1501 -- tables of a given type. 1502 -- 1503 -- Examples: 1504 -- 1505 -- package Pkg1 is | package Pkg2 is | package Pkg3 is 1506 -- type Typ is | type Typ is | type Typ is 1507 -- interface; | interface; | interface; 1508 -- end Pkg1; | end Pkg; | end Pkg3; 1509 -- 1510 -- with Pkg1, Pkg2, Pkg3; 1511 -- package Case_1 is 1512 -- type Typ is new Pkg1.Typ and Pkg2.Typ and Pkg3.Typ with ... 1513 -- end Case_1; 1514 -- 1515 -- with Case_1; 1516 -- package Case_2 is 1517 -- type Typ is new Case_1.Typ with ... 1518 -- end Case_2; 1519 -- 1520 -- These are the external names generated for Case_1.Typ (note that 1521 -- Pkg1.Typ is associated with the Primary Dispatch Table, because it 1522 -- is the parent of this type, and hence no external name is 1523 -- generated for it). 1524 -- case_1__typ0P (associated with Pkg2.Typ) 1525 -- case_1__typ1P (associated with Pkg3.Typ) 1526 -- 1527 -- These are the external names generated for Case_2.Typ: 1528 -- case_2__typ_case_1__typ0P 1529 -- case_2__typ_case_1__typ1P 1530 1531 ---------------------------- 1532 -- Effect of Optimization -- 1533 ---------------------------- 1534 1535 -- If the program is compiled with optimization on (e.g. -O1 switch 1536 -- specified), then there may be variations in the output from the above 1537 -- specification. In particular, objects may disappear from the output. 1538 -- This includes not only constants and variables that the program declares 1539 -- at the source level, but also the x___L and x___U constants created to 1540 -- describe the lower and upper bounds of subtypes with dynamic bounds. 1541 -- This means for example, that array bounds may disappear if optimization 1542 -- is turned on. The debugger is expected to recognize that these constants 1543 -- are missing and deal as best as it can with the limited information 1544 -- available. 1545 1546 --------------------------------- 1547 -- GNAT Extensions to DWARF2/3 -- 1548 --------------------------------- 1549 1550 -- If the compiler switch "-gdwarf+" is specified, GNAT Vendor extensions 1551 -- to DWARF2/3 are generated, with the following variations from the above 1552 -- specification. 1553 1554 -- Change in the contents of the DW_AT_name attribute 1555 1556 -- The operators are represented in their natural form. (for example, 1557 -- the addition operator is written as "+" instead of "Oadd"). The 1558 -- component separator is "." instead of "__" 1559 1560 -- Introduction of DW_AT_GNAT_encoding, encoded with value 0x2301 1561 1562 -- Any debugging information entry representing a program entity, named 1563 -- or implicit, may have a DW_AT_GNAT_encoding attribute. The value of 1564 -- this attribute is a string representing the suffix internally added 1565 -- by GNAT for various purposes, mainly for representing debug 1566 -- information compatible with other formats. In particular this is 1567 -- useful for IDEs which need to filter out information internal to 1568 -- GNAT from their graphical interfaces. 1569 1570 -- If a debugging information entry has multiple encodings, all of them 1571 -- will be listed in DW_AT_GNAT_encoding using the list separator ':'. 1572 1573 -- Introduction of DW_AT_GNAT_descriptive_type, encoded with value 0x2302 1574 1575 -- Any debugging information entry representing a type may have a 1576 -- DW_AT_GNAT_descriptive_type attribute whose value is a reference, 1577 -- pointing to a debugging information entry representing another type 1578 -- associated to the type. 1579 1580 -- Modification of the contents of the DW_AT_producer string 1581 1582 -- When emitting full GNAT Vendor extensions to DWARF2/3, "-gdwarf+" 1583 -- is appended to the DW_AT_producer string. 1584 -- 1585 -- When emitting only DW_AT_GNAT_descriptive_type, "-gdwarf+-" is 1586 -- appended to the DW_AT_producer string. 1587 1588end Exp_Dbug; 1589