1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.4 03/08/81"; */ 4 5 #define DEBUG 6 #define CONSETS 7 #define CHAR 8 #define STATIC 9 #define hp21mx 0 10 11 #include <stdio.h> 12 #include <sys/types.h> 13 14 typedef enum {FALSE, TRUE} bool; 15 16 /* 17 * Option flags 18 * 19 * The following options are recognized in the text of the program 20 * and also on the command line: 21 * 22 * b block buffer the file output 23 * 24 * i make a listing of the procedures and functions in 25 * the following include files 26 * 27 * l make a listing of the program 28 * 29 * n place each include file on a new page with a header 30 * 31 * p disable post mortem and statement limit counting 32 * 33 * t disable run-time tests 34 * 35 * u card image mode; only first 72 chars of input count 36 * 37 * w suppress special diagnostic warnings 38 * 39 * z generate counters for an execution profile 40 */ 41 #ifdef DEBUG 42 bool fulltrace, errtrace, testtrace, yyunique; 43 #endif DEBUG 44 45 /* 46 * Each option has a stack of 17 option values, with opts giving 47 * the current, top value, and optstk the value beneath it. 48 * One refers to option `l' as, e.g., opt('l') in the text for clarity. 49 */ 50 char opts[ 'z' - 'A' + 1]; 51 short optstk[ 'z' - 'A' + 1]; 52 53 #define opt(c) opts[c-'A'] 54 55 /* 56 * Monflg is set when we are generating 57 * a pxp profile. this is set by the -z command line option. 58 */ 59 bool monflg; 60 61 /* 62 * profflag is set when we are generating a prof profile. 63 * this is set by the -p command line option. 64 */ 65 bool profflag; 66 67 68 /* 69 * NOTES ON THE DYNAMIC NATURE OF THE DATA STRUCTURES 70 * 71 * Pi uses expandable tables for 72 * its namelist (symbol table), string table 73 * hash table, and parse tree space. The following 74 * definitions specify the size of the increments 75 * for these items in fundamental units so that 76 * each uses approximately 1024 bytes. 77 */ 78 79 #define STRINC 1024 /* string space increment */ 80 #define TRINC 512 /* tree space increment */ 81 #define HASHINC 509 /* hash table size in words, each increment */ 82 #define NLINC 56 /* namelist increment size in nl structs */ 83 84 /* 85 * The initial sizes of the structures. 86 * These should be large enough to compile 87 * an "average" sized program so as to minimize 88 * storage requests. 89 * On a small system or and 11/34 or 11/40 90 * these numbers can be trimmed to make the 91 * compiler smaller. 92 */ 93 #define ITREE 2000 94 #define INL 200 95 #define IHASH 509 96 97 /* 98 * The following limits on hash and tree tables currently 99 * allow approximately 1200 symbols and 20k words of tree 100 * space. The fundamental limit of 64k total data space 101 * should be exceeded well before these are full. 102 */ 103 /* 104 * TABLE_MULTIPLIER is for uniformly increasing the sizes of the tables 105 */ 106 #ifdef VAX 107 #define TABLE_MULTIPLIER 8 108 #else 109 #define TABLE_MULTIPLIER 1 110 #endif VAX 111 #define MAXHASH (4 * TABLE_MULTIPLIER) 112 #define MAXNL (12 * TABLE_MULTIPLIER) 113 #define MAXTREE (30 * TABLE_MULTIPLIER) 114 /* 115 * MAXDEPTH is the depth of the parse stack. 116 * STACK_MULTIPLIER is for increasing its size. 117 */ 118 #ifdef VAX 119 #define STACK_MULTIPLIER 8 120 #else 121 #define STACK_MULTIPLIER 1 122 #endif VAX 123 #define MAXDEPTH ( 150 * STACK_MULTIPLIER ) 124 125 /* 126 * ERROR RELATED DEFINITIONS 127 */ 128 129 /* 130 * Exit statuses to pexit 131 * 132 * AOK 133 * ERRS Compilation errors inhibit obj productin 134 * NOSTART Errors before we ever got started 135 * DIED We ran out of memory or some such 136 */ 137 #define AOK 0 138 #define ERRS 1 139 #define NOSTART 2 140 #define DIED 3 141 142 bool Recovery; 143 144 #define eholdnl() Eholdnl = 1 145 #define nocascade() Enocascade = 1 146 147 bool Eholdnl, Enocascade; 148 149 150 /* 151 * The flag eflg is set whenever we have a hard error. 152 * The character in errpfx will precede the next error message. 153 * When cgenflg is set code generation is suppressed. 154 * This happens whenver we have an error (i.e. if eflg is set) 155 * and when we are walking the tree to determine types only. 156 */ 157 bool eflg; 158 char errpfx; 159 160 #define setpfx(x) errpfx = x 161 162 #define standard() setpfx('s') 163 #define warning() setpfx('w') 164 #define recovered() setpfx('e') 165 166 int cgenflg; 167 168 169 /* 170 * The flag syneflg is used to suppress the diagnostics of the form 171 * E 10 a, defined in someprocedure, is neither used nor set 172 * when there were syntax errors in "someprocedure". 173 * In this case, it is likely that these warinings would be spurious. 174 */ 175 bool syneflg; 176 177 /* 178 * The compiler keeps its error messages in a file. 179 * The variable efil is the unit number on which 180 * this file is open for reading of error message text. 181 * Similarly, the file ofil is the unit of the file 182 * "obj" where we write the interpreter code. 183 */ 184 short efil; 185 short ofil; 186 short obuf[518]; 187 188 bool Enoline; 189 #define elineoff() Enoline = TRUE 190 #define elineon() Enoline = FALSE 191 192 193 /* 194 * SYMBOL TABLE STRUCTURE DEFINITIONS 195 * 196 * The symbol table is henceforth referred to as the "namelist". 197 * It consists of a number of structures of the form "nl" below. 198 * These are contained in a number of segments of the symbol 199 * table which are dynamically allocated as needed. 200 * The major namelist manipulation routines are contained in the 201 * file "nl.c". 202 * 203 * The major components of a namelist entry are the "symbol", giving 204 * a pointer into the string table for the string associated with this 205 * entry and the "class" which tells which of the (currently 19) 206 * possible types of structure this is. 207 * 208 * Many of the classes use the "type" field for a pointer to the type 209 * which the entry has. 210 * 211 * Other pieces of information in more than one class include the block 212 * in which the symbol is defined, flags indicating whether the symbol 213 * has been used and whether it has been assigned to, etc. 214 * 215 * A more complete discussion of the features of the namelist is impossible 216 * here as it would be too voluminous. Refer to the "PI 1.0 Implementation 217 * Notes" for more details. 218 */ 219 220 /* 221 * The basic namelist structure. 222 * There are also two other variants, defining the real 223 * field as longs or integers given below. 224 * 225 * The array disptab defines the hash header for the symbol table. 226 * Symbols are hashed based on the low 6 bits of their pointer into 227 * the string table; see the routines in the file "lookup.c" and also "fdec.c" 228 * especially "funcend". 229 */ 230 #ifdef PTREE 231 # include "pTree.h" 232 #endif PTREE 233 struct nl { 234 char *symbol; 235 char class, nl_flags; 236 #ifdef PC 237 char ext_flags; /* an extra flag is used for externals */ 238 #endif PC 239 struct nl *type; 240 struct nl *chain, *nl_next; 241 int *ptr[4]; 242 #ifdef PI 243 int entloc; 244 #endif PI 245 # ifdef PTREE 246 pPointer inTree; 247 # endif PTREE 248 } *nlp, *disptab[077+1]; 249 250 extern struct nl nl[INL]; 251 252 struct { 253 char *symbol; 254 char class, nl_flags; 255 #ifdef PC 256 char ext_flags; 257 #endif 258 struct nl *type; 259 struct nl *chain, *nl_next; 260 double real; 261 }; 262 263 struct { 264 char *symbol; 265 char class, nl_block; 266 #ifdef PC 267 char ext_flags; 268 #endif 269 struct nl *type; 270 struct nl *chain, *nl_next; 271 long range[2]; 272 }; 273 274 struct { 275 char *symbol; 276 char class, nl_flags; 277 #ifdef PC 278 char ext_flags; 279 #endif 280 struct nl *type; 281 struct nl *chain, *nl_next; 282 int value[5]; 283 }; 284 285 /* 286 * NL FLAGS BITS 287 * 288 * Definitions of the usage of the bits in 289 * the nl_flags byte. Note that the low 5 bits of the 290 * byte are the "nl_block" and that some classes make use 291 * of this byte as a "width". 292 * 293 * The only non-obvious bit definition here is "NFILES" 294 * which records whether a structure contains any files. 295 * Such structures are not allowed to be dynamically allocated. 296 */ 297 #define NUSED 0100 298 #define NMOD 0040 299 #define NFORWD 0200 300 #define NFILES 0200 301 302 #ifdef PC 303 #define NEXTERN 0001 /* flag used to mark external funcs and procs */ 304 #endif 305 306 /* 307 * Definition of the commonly used "value" fields. 308 * The most important one is NL_OFFS which gives 309 * the offset of a variable in its stack mark. 310 */ 311 #define NL_OFFS 0 312 313 #define NL_CNTR 1 314 #define NL_FVAR 3 315 316 #define NL_GOLEV 2 317 #define NL_GOLINE 3 318 #define NL_FORV 1 319 320 #define NL_FLDSZ 1 321 #define NL_VARNT 2 322 #define NL_VTOREC 2 323 #define NL_TAG 3 324 325 #define NL_ELABEL 4 326 327 /* 328 * For BADUSE nl structures, NL_KINDS is a bit vector 329 * indicating the kinds of illegal usages complained about 330 * so far. For kind of bad use "kind", "1 << kind" is set. 331 * The low bit is reserved as ISUNDEF to indicate whether 332 * this identifier is totally undefined. 333 */ 334 #define NL_KINDS 0 335 336 #define ISUNDEF 1 337 338 /* 339 * NAMELIST CLASSES 340 * 341 * The following are the namelist classes. 342 * Different classes make use of the value fields 343 * of the namelist in different ways. 344 * 345 * The namelist should be redesigned by providing 346 * a number of structure definitions with one corresponding 347 * to each namelist class, ala a variant record in Pascal. 348 */ 349 #define BADUSE 0 350 #define CONST 1 351 #define TYPE 2 352 #define VAR 3 353 #define ARRAY 4 354 #define PTRFILE 5 355 #define RECORD 6 356 #define FIELD 7 357 #define PROC 8 358 #define FUNC 9 359 #define FVAR 10 360 #define REF 11 361 #define PTR 12 362 #define FILET 13 363 #define SET 14 364 #define RANGE 15 365 #define LABEL 16 366 #define WITHPTR 17 367 #define SCAL 18 368 #define STR 19 369 #define PROG 20 370 #define IMPROPER 21 371 #define VARNT 22 372 #define FPROC 23 373 #define FFUNC 24 374 375 /* 376 * Clnames points to an array of names for the 377 * namelist classes. 378 */ 379 char **clnames; 380 381 /* 382 * PRE-DEFINED NAMELIST OFFSETS 383 * 384 * The following are the namelist offsets for the 385 * primitive types. The ones which are negative 386 * don't actually exist, but are generated and tested 387 * internally. These definitions are sensitive to the 388 * initializations in nl.c. 389 */ 390 #define TFIRST -7 391 #define TFILE -7 392 #define TREC -6 393 #define TARY -5 394 #define TSCAL -4 395 #define TPTR -3 396 #define TSET -2 397 #define TSTR -1 398 #define NIL 0 399 #define TBOOL 1 400 #define TCHAR 2 401 #define TINT 3 402 #define TDOUBLE 4 403 #define TNIL 5 404 #define T1INT 6 405 #define T2INT 7 406 #define T4INT 8 407 #define T1CHAR 9 408 #define T1BOOL 10 409 #define T8REAL 11 410 #define TLAST 11 411 412 /* 413 * SEMANTIC DEFINITIONS 414 */ 415 416 /* 417 * NOCON and SAWCON are flags in the tree telling whether 418 * a constant set is part of an expression. 419 */ 420 #define NOCON 0 421 #define SAWCON 1 422 423 /* 424 * The variable cbn gives the current block number, 425 * the variable bn is set as a side effect of a call to 426 * lookup, and is the block number of the variable which 427 * was found. 428 */ 429 short bn, cbn; 430 431 /* 432 * The variable line is the current semantic 433 * line and is set in stat.c from the numbers 434 * embedded in statement type tree nodes. 435 */ 436 short line; 437 438 /* 439 * The size of the display 440 * which defines the maximum nesting 441 * of procedures and functions allowed. 442 * Because of the flags in the current namelist 443 * this must be no greater than 32. 444 */ 445 #define DSPLYSZ 20 446 447 /* 448 * The following structure is used 449 * to keep track of the amount of variable 450 * storage required by each block. 451 * "Max" is the high water mark, "off" 452 * the current need. Temporaries for "for" 453 * loops and "with" statements are allocated 454 * in the local variable area and these 455 * numbers are thereby changed if necessary. 456 */ 457 struct om { 458 long om_off; 459 long om_max; 460 } sizes[DSPLYSZ]; 461 462 /* 463 * the following structure records whether a level declares 464 * any variables which are (or contain) files. 465 * this so that the runtime routines for file cleanup can be invoked. 466 */ 467 bool dfiles[ DSPLYSZ ]; 468 469 /* 470 * Structure recording information about a constant 471 * declaration. It is actually the return value from 472 * the routine "gconst", but since C doesn't support 473 * record valued functions, this is more convenient. 474 */ 475 struct { 476 struct nl *ctype; 477 short cival; 478 double crval; 479 int *cpval; 480 } con; 481 482 /* 483 * The set structure records the lower bound 484 * and upper bound with the lower bound normalized 485 * to zero when working with a set. It is set by 486 * the routine setran in var.c. 487 */ 488 struct { 489 short lwrb, uprbp; 490 } set; 491 492 /* 493 * structures of this kind are filled in by precset and used by postcset 494 * to indicate things about constant sets. 495 */ 496 struct csetstr { 497 struct nl *csettype; 498 long paircnt; 499 long singcnt; 500 bool comptime; 501 }; 502 /* 503 * The following flags are passed on calls to lvalue 504 * to indicate how the reference is to affect the usage 505 * information for the variable being referenced. 506 * MOD is used to set the NMOD flag in the namelist 507 * entry for the variable, ASGN permits diagnostics 508 * to be formed when a for variable is assigned to in 509 * the range of the loop. 510 */ 511 #define NOFLAGS 0 512 #define MOD 01 513 #define ASGN 02 514 #define NOUSE 04 515 516 /* 517 * the following flags are passed to lvalue and rvalue 518 * to tell them whether an lvalue or rvalue is required. 519 * the semantics checking is done according to the function called, 520 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 521 * and rvalue may stop short of putting out the indirection. 522 */ 523 #define LREQ 01 524 #define RREQ 02 525 526 double MAXINT; 527 double MININT; 528 529 /* 530 * Variables for generation of profile information. 531 * Monflg is set when we want to generate a profile. 532 * Gocnt record the total number of goto's and 533 * cnts records the current counter for generating 534 * COUNT operators. 535 */ 536 short gocnt; 537 short cnts; 538 539 /* 540 * Most routines call "incompat" rather than asking "!compat" 541 * for historical reasons. 542 */ 543 #define incompat !compat 544 545 /* 546 * Parts records which declaration parts have been seen. 547 * The grammar allows the "label" "const" "type" "var" and routine 548 * parts to be repeated and to be in any order, so that 549 * they can be detected semantically to give better 550 * error diagnostics. 551 */ 552 int parts[ DSPLYSZ ]; 553 554 #define LPRT 1 555 #define CPRT 2 556 #define TPRT 4 557 #define VPRT 8 558 #define RPRT 16 559 560 /* 561 * Flags for the "you used / instead of div" diagnostic 562 */ 563 bool divchk; 564 bool divflg; 565 566 bool errcnt[DSPLYSZ]; 567 568 /* 569 * Forechain links those types which are 570 * ^ sometype 571 * so that they can be evaluated later, permitting 572 * circular, recursive list structures to be defined. 573 */ 574 struct nl *forechain; 575 576 /* 577 * Withlist links all the records which are currently 578 * opened scopes because of with statements. 579 */ 580 struct nl *withlist; 581 582 struct nl *intset; 583 struct nl *input, *output; 584 struct nl *program; 585 586 /* progseen flag used by PC to determine if 587 * a routine segment is being compiled (and 588 * therefore no program statement seen) 589 */ 590 bool progseen; 591 592 593 /* 594 * STRUCTURED STATEMENT GOTO CHECKING 595 * 596 * The variable level keeps track of the current 597 * "structured statement level" when processing the statement 598 * body of blocks. This is used in the detection of goto's into 599 * structured statements in a block. 600 * 601 * Each label's namelist entry contains two pieces of information 602 * related to this check. The first `NL_GOLEV' either contains 603 * the level at which the label was declared, `NOTYET' if the label 604 * has not yet been declared, or `DEAD' if the label is dead, i.e. 605 * if we have exited the level in which the label was defined. 606 * 607 * When we discover a "goto" statement, if the label has not 608 * been defined yet, then we record the current level and the current line 609 * for a later error check. If the label has been already become "DEAD" 610 * then a reference to it is an error. Now the compiler maintains, 611 * for each block, a linked list of the labels headed by "gotos[bn]". 612 * When we exit a structured level, we perform the routine 613 * ungoto in stat.c. It notices labels whose definition levels have been 614 * exited and makes them be dead. For labels which have not yet been 615 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 616 * since the first usage of the label. It is not hard to see that the label 617 * must eventually be declared at this level or an outer level to this 618 * one or a goto into a structured statement will exist. 619 */ 620 short level; 621 struct nl *gotos[DSPLYSZ]; 622 623 #define NOTYET 10000 624 #define DEAD 10000 625 626 /* 627 * Noreach is true when the next statement will 628 * be unreachable unless something happens along 629 * (like exiting a looping construct) to save 630 * the day. 631 */ 632 bool noreach; 633 634 /* 635 * UNDEFINED VARIABLE REFERENCE STRUCTURES 636 */ 637 struct udinfo { 638 int ud_line; 639 struct udinfo *ud_next; 640 char nullch; 641 }; 642 643 /* 644 * CODE GENERATION DEFINITIONS 645 */ 646 647 /* 648 * NSTAND is or'ed onto the abstract machine opcode 649 * for non-standard built-in procedures and functions. 650 */ 651 #define NSTAND 0400 652 653 #define codeon() cgenflg++ 654 #define codeoff() --cgenflg 655 656 /* 657 * Codeline is the last lino output in the code generator. 658 * It used to be used to suppress LINO operators but no 659 * more since we now count statements. 660 * Lc is the intepreter code location counter. 661 * 662 short codeline; 663 */ 664 char *lc; 665 666 667 /* 668 * Routines which need types 669 * other than "integer" to be 670 * assumed by the compiler. 671 */ 672 double atof(); 673 long lwidth(); 674 long leven(); 675 long aryconst(); 676 long a8tol(); 677 long roundup(); 678 struct nl *lookup(); 679 double atof(); 680 int *tree(); 681 int *hash(); 682 char *alloc(); 683 int *calloc(); 684 char *savestr(); 685 struct nl *lookup1(); 686 struct nl *hdefnl(); 687 struct nl *defnl(); 688 struct nl *enter(); 689 struct nl *nlcopy(); 690 struct nl *tyrecl(); 691 struct nl *tyary(); 692 struct nl *fields(); 693 struct nl *variants(); 694 struct nl *deffld(); 695 struct nl *defvnt(); 696 struct nl *tyrec1(); 697 struct nl *reclook(); 698 struct nl *asgnop1(); 699 struct nl *gtype(); 700 struct nl *call(); 701 struct nl *lvalue(); 702 struct nl *rvalue(); 703 struct nl *cset(); 704 705 /* 706 * type cast NIL to keep lint happy (which is not so bad) 707 */ 708 #define NLNIL ( (struct nl *) NIL ) 709 710 /* 711 * Funny structures to use 712 * pointers in wild and wooly ways 713 */ 714 struct { 715 char pchar; 716 }; 717 struct { 718 short pint; 719 short pint2; 720 }; 721 struct { 722 long plong; 723 }; 724 struct { 725 double pdouble; 726 }; 727 728 #define OCT 1 729 #define HEX 2 730 731 /* 732 * MAIN PROGRAM VARIABLES, MISCELLANY 733 */ 734 735 /* 736 * Variables forming a data base referencing 737 * the command line arguments with the "i" option, e.g. 738 * in "pi -i scanner.i compiler.p". 739 */ 740 char **pflist; 741 short pflstc; 742 short pfcnt; 743 744 char *filename; /* current source file name */ 745 long tvec; 746 extern char *snark; /* SNARK */ 747 extern char *classes[ ]; /* maps namelist classes to string names */ 748 749 #define derror error 750 751 #ifdef PC 752 753 /* 754 * the current function number, for [ lines 755 */ 756 int ftnno; 757 758 /* 759 * the pc output stream 760 */ 761 FILE *pcstream; 762 763 #endif PC 764