1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.5 03/11/81"; */ 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 extern struct nl *Fp; 231 extern int pnumcnt; 232 233 #ifdef PTREE 234 # include "pTree.h" 235 #endif PTREE 236 struct nl { 237 char *symbol; 238 char class, nl_flags; 239 #ifdef PC 240 char ext_flags; /* an extra flag is used for externals */ 241 #endif PC 242 struct nl *type; 243 struct nl *chain, *nl_next; 244 int *ptr[4]; 245 #ifdef PI 246 int entloc; 247 #endif PI 248 # ifdef PTREE 249 pPointer inTree; 250 # endif PTREE 251 } *nlp, *disptab[077+1]; 252 253 extern struct nl nl[INL]; 254 255 struct { 256 char *symbol; 257 char class, nl_flags; 258 #ifdef PC 259 char ext_flags; 260 #endif 261 struct nl *type; 262 struct nl *chain, *nl_next; 263 double real; 264 }; 265 266 struct { 267 char *symbol; 268 char class, nl_block; 269 #ifdef PC 270 char ext_flags; 271 #endif 272 struct nl *type; 273 struct nl *chain, *nl_next; 274 long range[2]; 275 }; 276 277 struct { 278 char *symbol; 279 char class, nl_flags; 280 #ifdef PC 281 char ext_flags; 282 #endif 283 struct nl *type; 284 struct nl *chain, *nl_next; 285 int value[5]; 286 }; 287 288 /* 289 * NL FLAGS BITS 290 * 291 * Definitions of the usage of the bits in 292 * the nl_flags byte. Note that the low 5 bits of the 293 * byte are the "nl_block" and that some classes make use 294 * of this byte as a "width". 295 * 296 * The only non-obvious bit definition here is "NFILES" 297 * which records whether a structure contains any files. 298 * Such structures are not allowed to be dynamically allocated. 299 */ 300 #define NUSED 0100 301 #define NMOD 0040 302 #define NFORWD 0200 303 #define NFILES 0200 304 305 #ifdef PC 306 #define NEXTERN 0001 /* flag used to mark external funcs and procs */ 307 #endif 308 309 /* 310 * Definition of the commonly used "value" fields. 311 * The most important one is NL_OFFS which gives 312 * the offset of a variable in its stack mark. 313 */ 314 #define NL_OFFS 0 315 316 #define NL_CNTR 1 317 #define NL_FVAR 3 318 319 #define NL_GOLEV 2 320 #define NL_GOLINE 3 321 #define NL_FORV 1 322 323 #define NL_FLDSZ 1 324 #define NL_VARNT 2 325 #define NL_VTOREC 2 326 #define NL_TAG 3 327 328 #define NL_ELABEL 4 329 330 /* 331 * For BADUSE nl structures, NL_KINDS is a bit vector 332 * indicating the kinds of illegal usages complained about 333 * so far. For kind of bad use "kind", "1 << kind" is set. 334 * The low bit is reserved as ISUNDEF to indicate whether 335 * this identifier is totally undefined. 336 */ 337 #define NL_KINDS 0 338 339 #define ISUNDEF 1 340 341 /* 342 * NAMELIST CLASSES 343 * 344 * The following are the namelist classes. 345 * Different classes make use of the value fields 346 * of the namelist in different ways. 347 * 348 * The namelist should be redesigned by providing 349 * a number of structure definitions with one corresponding 350 * to each namelist class, ala a variant record in Pascal. 351 */ 352 #define BADUSE 0 353 #define CONST 1 354 #define TYPE 2 355 #define VAR 3 356 #define ARRAY 4 357 #define PTRFILE 5 358 #define RECORD 6 359 #define FIELD 7 360 #define PROC 8 361 #define FUNC 9 362 #define FVAR 10 363 #define REF 11 364 #define PTR 12 365 #define FILET 13 366 #define SET 14 367 #define RANGE 15 368 #define LABEL 16 369 #define WITHPTR 17 370 #define SCAL 18 371 #define STR 19 372 #define PROG 20 373 #define IMPROPER 21 374 #define VARNT 22 375 #define FPROC 23 376 #define FFUNC 24 377 378 /* 379 * Clnames points to an array of names for the 380 * namelist classes. 381 */ 382 char **clnames; 383 384 /* 385 * PRE-DEFINED NAMELIST OFFSETS 386 * 387 * The following are the namelist offsets for the 388 * primitive types. The ones which are negative 389 * don't actually exist, but are generated and tested 390 * internally. These definitions are sensitive to the 391 * initializations in nl.c. 392 */ 393 #define TFIRST -7 394 #define TFILE -7 395 #define TREC -6 396 #define TARY -5 397 #define TSCAL -4 398 #define TPTR -3 399 #define TSET -2 400 #define TSTR -1 401 #define NIL 0 402 #define TBOOL 1 403 #define TCHAR 2 404 #define TINT 3 405 #define TDOUBLE 4 406 #define TNIL 5 407 #define T1INT 6 408 #define T2INT 7 409 #define T4INT 8 410 #define T1CHAR 9 411 #define T1BOOL 10 412 #define T8REAL 11 413 #define TLAST 11 414 415 /* 416 * SEMANTIC DEFINITIONS 417 */ 418 419 /* 420 * NOCON and SAWCON are flags in the tree telling whether 421 * a constant set is part of an expression. 422 */ 423 #define NOCON 0 424 #define SAWCON 1 425 426 /* 427 * The variable cbn gives the current block number, 428 * the variable bn is set as a side effect of a call to 429 * lookup, and is the block number of the variable which 430 * was found. 431 */ 432 short bn, cbn; 433 434 /* 435 * The variable line is the current semantic 436 * line and is set in stat.c from the numbers 437 * embedded in statement type tree nodes. 438 */ 439 short line; 440 441 /* 442 * The size of the display 443 * which defines the maximum nesting 444 * of procedures and functions allowed. 445 * Because of the flags in the current namelist 446 * this must be no greater than 32. 447 */ 448 #define DSPLYSZ 20 449 450 /* 451 * The following structure is used 452 * to keep track of the amount of variable 453 * storage required by each block. 454 * "Max" is the high water mark, "off" 455 * the current need. Temporaries for "for" 456 * loops and "with" statements are allocated 457 * in the local variable area and these 458 * numbers are thereby changed if necessary. 459 */ 460 struct om { 461 long om_max; 462 long reg_max; 463 struct tmps { 464 long om_off; 465 long reg_off; 466 } curtmps; 467 } sizes[DSPLYSZ]; 468 #define NOREG 0 469 #define REGOK 1 470 471 /* 472 * the following structure records whether a level declares 473 * any variables which are (or contain) files. 474 * this so that the runtime routines for file cleanup can be invoked. 475 */ 476 bool dfiles[ DSPLYSZ ]; 477 478 /* 479 * Structure recording information about a constant 480 * declaration. It is actually the return value from 481 * the routine "gconst", but since C doesn't support 482 * record valued functions, this is more convenient. 483 */ 484 struct { 485 struct nl *ctype; 486 short cival; 487 double crval; 488 int *cpval; 489 } con; 490 491 /* 492 * The set structure records the lower bound 493 * and upper bound with the lower bound normalized 494 * to zero when working with a set. It is set by 495 * the routine setran in var.c. 496 */ 497 struct { 498 short lwrb, uprbp; 499 } set; 500 501 /* 502 * structures of this kind are filled in by precset and used by postcset 503 * to indicate things about constant sets. 504 */ 505 struct csetstr { 506 struct nl *csettype; 507 long paircnt; 508 long singcnt; 509 bool comptime; 510 }; 511 /* 512 * The following flags are passed on calls to lvalue 513 * to indicate how the reference is to affect the usage 514 * information for the variable being referenced. 515 * MOD is used to set the NMOD flag in the namelist 516 * entry for the variable, ASGN permits diagnostics 517 * to be formed when a for variable is assigned to in 518 * the range of the loop. 519 */ 520 #define NOFLAGS 0 521 #define MOD 01 522 #define ASGN 02 523 #define NOUSE 04 524 525 /* 526 * the following flags are passed to lvalue and rvalue 527 * to tell them whether an lvalue or rvalue is required. 528 * the semantics checking is done according to the function called, 529 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 530 * and rvalue may stop short of putting out the indirection. 531 */ 532 #define LREQ 01 533 #define RREQ 02 534 535 double MAXINT; 536 double MININT; 537 538 /* 539 * Variables for generation of profile information. 540 * Monflg is set when we want to generate a profile. 541 * Gocnt record the total number of goto's and 542 * cnts records the current counter for generating 543 * COUNT operators. 544 */ 545 short gocnt; 546 short cnts; 547 548 /* 549 * Most routines call "incompat" rather than asking "!compat" 550 * for historical reasons. 551 */ 552 #define incompat !compat 553 554 /* 555 * Parts records which declaration parts have been seen. 556 * The grammar allows the "label" "const" "type" "var" and routine 557 * parts to be repeated and to be in any order, so that 558 * they can be detected semantically to give better 559 * error diagnostics. 560 */ 561 int parts[ DSPLYSZ ]; 562 563 #define LPRT 1 564 #define CPRT 2 565 #define TPRT 4 566 #define VPRT 8 567 #define RPRT 16 568 569 /* 570 * Flags for the "you used / instead of div" diagnostic 571 */ 572 bool divchk; 573 bool divflg; 574 575 bool errcnt[DSPLYSZ]; 576 577 /* 578 * Forechain links those types which are 579 * ^ sometype 580 * so that they can be evaluated later, permitting 581 * circular, recursive list structures to be defined. 582 */ 583 struct nl *forechain; 584 585 /* 586 * Withlist links all the records which are currently 587 * opened scopes because of with statements. 588 */ 589 struct nl *withlist; 590 591 struct nl *intset; 592 struct nl *input, *output; 593 struct nl *program; 594 595 /* progseen flag used by PC to determine if 596 * a routine segment is being compiled (and 597 * therefore no program statement seen) 598 */ 599 bool progseen; 600 601 602 /* 603 * STRUCTURED STATEMENT GOTO CHECKING 604 * 605 * The variable level keeps track of the current 606 * "structured statement level" when processing the statement 607 * body of blocks. This is used in the detection of goto's into 608 * structured statements in a block. 609 * 610 * Each label's namelist entry contains two pieces of information 611 * related to this check. The first `NL_GOLEV' either contains 612 * the level at which the label was declared, `NOTYET' if the label 613 * has not yet been declared, or `DEAD' if the label is dead, i.e. 614 * if we have exited the level in which the label was defined. 615 * 616 * When we discover a "goto" statement, if the label has not 617 * been defined yet, then we record the current level and the current line 618 * for a later error check. If the label has been already become "DEAD" 619 * then a reference to it is an error. Now the compiler maintains, 620 * for each block, a linked list of the labels headed by "gotos[bn]". 621 * When we exit a structured level, we perform the routine 622 * ungoto in stat.c. It notices labels whose definition levels have been 623 * exited and makes them be dead. For labels which have not yet been 624 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 625 * since the first usage of the label. It is not hard to see that the label 626 * must eventually be declared at this level or an outer level to this 627 * one or a goto into a structured statement will exist. 628 */ 629 short level; 630 struct nl *gotos[DSPLYSZ]; 631 632 #define NOTYET 10000 633 #define DEAD 10000 634 635 /* 636 * Noreach is true when the next statement will 637 * be unreachable unless something happens along 638 * (like exiting a looping construct) to save 639 * the day. 640 */ 641 bool noreach; 642 643 /* 644 * UNDEFINED VARIABLE REFERENCE STRUCTURES 645 */ 646 struct udinfo { 647 int ud_line; 648 struct udinfo *ud_next; 649 char nullch; 650 }; 651 652 /* 653 * CODE GENERATION DEFINITIONS 654 */ 655 656 /* 657 * NSTAND is or'ed onto the abstract machine opcode 658 * for non-standard built-in procedures and functions. 659 */ 660 #define NSTAND 0400 661 662 #define codeon() cgenflg++ 663 #define codeoff() --cgenflg 664 665 /* 666 * Codeline is the last lino output in the code generator. 667 * It used to be used to suppress LINO operators but no 668 * more since we now count statements. 669 * Lc is the intepreter code location counter. 670 * 671 short codeline; 672 */ 673 char *lc; 674 675 676 /* 677 * Routines which need types 678 * other than "integer" to be 679 * assumed by the compiler. 680 */ 681 double atof(); 682 long lwidth(); 683 long leven(); 684 long aryconst(); 685 long a8tol(); 686 long roundup(); 687 struct nl *lookup(); 688 double atof(); 689 int *tree(); 690 int *hash(); 691 char *alloc(); 692 int *calloc(); 693 char *savestr(); 694 struct nl *lookup1(); 695 struct nl *hdefnl(); 696 struct nl *defnl(); 697 struct nl *enter(); 698 struct nl *nlcopy(); 699 struct nl *tyrecl(); 700 struct nl *tyary(); 701 struct nl *fields(); 702 struct nl *variants(); 703 struct nl *deffld(); 704 struct nl *defvnt(); 705 struct nl *tyrec1(); 706 struct nl *reclook(); 707 struct nl *asgnop1(); 708 struct nl *gtype(); 709 struct nl *call(); 710 struct nl *lvalue(); 711 struct nl *rvalue(); 712 struct nl *cset(); 713 714 /* 715 * type cast NIL to keep lint happy (which is not so bad) 716 */ 717 #define NLNIL ( (struct nl *) NIL ) 718 719 /* 720 * Funny structures to use 721 * pointers in wild and wooly ways 722 */ 723 struct { 724 char pchar; 725 }; 726 struct { 727 short pint; 728 short pint2; 729 }; 730 struct { 731 long plong; 732 }; 733 struct { 734 double pdouble; 735 }; 736 737 #define OCT 1 738 #define HEX 2 739 740 /* 741 * MAIN PROGRAM VARIABLES, MISCELLANY 742 */ 743 744 /* 745 * Variables forming a data base referencing 746 * the command line arguments with the "i" option, e.g. 747 * in "pi -i scanner.i compiler.p". 748 */ 749 char **pflist; 750 short pflstc; 751 short pfcnt; 752 753 char *filename; /* current source file name */ 754 long tvec; 755 extern char *snark; /* SNARK */ 756 extern char *classes[ ]; /* maps namelist classes to string names */ 757 758 #define derror error 759 760 #ifdef PC 761 762 /* 763 * the current function number, for [ lines 764 */ 765 int ftnno; 766 767 /* 768 * the pc output stream 769 */ 770 FILE *pcstream; 771 772 #endif PC 773