1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.7 03/16/81"; */ 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 #define NL_SOFFS 4 323 324 #define NL_FLDSZ 1 325 #define NL_VARNT 2 326 #define NL_VTOREC 2 327 #define NL_TAG 3 328 329 #define NL_ELABEL 4 330 331 /* 332 * For BADUSE nl structures, NL_KINDS is a bit vector 333 * indicating the kinds of illegal usages complained about 334 * so far. For kind of bad use "kind", "1 << kind" is set. 335 * The low bit is reserved as ISUNDEF to indicate whether 336 * this identifier is totally undefined. 337 */ 338 #define NL_KINDS 0 339 340 #define ISUNDEF 1 341 342 /* 343 * Variables may reside on the stack as formals or as locals, 344 * or as register temporaries 345 */ 346 #define PARAMVAR 1 347 #define LOCALVAR 2 348 #define REGVAR 3 349 350 /* 351 * NAMELIST CLASSES 352 * 353 * The following are the namelist classes. 354 * Different classes make use of the value fields 355 * of the namelist in different ways. 356 * 357 * The namelist should be redesigned by providing 358 * a number of structure definitions with one corresponding 359 * to each namelist class, ala a variant record in Pascal. 360 */ 361 #define BADUSE 0 362 #define CONST 1 363 #define TYPE 2 364 #define VAR 3 365 #define ARRAY 4 366 #define PTRFILE 5 367 #define RECORD 6 368 #define FIELD 7 369 #define PROC 8 370 #define FUNC 9 371 #define FVAR 10 372 #define REF 11 373 #define PTR 12 374 #define FILET 13 375 #define SET 14 376 #define RANGE 15 377 #define LABEL 16 378 #define WITHPTR 17 379 #define SCAL 18 380 #define STR 19 381 #define PROG 20 382 #define IMPROPER 21 383 #define VARNT 22 384 #define FPROC 23 385 #define FFUNC 24 386 387 /* 388 * Clnames points to an array of names for the 389 * namelist classes. 390 */ 391 char **clnames; 392 393 /* 394 * PRE-DEFINED NAMELIST OFFSETS 395 * 396 * The following are the namelist offsets for the 397 * primitive types. The ones which are negative 398 * don't actually exist, but are generated and tested 399 * internally. These definitions are sensitive to the 400 * initializations in nl.c. 401 */ 402 #define TFIRST -7 403 #define TFILE -7 404 #define TREC -6 405 #define TARY -5 406 #define TSCAL -4 407 #define TPTR -3 408 #define TSET -2 409 #define TSTR -1 410 #define NIL 0 411 #define TBOOL 1 412 #define TCHAR 2 413 #define TINT 3 414 #define TDOUBLE 4 415 #define TNIL 5 416 #define T1INT 6 417 #define T2INT 7 418 #define T4INT 8 419 #define T1CHAR 9 420 #define T1BOOL 10 421 #define T8REAL 11 422 #define TLAST 11 423 424 /* 425 * SEMANTIC DEFINITIONS 426 */ 427 428 /* 429 * NOCON and SAWCON are flags in the tree telling whether 430 * a constant set is part of an expression. 431 */ 432 #define NOCON 0 433 #define SAWCON 1 434 435 /* 436 * The variable cbn gives the current block number, 437 * the variable bn is set as a side effect of a call to 438 * lookup, and is the block number of the variable which 439 * was found. 440 */ 441 short bn, cbn; 442 443 /* 444 * The variable line is the current semantic 445 * line and is set in stat.c from the numbers 446 * embedded in statement type tree nodes. 447 */ 448 short line; 449 450 /* 451 * The size of the display 452 * which defines the maximum nesting 453 * of procedures and functions allowed. 454 * Because of the flags in the current namelist 455 * this must be no greater than 32. 456 */ 457 #define DSPLYSZ 20 458 459 /* 460 * The following structure is used 461 * to keep track of the amount of variable 462 * storage required by each block. 463 * "Max" is the high water mark, "off" 464 * the current need. Temporaries for "for" 465 * loops and "with" statements are allocated 466 * in the local variable area and these 467 * numbers are thereby changed if necessary. 468 */ 469 struct om { 470 long om_max; 471 long reg_max; 472 struct tmps { 473 long om_off; 474 long reg_off; 475 } curtmps; 476 } sizes[DSPLYSZ]; 477 #define NOREG 0 478 #define REGOK 1 479 480 /* 481 * the following structure records whether a level declares 482 * any variables which are (or contain) files. 483 * this so that the runtime routines for file cleanup can be invoked. 484 */ 485 bool dfiles[ DSPLYSZ ]; 486 487 /* 488 * Structure recording information about a constant 489 * declaration. It is actually the return value from 490 * the routine "gconst", but since C doesn't support 491 * record valued functions, this is more convenient. 492 */ 493 struct { 494 struct nl *ctype; 495 short cival; 496 double crval; 497 int *cpval; 498 } con; 499 500 /* 501 * The set structure records the lower bound 502 * and upper bound with the lower bound normalized 503 * to zero when working with a set. It is set by 504 * the routine setran in var.c. 505 */ 506 struct { 507 short lwrb, uprbp; 508 } set; 509 510 /* 511 * structures of this kind are filled in by precset and used by postcset 512 * to indicate things about constant sets. 513 */ 514 struct csetstr { 515 struct nl *csettype; 516 long paircnt; 517 long singcnt; 518 bool comptime; 519 }; 520 /* 521 * The following flags are passed on calls to lvalue 522 * to indicate how the reference is to affect the usage 523 * information for the variable being referenced. 524 * MOD is used to set the NMOD flag in the namelist 525 * entry for the variable, ASGN permits diagnostics 526 * to be formed when a for variable is assigned to in 527 * the range of the loop. 528 */ 529 #define NOFLAGS 0 530 #define MOD 01 531 #define ASGN 02 532 #define NOUSE 04 533 534 /* 535 * the following flags are passed to lvalue and rvalue 536 * to tell them whether an lvalue or rvalue is required. 537 * the semantics checking is done according to the function called, 538 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 539 * and rvalue may stop short of putting out the indirection. 540 */ 541 #define LREQ 01 542 #define RREQ 02 543 544 double MAXINT; 545 double MININT; 546 547 /* 548 * Variables for generation of profile information. 549 * Monflg is set when we want to generate a profile. 550 * Gocnt record the total number of goto's and 551 * cnts records the current counter for generating 552 * COUNT operators. 553 */ 554 short gocnt; 555 short cnts; 556 557 /* 558 * Most routines call "incompat" rather than asking "!compat" 559 * for historical reasons. 560 */ 561 #define incompat !compat 562 563 /* 564 * Parts records which declaration parts have been seen. 565 * The grammar allows the "label" "const" "type" "var" and routine 566 * parts to be repeated and to be in any order, so that 567 * they can be detected semantically to give better 568 * error diagnostics. 569 */ 570 int parts[ DSPLYSZ ]; 571 572 #define LPRT 1 573 #define CPRT 2 574 #define TPRT 4 575 #define VPRT 8 576 #define RPRT 16 577 578 /* 579 * Flags for the "you used / instead of div" diagnostic 580 */ 581 bool divchk; 582 bool divflg; 583 584 bool errcnt[DSPLYSZ]; 585 586 /* 587 * Forechain links those types which are 588 * ^ sometype 589 * so that they can be evaluated later, permitting 590 * circular, recursive list structures to be defined. 591 */ 592 struct nl *forechain; 593 594 /* 595 * Withlist links all the records which are currently 596 * opened scopes because of with statements. 597 */ 598 struct nl *withlist; 599 600 struct nl *intset; 601 struct nl *input, *output; 602 struct nl *program; 603 604 /* progseen flag used by PC to determine if 605 * a routine segment is being compiled (and 606 * therefore no program statement seen) 607 */ 608 bool progseen; 609 610 611 /* 612 * STRUCTURED STATEMENT GOTO CHECKING 613 * 614 * The variable level keeps track of the current 615 * "structured statement level" when processing the statement 616 * body of blocks. This is used in the detection of goto's into 617 * structured statements in a block. 618 * 619 * Each label's namelist entry contains two pieces of information 620 * related to this check. The first `NL_GOLEV' either contains 621 * the level at which the label was declared, `NOTYET' if the label 622 * has not yet been declared, or `DEAD' if the label is dead, i.e. 623 * if we have exited the level in which the label was defined. 624 * 625 * When we discover a "goto" statement, if the label has not 626 * been defined yet, then we record the current level and the current line 627 * for a later error check. If the label has been already become "DEAD" 628 * then a reference to it is an error. Now the compiler maintains, 629 * for each block, a linked list of the labels headed by "gotos[bn]". 630 * When we exit a structured level, we perform the routine 631 * ungoto in stat.c. It notices labels whose definition levels have been 632 * exited and makes them be dead. For labels which have not yet been 633 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 634 * since the first usage of the label. It is not hard to see that the label 635 * must eventually be declared at this level or an outer level to this 636 * one or a goto into a structured statement will exist. 637 */ 638 short level; 639 struct nl *gotos[DSPLYSZ]; 640 641 #define NOTYET 10000 642 #define DEAD 10000 643 644 /* 645 * Noreach is true when the next statement will 646 * be unreachable unless something happens along 647 * (like exiting a looping construct) to save 648 * the day. 649 */ 650 bool noreach; 651 652 /* 653 * UNDEFINED VARIABLE REFERENCE STRUCTURES 654 */ 655 struct udinfo { 656 int ud_line; 657 struct udinfo *ud_next; 658 char nullch; 659 }; 660 661 /* 662 * CODE GENERATION DEFINITIONS 663 */ 664 665 /* 666 * NSTAND is or'ed onto the abstract machine opcode 667 * for non-standard built-in procedures and functions. 668 */ 669 #define NSTAND 0400 670 671 #define codeon() cgenflg++ 672 #define codeoff() --cgenflg 673 674 /* 675 * Codeline is the last lino output in the code generator. 676 * It used to be used to suppress LINO operators but no 677 * more since we now count statements. 678 * Lc is the intepreter code location counter. 679 * 680 short codeline; 681 */ 682 char *lc; 683 684 685 /* 686 * Routines which need types 687 * other than "integer" to be 688 * assumed by the compiler. 689 */ 690 double atof(); 691 long lwidth(); 692 long leven(); 693 long aryconst(); 694 long a8tol(); 695 long roundup(); 696 long tmpalloc(); 697 struct nl *lookup(); 698 double atof(); 699 int *tree(); 700 int *hash(); 701 char *alloc(); 702 int *calloc(); 703 char *savestr(); 704 char *parnam(); 705 bool fcompat(); 706 struct nl *lookup1(); 707 struct nl *hdefnl(); 708 struct nl *defnl(); 709 struct nl *enter(); 710 struct nl *nlcopy(); 711 struct nl *tyrecl(); 712 struct nl *tyary(); 713 struct nl *fields(); 714 struct nl *variants(); 715 struct nl *deffld(); 716 struct nl *defvnt(); 717 struct nl *tyrec1(); 718 struct nl *reclook(); 719 struct nl *asgnop1(); 720 struct nl *gtype(); 721 struct nl *call(); 722 struct nl *lvalue(); 723 struct nl *rvalue(); 724 struct nl *cset(); 725 726 /* 727 * type cast NIL to keep lint happy (which is not so bad) 728 */ 729 #define NLNIL ( (struct nl *) NIL ) 730 731 /* 732 * Funny structures to use 733 * pointers in wild and wooly ways 734 */ 735 struct { 736 char pchar; 737 }; 738 struct { 739 short pint; 740 short pint2; 741 }; 742 struct { 743 long plong; 744 }; 745 struct { 746 double pdouble; 747 }; 748 749 #define OCT 1 750 #define HEX 2 751 752 /* 753 * MAIN PROGRAM VARIABLES, MISCELLANY 754 */ 755 756 /* 757 * Variables forming a data base referencing 758 * the command line arguments with the "i" option, e.g. 759 * in "pi -i scanner.i compiler.p". 760 */ 761 char **pflist; 762 short pflstc; 763 short pfcnt; 764 765 char *filename; /* current source file name */ 766 long tvec; 767 extern char *snark; /* SNARK */ 768 extern char *classes[ ]; /* maps namelist classes to string names */ 769 770 #define derror error 771 772 #ifdef PC 773 774 /* 775 * the current function number, for [ lines 776 */ 777 int ftnno; 778 779 /* 780 * the pc output stream 781 */ 782 FILE *pcstream; 783 784 #endif PC 785