1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.14 03/29/82"; */ 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 #define continuation() setpfx(' ') 166 167 int cgenflg; 168 169 170 /* 171 * The flag syneflg is used to suppress the diagnostics of the form 172 * E 10 a, defined in someprocedure, is neither used nor set 173 * when there were syntax errors in "someprocedure". 174 * In this case, it is likely that these warinings would be spurious. 175 */ 176 bool syneflg; 177 178 /* 179 * The compiler keeps its error messages in a file. 180 * The variable efil is the unit number on which 181 * this file is open for reading of error message text. 182 * Similarly, the file ofil is the unit of the file 183 * "obj" where we write the interpreter code. 184 */ 185 short efil; 186 short ofil; 187 short obuf[518]; 188 189 bool Enoline; 190 #define elineoff() Enoline = TRUE 191 #define elineon() Enoline = FALSE 192 193 194 /* 195 * SYMBOL TABLE STRUCTURE DEFINITIONS 196 * 197 * The symbol table is henceforth referred to as the "namelist". 198 * It consists of a number of structures of the form "nl" below. 199 * These are contained in a number of segments of the symbol 200 * table which are dynamically allocated as needed. 201 * The major namelist manipulation routines are contained in the 202 * file "nl.c". 203 * 204 * The major components of a namelist entry are the "symbol", giving 205 * a pointer into the string table for the string associated with this 206 * entry and the "class" which tells which of the (currently 19) 207 * possible types of structure this is. 208 * 209 * Many of the classes use the "type" field for a pointer to the type 210 * which the entry has. 211 * 212 * Other pieces of information in more than one class include the block 213 * in which the symbol is defined, flags indicating whether the symbol 214 * has been used and whether it has been assigned to, etc. 215 * 216 * A more complete discussion of the features of the namelist is impossible 217 * here as it would be too voluminous. Refer to the "PI 1.0 Implementation 218 * Notes" for more details. 219 */ 220 221 /* 222 * The basic namelist structure. 223 * There are also two other variants, defining the real 224 * field as longs or integers given below. 225 * 226 * The array disptab defines the hash header for the symbol table. 227 * Symbols are hashed based on the low 6 bits of their pointer into 228 * the string table; see the routines in the file "lookup.c" and also "fdec.c" 229 * especially "funcend". 230 */ 231 extern struct nl *Fp; 232 extern int pnumcnt; 233 234 #ifdef PTREE 235 # include "pTree.h" 236 #endif PTREE 237 struct nl { 238 char *symbol; 239 char class, nl_flags; 240 #ifdef PC 241 char extra_flags; /* for where things are */ 242 #endif PC 243 struct nl *type; 244 struct nl *chain, *nl_next; 245 int value[5]; 246 # ifdef PTREE 247 pPointer inTree; 248 # endif PTREE 249 } *nlp, *disptab[077+1]; 250 251 extern struct nl nl[INL]; 252 253 struct { 254 char *symbol; 255 char class, nl_flags; 256 #ifdef PC 257 char extra_flags; 258 #endif 259 struct nl *type; 260 struct nl *chain, *nl_next; 261 double real; 262 }; 263 264 struct { 265 char *symbol; 266 char class, nl_block; 267 #ifdef PC 268 char extra_flags; 269 #endif 270 struct nl *type; 271 struct nl *chain, *nl_next; 272 long range[2]; 273 }; 274 275 struct { 276 char *symbol; 277 char class, nl_flags; 278 #ifdef PC 279 char extra_flags; 280 #endif 281 struct nl *type; 282 struct nl *chain, *nl_next; 283 int *ptr[4]; 284 #ifdef PI 285 int entloc; 286 #endif PI 287 }; 288 289 /* 290 * NL FLAGS BITS 291 * 292 * Definitions of the usage of the bits in 293 * the nl_flags byte. Note that the low 5 bits of the 294 * byte are the "nl_block" and that some classes make use 295 * of this byte as a "width". 296 * 297 * The only non-obvious bit definition here is "NFILES" 298 * which records whether a structure contains any files. 299 * Such structures are not allowed to be dynamically allocated. 300 */ 301 302 #define BLOCKNO( flag ) ( flag & 037 ) 303 #define NLFLAGS( flag ) ( flag &~ 037 ) 304 305 #define NUSED 0100 306 #define NMOD 0040 307 #define NFORWD 0200 308 #define NFILES 0200 309 #ifdef PC 310 #define NEXTERN 0001 /* flag used to mark external funcs and procs */ 311 #define NLOCAL 0002 /* variable is a local */ 312 #define NPARAM 0004 /* variable is a parameter */ 313 #define NGLOBAL 0010 /* variable is a global */ 314 #define NREGVAR 0020 /* or'ed in if variable is in a register */ 315 #endif PC 316 317 /* 318 * used to mark value[ NL_FORV ] for loop variables 319 */ 320 #define FORVAR 1 321 322 /* 323 * Definition of the commonly used "value" fields. 324 * The most important one is NL_OFFS which gives 325 * the offset of a variable in its stack mark. 326 */ 327 #define NL_OFFS 0 328 329 #define NL_CNTR 1 330 #define NL_NLSTRT 2 331 #define NL_LINENO 3 332 #define NL_FVAR 3 333 #define NL_FCHAIN 4 334 335 #define NL_GOLEV 2 336 #define NL_GOLINE 3 337 #define NL_FORV 1 338 339 #define NL_FLDSZ 1 340 #define NL_VARNT 2 341 #define NL_VTOREC 2 342 #define NL_TAG 3 343 344 #define NL_ELABEL 4 345 346 /* 347 * For BADUSE nl structures, NL_KINDS is a bit vector 348 * indicating the kinds of illegal usages complained about 349 * so far. For kind of bad use "kind", "1 << kind" is set. 350 * The low bit is reserved as ISUNDEF to indicate whether 351 * this identifier is totally undefined. 352 */ 353 #define NL_KINDS 0 354 355 #define ISUNDEF 1 356 357 /* 358 * variables come in three flavors: globals, parameters, locals; 359 * they can also hide in registers, but that's a different flag 360 */ 361 #define PARAMVAR 1 362 #define LOCALVAR 2 363 #define GLOBALVAR 3 364 365 /* 366 * NAMELIST CLASSES 367 * 368 * The following are the namelist classes. 369 * Different classes make use of the value fields 370 * of the namelist in different ways. 371 * 372 * The namelist should be redesigned by providing 373 * a number of structure definitions with one corresponding 374 * to each namelist class, ala a variant record in Pascal. 375 */ 376 #define BADUSE 0 377 #define CONST 1 378 #define TYPE 2 379 #define VAR 3 380 #define ARRAY 4 381 #define PTRFILE 5 382 #define RECORD 6 383 #define FIELD 7 384 #define PROC 8 385 #define FUNC 9 386 #define FVAR 10 387 #define REF 11 388 #define PTR 12 389 #define FILET 13 390 #define SET 14 391 #define RANGE 15 392 #define LABEL 16 393 #define WITHPTR 17 394 #define SCAL 18 395 #define STR 19 396 #define PROG 20 397 #define IMPROPER 21 398 #define VARNT 22 399 #define FPROC 23 400 #define FFUNC 24 401 402 /* 403 * Clnames points to an array of names for the 404 * namelist classes. 405 */ 406 char **clnames; 407 408 /* 409 * PRE-DEFINED NAMELIST OFFSETS 410 * 411 * The following are the namelist offsets for the 412 * primitive types. The ones which are negative 413 * don't actually exist, but are generated and tested 414 * internally. These definitions are sensitive to the 415 * initializations in nl.c. 416 */ 417 #define TFIRST -7 418 #define TFILE -7 419 #define TREC -6 420 #define TARY -5 421 #define TSCAL -4 422 #define TPTR -3 423 #define TSET -2 424 #define TSTR -1 425 #define NIL 0 426 #define TBOOL 1 427 #define TCHAR 2 428 #define TINT 3 429 #define TDOUBLE 4 430 #define TNIL 5 431 #define T1INT 6 432 #define T2INT 7 433 #define T4INT 8 434 #define T1CHAR 9 435 #define T1BOOL 10 436 #define T8REAL 11 437 #define TLAST 11 438 439 /* 440 * SEMANTIC DEFINITIONS 441 */ 442 443 /* 444 * NOCON and SAWCON are flags in the tree telling whether 445 * a constant set is part of an expression. 446 * these are no longer used, 447 * since we now do constant sets at compile time. 448 */ 449 #define NOCON 0 450 #define SAWCON 1 451 452 /* 453 * The variable cbn gives the current block number, 454 * the variable bn is set as a side effect of a call to 455 * lookup, and is the block number of the variable which 456 * was found. 457 */ 458 short bn, cbn; 459 460 /* 461 * The variable line is the current semantic 462 * line and is set in stat.c from the numbers 463 * embedded in statement type tree nodes. 464 */ 465 short line; 466 467 /* 468 * The size of the display 469 * which defines the maximum nesting 470 * of procedures and functions allowed. 471 * Because of the flags in the current namelist 472 * this must be no greater than 32. 473 */ 474 #define DSPLYSZ 20 475 476 /* 477 * the display is made up of saved AP's and FP's. 478 * FP's are used to find locals, and AP's are used to find parameters. 479 * FP and AP are untyped pointers, but are used throughout as (char *). 480 * the display is used by adding AP_OFFSET or FP_OFFSET to the 481 * address of the approriate display entry. 482 */ 483 struct dispsave { 484 char *savedAP; 485 char *savedFP; 486 } display[ DSPLYSZ ]; 487 488 #define AP_OFFSET ( 0 ) 489 #define FP_OFFSET ( sizeof(char *) ) 490 491 /* 492 * formal routine structure: 493 */ 494 struct formalrtn { 495 long (*fentryaddr)(); /* formal entry point */ 496 long fbn; /* block number of function */ 497 struct dispsave fdisp[ DSPLYSZ ]; /* saved at first passing */ 498 } frtn; 499 500 #define FENTRYOFFSET 0 501 #define FBNOFFSET ( FENTRYOFFSET + sizeof frtn.fentryaddr ) 502 #define FDISPOFFSET ( FBNOFFSET + sizeof frtn.fbn ) 503 504 /* 505 * The following structure is used 506 * to keep track of the amount of variable 507 * storage required by each block. 508 * "Max" is the high water mark, "off" 509 * the current need. Temporaries for "for" 510 * loops and "with" statements are allocated 511 * in the local variable area and these 512 * numbers are thereby changed if necessary. 513 */ 514 struct om { 515 long om_max; 516 long reg_max; 517 struct tmps { 518 long om_off; 519 long reg_off; 520 } curtmps; 521 } sizes[DSPLYSZ]; 522 #define NOREG 0 523 #define REGOK 1 524 525 /* 526 * the following structure records whether a level declares 527 * any variables which are (or contain) files. 528 * this so that the runtime routines for file cleanup can be invoked. 529 */ 530 bool dfiles[ DSPLYSZ ]; 531 532 /* 533 * Structure recording information about a constant 534 * declaration. It is actually the return value from 535 * the routine "gconst", but since C doesn't support 536 * record valued functions, this is more convenient. 537 */ 538 struct { 539 struct nl *ctype; 540 short cival; 541 double crval; 542 int *cpval; 543 } con; 544 545 /* 546 * The set structure records the lower bound 547 * and upper bound with the lower bound normalized 548 * to zero when working with a set. It is set by 549 * the routine setran in var.c. 550 */ 551 struct { 552 short lwrb, uprbp; 553 } set; 554 555 /* 556 * structures of this kind are filled in by precset and used by postcset 557 * to indicate things about constant sets. 558 */ 559 struct csetstr { 560 struct nl *csettype; 561 long paircnt; 562 long singcnt; 563 bool comptime; 564 }; 565 /* 566 * The following flags are passed on calls to lvalue 567 * to indicate how the reference is to affect the usage 568 * information for the variable being referenced. 569 * MOD is used to set the NMOD flag in the namelist 570 * entry for the variable, ASGN permits diagnostics 571 * to be formed when a for variable is assigned to in 572 * the range of the loop. 573 */ 574 #define NOFLAGS 0 575 #define MOD 01 576 #define ASGN 02 577 #define NOUSE 04 578 579 /* 580 * the following flags are passed to lvalue and rvalue 581 * to tell them whether an lvalue or rvalue is required. 582 * the semantics checking is done according to the function called, 583 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 584 * and rvalue may stop short of putting out the indirection. 585 */ 586 #define LREQ 01 587 #define RREQ 02 588 589 double MAXINT; 590 double MININT; 591 592 /* 593 * Variables for generation of profile information. 594 * Monflg is set when we want to generate a profile. 595 * Gocnt record the total number of goto's and 596 * cnts records the current counter for generating 597 * COUNT operators. 598 */ 599 short gocnt; 600 short cnts; 601 602 /* 603 * Most routines call "incompat" rather than asking "!compat" 604 * for historical reasons. 605 */ 606 #define incompat !compat 607 608 /* 609 * Parts records which declaration parts have been seen. 610 * The grammar allows the "label" "const" "type" "var" and routine 611 * parts to be repeated and to be in any order, so that 612 * they can be detected semantically to give better 613 * error diagnostics. 614 */ 615 int parts[ DSPLYSZ ]; 616 617 #define LPRT 1 618 #define CPRT 2 619 #define TPRT 4 620 #define VPRT 8 621 #define RPRT 16 622 623 /* 624 * Flags for the "you used / instead of div" diagnostic 625 */ 626 bool divchk; 627 bool divflg; 628 629 bool errcnt[DSPLYSZ]; 630 631 /* 632 * Forechain links those types which are 633 * ^ sometype 634 * so that they can be evaluated later, permitting 635 * circular, recursive list structures to be defined. 636 */ 637 struct nl *forechain; 638 639 /* 640 * Withlist links all the records which are currently 641 * opened scopes because of with statements. 642 */ 643 struct nl *withlist; 644 645 struct nl *intset; 646 struct nl *input, *output; 647 struct nl *program; 648 649 /* progseen flag used by PC to determine if 650 * a routine segment is being compiled (and 651 * therefore no program statement seen) 652 */ 653 bool progseen; 654 655 656 /* 657 * STRUCTURED STATEMENT GOTO CHECKING 658 * 659 * The variable level keeps track of the current 660 * "structured statement level" when processing the statement 661 * body of blocks. This is used in the detection of goto's into 662 * structured statements in a block. 663 * 664 * Each label's namelist entry contains two pieces of information 665 * related to this check. The first `NL_GOLEV' either contains 666 * the level at which the label was declared, `NOTYET' if the label 667 * has not yet been declared, or `DEAD' if the label is dead, i.e. 668 * if we have exited the level in which the label was defined. 669 * 670 * When we discover a "goto" statement, if the label has not 671 * been defined yet, then we record the current level and the current line 672 * for a later error check. If the label has been already become "DEAD" 673 * then a reference to it is an error. Now the compiler maintains, 674 * for each block, a linked list of the labels headed by "gotos[bn]". 675 * When we exit a structured level, we perform the routine 676 * ungoto in stat.c. It notices labels whose definition levels have been 677 * exited and makes them be dead. For labels which have not yet been 678 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 679 * since the first usage of the label. It is not hard to see that the label 680 * must eventually be declared at this level or an outer level to this 681 * one or a goto into a structured statement will exist. 682 */ 683 short level; 684 struct nl *gotos[DSPLYSZ]; 685 686 #define NOTYET 10000 687 #define DEAD 10000 688 689 /* 690 * Noreach is true when the next statement will 691 * be unreachable unless something happens along 692 * (like exiting a looping construct) to save 693 * the day. 694 */ 695 bool noreach; 696 697 /* 698 * UNDEFINED VARIABLE REFERENCE STRUCTURES 699 */ 700 struct udinfo { 701 int ud_line; 702 struct udinfo *ud_next; 703 char nullch; 704 }; 705 706 /* 707 * CODE GENERATION DEFINITIONS 708 */ 709 710 /* 711 * NSTAND is or'ed onto the abstract machine opcode 712 * for non-standard built-in procedures and functions. 713 */ 714 #define NSTAND 0400 715 716 #define codeon() cgenflg++ 717 #define codeoff() --cgenflg 718 #define CGENNING ( cgenflg >= 0 ) 719 720 /* 721 * Codeline is the last lino output in the code generator. 722 * It used to be used to suppress LINO operators but no 723 * more since we now count statements. 724 * Lc is the intepreter code location counter. 725 * 726 short codeline; 727 */ 728 char *lc; 729 730 731 /* 732 * Routines which need types 733 * other than "integer" to be 734 * assumed by the compiler. 735 */ 736 double atof(); 737 long lwidth(); 738 long leven(); 739 long aryconst(); 740 long a8tol(); 741 long roundup(); 742 struct nl *tmpalloc(); 743 struct nl *lookup(); 744 double atof(); 745 int *tree(); 746 int *hash(); 747 char *alloc(); 748 int *calloc(); 749 char *savestr(); 750 char *parnam(); 751 bool fcompat(); 752 struct nl *lookup1(); 753 struct nl *hdefnl(); 754 struct nl *defnl(); 755 struct nl *enter(); 756 struct nl *nlcopy(); 757 struct nl *tyrecl(); 758 struct nl *tyary(); 759 struct nl *fields(); 760 struct nl *variants(); 761 struct nl *deffld(); 762 struct nl *defvnt(); 763 struct nl *tyrec1(); 764 struct nl *reclook(); 765 struct nl *asgnop1(); 766 struct nl *gtype(); 767 struct nl *call(); 768 struct nl *lvalue(); 769 struct nl *rvalue(); 770 struct nl *cset(); 771 772 /* 773 * type cast NIL to keep lint happy (which is not so bad) 774 */ 775 #define NLNIL ( (struct nl *) NIL ) 776 777 /* 778 * Funny structures to use 779 * pointers in wild and wooly ways 780 */ 781 struct { 782 char pchar; 783 }; 784 struct { 785 short pint; 786 short pint2; 787 }; 788 struct { 789 long plong; 790 }; 791 struct { 792 double pdouble; 793 }; 794 795 #define OCT 1 796 #define HEX 2 797 798 /* 799 * MAIN PROGRAM VARIABLES, MISCELLANY 800 */ 801 802 /* 803 * Variables forming a data base referencing 804 * the command line arguments with the "i" option, e.g. 805 * in "pi -i scanner.i compiler.p". 806 */ 807 char **pflist; 808 short pflstc; 809 short pfcnt; 810 811 char *filename; /* current source file name */ 812 long tvec; 813 extern char *snark; /* SNARK */ 814 extern char *classes[ ]; /* maps namelist classes to string names */ 815 816 #define derror error 817 818 #ifdef PC 819 820 /* 821 * the current function number, for [ lines 822 */ 823 int ftnno; 824 825 /* 826 * the pc output stream 827 */ 828 FILE *pcstream; 829 830 #endif PC 831