1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.13 06/01/81"; */ 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 extra_flags; /* for where things are */ 241 #endif PC 242 struct nl *type; 243 struct nl *chain, *nl_next; 244 int value[5]; 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 extra_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 extra_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 extra_flags; 279 #endif 280 struct nl *type; 281 struct nl *chain, *nl_next; 282 int *ptr[4]; 283 #ifdef PI 284 int entloc; 285 #endif PI 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 */ 300 301 #define BLOCKNO( flag ) ( flag & 037 ) 302 #define NLFLAGS( flag ) ( flag &~ 037 ) 303 304 #define NUSED 0100 305 #define NMOD 0040 306 #define NFORWD 0200 307 #define NFILES 0200 308 #ifdef PC 309 #define NEXTERN 0001 /* flag used to mark external funcs and procs */ 310 #define NLOCAL 0002 /* variable is a local */ 311 #define NPARAM 0004 /* variable is a parameter */ 312 #define NGLOBAL 0010 /* variable is a global */ 313 #define NREGVAR 0020 /* or'ed in if variable is in a register */ 314 #endif PC 315 316 /* 317 * used to mark value[ NL_FORV ] for loop variables 318 */ 319 #define FORVAR 1 320 321 /* 322 * Definition of the commonly used "value" fields. 323 * The most important one is NL_OFFS which gives 324 * the offset of a variable in its stack mark. 325 */ 326 #define NL_OFFS 0 327 328 #define NL_CNTR 1 329 #define NL_NLSTRT 2 330 #define NL_LINENO 3 331 #define NL_FVAR 3 332 #define NL_FCHAIN 4 333 334 #define NL_GOLEV 2 335 #define NL_GOLINE 3 336 #define NL_FORV 1 337 338 #define NL_FLDSZ 1 339 #define NL_VARNT 2 340 #define NL_VTOREC 2 341 #define NL_TAG 3 342 343 #define NL_ELABEL 4 344 345 /* 346 * For BADUSE nl structures, NL_KINDS is a bit vector 347 * indicating the kinds of illegal usages complained about 348 * so far. For kind of bad use "kind", "1 << kind" is set. 349 * The low bit is reserved as ISUNDEF to indicate whether 350 * this identifier is totally undefined. 351 */ 352 #define NL_KINDS 0 353 354 #define ISUNDEF 1 355 356 /* 357 * variables come in three flavors: globals, parameters, locals; 358 * they can also hide in registers, but that's a different flag 359 */ 360 #define PARAMVAR 1 361 #define LOCALVAR 2 362 #define GLOBALVAR 3 363 364 /* 365 * NAMELIST CLASSES 366 * 367 * The following are the namelist classes. 368 * Different classes make use of the value fields 369 * of the namelist in different ways. 370 * 371 * The namelist should be redesigned by providing 372 * a number of structure definitions with one corresponding 373 * to each namelist class, ala a variant record in Pascal. 374 */ 375 #define BADUSE 0 376 #define CONST 1 377 #define TYPE 2 378 #define VAR 3 379 #define ARRAY 4 380 #define PTRFILE 5 381 #define RECORD 6 382 #define FIELD 7 383 #define PROC 8 384 #define FUNC 9 385 #define FVAR 10 386 #define REF 11 387 #define PTR 12 388 #define FILET 13 389 #define SET 14 390 #define RANGE 15 391 #define LABEL 16 392 #define WITHPTR 17 393 #define SCAL 18 394 #define STR 19 395 #define PROG 20 396 #define IMPROPER 21 397 #define VARNT 22 398 #define FPROC 23 399 #define FFUNC 24 400 401 /* 402 * Clnames points to an array of names for the 403 * namelist classes. 404 */ 405 char **clnames; 406 407 /* 408 * PRE-DEFINED NAMELIST OFFSETS 409 * 410 * The following are the namelist offsets for the 411 * primitive types. The ones which are negative 412 * don't actually exist, but are generated and tested 413 * internally. These definitions are sensitive to the 414 * initializations in nl.c. 415 */ 416 #define TFIRST -7 417 #define TFILE -7 418 #define TREC -6 419 #define TARY -5 420 #define TSCAL -4 421 #define TPTR -3 422 #define TSET -2 423 #define TSTR -1 424 #define NIL 0 425 #define TBOOL 1 426 #define TCHAR 2 427 #define TINT 3 428 #define TDOUBLE 4 429 #define TNIL 5 430 #define T1INT 6 431 #define T2INT 7 432 #define T4INT 8 433 #define T1CHAR 9 434 #define T1BOOL 10 435 #define T8REAL 11 436 #define TLAST 11 437 438 /* 439 * SEMANTIC DEFINITIONS 440 */ 441 442 /* 443 * NOCON and SAWCON are flags in the tree telling whether 444 * a constant set is part of an expression. 445 * these are no longer used, 446 * since we now do constant sets at compile time. 447 */ 448 #define NOCON 0 449 #define SAWCON 1 450 451 /* 452 * The variable cbn gives the current block number, 453 * the variable bn is set as a side effect of a call to 454 * lookup, and is the block number of the variable which 455 * was found. 456 */ 457 short bn, cbn; 458 459 /* 460 * The variable line is the current semantic 461 * line and is set in stat.c from the numbers 462 * embedded in statement type tree nodes. 463 */ 464 short line; 465 466 /* 467 * The size of the display 468 * which defines the maximum nesting 469 * of procedures and functions allowed. 470 * Because of the flags in the current namelist 471 * this must be no greater than 32. 472 */ 473 #define DSPLYSZ 20 474 475 /* 476 * the display is made up of saved AP's and FP's. 477 * FP's are used to find locals, and AP's are used to find parameters. 478 * FP and AP are untyped pointers, but are used throughout as (char *). 479 * the display is used by adding AP_OFFSET or FP_OFFSET to the 480 * address of the approriate display entry. 481 */ 482 struct dispsave { 483 char *savedAP; 484 char *savedFP; 485 } display[ DSPLYSZ ]; 486 487 #define AP_OFFSET ( 0 ) 488 #define FP_OFFSET ( sizeof(char *) ) 489 490 /* 491 * formal routine structure: 492 */ 493 struct formalrtn { 494 long (*fentryaddr)(); /* formal entry point */ 495 long fbn; /* block number of function */ 496 struct dispsave fdisp[ DSPLYSZ ]; /* saved at first passing */ 497 } frtn; 498 499 #define FENTRYOFFSET 0 500 #define FBNOFFSET ( FENTRYOFFSET + sizeof frtn.fentryaddr ) 501 #define FDISPOFFSET ( FBNOFFSET + sizeof frtn.fbn ) 502 503 /* 504 * The following structure is used 505 * to keep track of the amount of variable 506 * storage required by each block. 507 * "Max" is the high water mark, "off" 508 * the current need. Temporaries for "for" 509 * loops and "with" statements are allocated 510 * in the local variable area and these 511 * numbers are thereby changed if necessary. 512 */ 513 struct om { 514 long om_max; 515 long reg_max; 516 struct tmps { 517 long om_off; 518 long reg_off; 519 } curtmps; 520 } sizes[DSPLYSZ]; 521 #define NOREG 0 522 #define REGOK 1 523 524 /* 525 * the following structure records whether a level declares 526 * any variables which are (or contain) files. 527 * this so that the runtime routines for file cleanup can be invoked. 528 */ 529 bool dfiles[ DSPLYSZ ]; 530 531 /* 532 * Structure recording information about a constant 533 * declaration. It is actually the return value from 534 * the routine "gconst", but since C doesn't support 535 * record valued functions, this is more convenient. 536 */ 537 struct { 538 struct nl *ctype; 539 short cival; 540 double crval; 541 int *cpval; 542 } con; 543 544 /* 545 * The set structure records the lower bound 546 * and upper bound with the lower bound normalized 547 * to zero when working with a set. It is set by 548 * the routine setran in var.c. 549 */ 550 struct { 551 short lwrb, uprbp; 552 } set; 553 554 /* 555 * structures of this kind are filled in by precset and used by postcset 556 * to indicate things about constant sets. 557 */ 558 struct csetstr { 559 struct nl *csettype; 560 long paircnt; 561 long singcnt; 562 bool comptime; 563 }; 564 /* 565 * The following flags are passed on calls to lvalue 566 * to indicate how the reference is to affect the usage 567 * information for the variable being referenced. 568 * MOD is used to set the NMOD flag in the namelist 569 * entry for the variable, ASGN permits diagnostics 570 * to be formed when a for variable is assigned to in 571 * the range of the loop. 572 */ 573 #define NOFLAGS 0 574 #define MOD 01 575 #define ASGN 02 576 #define NOUSE 04 577 578 /* 579 * the following flags are passed to lvalue and rvalue 580 * to tell them whether an lvalue or rvalue is required. 581 * the semantics checking is done according to the function called, 582 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 583 * and rvalue may stop short of putting out the indirection. 584 */ 585 #define LREQ 01 586 #define RREQ 02 587 588 double MAXINT; 589 double MININT; 590 591 /* 592 * Variables for generation of profile information. 593 * Monflg is set when we want to generate a profile. 594 * Gocnt record the total number of goto's and 595 * cnts records the current counter for generating 596 * COUNT operators. 597 */ 598 short gocnt; 599 short cnts; 600 601 /* 602 * Most routines call "incompat" rather than asking "!compat" 603 * for historical reasons. 604 */ 605 #define incompat !compat 606 607 /* 608 * Parts records which declaration parts have been seen. 609 * The grammar allows the "label" "const" "type" "var" and routine 610 * parts to be repeated and to be in any order, so that 611 * they can be detected semantically to give better 612 * error diagnostics. 613 */ 614 int parts[ DSPLYSZ ]; 615 616 #define LPRT 1 617 #define CPRT 2 618 #define TPRT 4 619 #define VPRT 8 620 #define RPRT 16 621 622 /* 623 * Flags for the "you used / instead of div" diagnostic 624 */ 625 bool divchk; 626 bool divflg; 627 628 bool errcnt[DSPLYSZ]; 629 630 /* 631 * Forechain links those types which are 632 * ^ sometype 633 * so that they can be evaluated later, permitting 634 * circular, recursive list structures to be defined. 635 */ 636 struct nl *forechain; 637 638 /* 639 * Withlist links all the records which are currently 640 * opened scopes because of with statements. 641 */ 642 struct nl *withlist; 643 644 struct nl *intset; 645 struct nl *input, *output; 646 struct nl *program; 647 648 /* progseen flag used by PC to determine if 649 * a routine segment is being compiled (and 650 * therefore no program statement seen) 651 */ 652 bool progseen; 653 654 655 /* 656 * STRUCTURED STATEMENT GOTO CHECKING 657 * 658 * The variable level keeps track of the current 659 * "structured statement level" when processing the statement 660 * body of blocks. This is used in the detection of goto's into 661 * structured statements in a block. 662 * 663 * Each label's namelist entry contains two pieces of information 664 * related to this check. The first `NL_GOLEV' either contains 665 * the level at which the label was declared, `NOTYET' if the label 666 * has not yet been declared, or `DEAD' if the label is dead, i.e. 667 * if we have exited the level in which the label was defined. 668 * 669 * When we discover a "goto" statement, if the label has not 670 * been defined yet, then we record the current level and the current line 671 * for a later error check. If the label has been already become "DEAD" 672 * then a reference to it is an error. Now the compiler maintains, 673 * for each block, a linked list of the labels headed by "gotos[bn]". 674 * When we exit a structured level, we perform the routine 675 * ungoto in stat.c. It notices labels whose definition levels have been 676 * exited and makes them be dead. For labels which have not yet been 677 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 678 * since the first usage of the label. It is not hard to see that the label 679 * must eventually be declared at this level or an outer level to this 680 * one or a goto into a structured statement will exist. 681 */ 682 short level; 683 struct nl *gotos[DSPLYSZ]; 684 685 #define NOTYET 10000 686 #define DEAD 10000 687 688 /* 689 * Noreach is true when the next statement will 690 * be unreachable unless something happens along 691 * (like exiting a looping construct) to save 692 * the day. 693 */ 694 bool noreach; 695 696 /* 697 * UNDEFINED VARIABLE REFERENCE STRUCTURES 698 */ 699 struct udinfo { 700 int ud_line; 701 struct udinfo *ud_next; 702 char nullch; 703 }; 704 705 /* 706 * CODE GENERATION DEFINITIONS 707 */ 708 709 /* 710 * NSTAND is or'ed onto the abstract machine opcode 711 * for non-standard built-in procedures and functions. 712 */ 713 #define NSTAND 0400 714 715 #define codeon() cgenflg++ 716 #define codeoff() --cgenflg 717 #define CGENNING ( cgenflg >= 0 ) 718 719 /* 720 * Codeline is the last lino output in the code generator. 721 * It used to be used to suppress LINO operators but no 722 * more since we now count statements. 723 * Lc is the intepreter code location counter. 724 * 725 short codeline; 726 */ 727 char *lc; 728 729 730 /* 731 * Routines which need types 732 * other than "integer" to be 733 * assumed by the compiler. 734 */ 735 double atof(); 736 long lwidth(); 737 long leven(); 738 long aryconst(); 739 long a8tol(); 740 long roundup(); 741 struct nl *tmpalloc(); 742 struct nl *lookup(); 743 double atof(); 744 int *tree(); 745 int *hash(); 746 char *alloc(); 747 int *calloc(); 748 char *savestr(); 749 char *parnam(); 750 bool fcompat(); 751 struct nl *lookup1(); 752 struct nl *hdefnl(); 753 struct nl *defnl(); 754 struct nl *enter(); 755 struct nl *nlcopy(); 756 struct nl *tyrecl(); 757 struct nl *tyary(); 758 struct nl *fields(); 759 struct nl *variants(); 760 struct nl *deffld(); 761 struct nl *defvnt(); 762 struct nl *tyrec1(); 763 struct nl *reclook(); 764 struct nl *asgnop1(); 765 struct nl *gtype(); 766 struct nl *call(); 767 struct nl *lvalue(); 768 struct nl *rvalue(); 769 struct nl *cset(); 770 771 /* 772 * type cast NIL to keep lint happy (which is not so bad) 773 */ 774 #define NLNIL ( (struct nl *) NIL ) 775 776 /* 777 * Funny structures to use 778 * pointers in wild and wooly ways 779 */ 780 struct { 781 char pchar; 782 }; 783 struct { 784 short pint; 785 short pint2; 786 }; 787 struct { 788 long plong; 789 }; 790 struct { 791 double pdouble; 792 }; 793 794 #define OCT 1 795 #define HEX 2 796 797 /* 798 * MAIN PROGRAM VARIABLES, MISCELLANY 799 */ 800 801 /* 802 * Variables forming a data base referencing 803 * the command line arguments with the "i" option, e.g. 804 * in "pi -i scanner.i compiler.p". 805 */ 806 char **pflist; 807 short pflstc; 808 short pfcnt; 809 810 char *filename; /* current source file name */ 811 long tvec; 812 extern char *snark; /* SNARK */ 813 extern char *classes[ ]; /* maps namelist classes to string names */ 814 815 #define derror error 816 817 #ifdef PC 818 819 /* 820 * the current function number, for [ lines 821 */ 822 int ftnno; 823 824 /* 825 * the pc output stream 826 */ 827 FILE *pcstream; 828 829 #endif PC 830