1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.12 04/17/81"; */ 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 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 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 *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 #endif 311 312 /* 313 * used to mark value[ NL_FORV ] for loop variables 314 */ 315 #define FORVAR 1 316 317 /* 318 * Definition of the commonly used "value" fields. 319 * The most important one is NL_OFFS which gives 320 * the offset of a variable in its stack mark. 321 */ 322 #define NL_OFFS 0 323 324 #define NL_CNTR 1 325 #define NL_NLSTRT 2 326 #define NL_LINENO 3 327 #define NL_FVAR 3 328 #define NL_FCHAIN 4 329 330 #define NL_GOLEV 2 331 #define NL_GOLINE 3 332 #define NL_FORV 1 333 334 #define NL_FLDSZ 1 335 #define NL_VARNT 2 336 #define NL_VTOREC 2 337 #define NL_TAG 3 338 339 #define NL_ELABEL 4 340 341 /* 342 * For BADUSE nl structures, NL_KINDS is a bit vector 343 * indicating the kinds of illegal usages complained about 344 * so far. For kind of bad use "kind", "1 << kind" is set. 345 * The low bit is reserved as ISUNDEF to indicate whether 346 * this identifier is totally undefined. 347 */ 348 #define NL_KINDS 0 349 350 #define ISUNDEF 1 351 352 /* 353 * Variables may reside on the stack as formals or as locals, 354 * or as register temporaries 355 */ 356 #define PARAMVAR 1 357 #define LOCALVAR 2 358 #define REGVAR 3 359 360 /* 361 * NAMELIST CLASSES 362 * 363 * The following are the namelist classes. 364 * Different classes make use of the value fields 365 * of the namelist in different ways. 366 * 367 * The namelist should be redesigned by providing 368 * a number of structure definitions with one corresponding 369 * to each namelist class, ala a variant record in Pascal. 370 */ 371 #define BADUSE 0 372 #define CONST 1 373 #define TYPE 2 374 #define VAR 3 375 #define ARRAY 4 376 #define PTRFILE 5 377 #define RECORD 6 378 #define FIELD 7 379 #define PROC 8 380 #define FUNC 9 381 #define FVAR 10 382 #define REF 11 383 #define PTR 12 384 #define FILET 13 385 #define SET 14 386 #define RANGE 15 387 #define LABEL 16 388 #define WITHPTR 17 389 #define SCAL 18 390 #define STR 19 391 #define PROG 20 392 #define IMPROPER 21 393 #define VARNT 22 394 #define FPROC 23 395 #define FFUNC 24 396 397 /* 398 * Clnames points to an array of names for the 399 * namelist classes. 400 */ 401 char **clnames; 402 403 /* 404 * PRE-DEFINED NAMELIST OFFSETS 405 * 406 * The following are the namelist offsets for the 407 * primitive types. The ones which are negative 408 * don't actually exist, but are generated and tested 409 * internally. These definitions are sensitive to the 410 * initializations in nl.c. 411 */ 412 #define TFIRST -7 413 #define TFILE -7 414 #define TREC -6 415 #define TARY -5 416 #define TSCAL -4 417 #define TPTR -3 418 #define TSET -2 419 #define TSTR -1 420 #define NIL 0 421 #define TBOOL 1 422 #define TCHAR 2 423 #define TINT 3 424 #define TDOUBLE 4 425 #define TNIL 5 426 #define T1INT 6 427 #define T2INT 7 428 #define T4INT 8 429 #define T1CHAR 9 430 #define T1BOOL 10 431 #define T8REAL 11 432 #define TLAST 11 433 434 /* 435 * SEMANTIC DEFINITIONS 436 */ 437 438 /* 439 * NOCON and SAWCON are flags in the tree telling whether 440 * a constant set is part of an expression. 441 * these are no longer used, 442 * since we now do constant sets at compile time. 443 */ 444 #define NOCON 0 445 #define SAWCON 1 446 447 /* 448 * The variable cbn gives the current block number, 449 * the variable bn is set as a side effect of a call to 450 * lookup, and is the block number of the variable which 451 * was found. 452 */ 453 short bn, cbn; 454 455 /* 456 * The variable line is the current semantic 457 * line and is set in stat.c from the numbers 458 * embedded in statement type tree nodes. 459 */ 460 short line; 461 462 /* 463 * The size of the display 464 * which defines the maximum nesting 465 * of procedures and functions allowed. 466 * Because of the flags in the current namelist 467 * this must be no greater than 32. 468 */ 469 #define DSPLYSZ 20 470 471 /* 472 * the display is made up of saved AP's and FP's. 473 * FP's are used to find locals, and AP's are used to find parameters. 474 * FP and AP are untyped pointers, but are used throughout as (char *). 475 * the display is used by adding AP_OFFSET or FP_OFFSET to the 476 * address of the approriate display entry. 477 */ 478 struct dispsave { 479 char *savedAP; 480 char *savedFP; 481 } display[ DSPLYSZ ]; 482 483 #define AP_OFFSET ( 0 ) 484 #define FP_OFFSET ( sizeof(char *) ) 485 486 /* 487 * formal routine structure: 488 */ 489 struct formalrtn { 490 long (*fentryaddr)(); /* formal entry point */ 491 long fbn; /* block number of function */ 492 struct dispsave fdisp[ DSPLYSZ ]; /* saved at first passing */ 493 } frtn; 494 495 #define FENTRYOFFSET 0 496 #define FBNOFFSET ( FENTRYOFFSET + sizeof frtn.fentryaddr ) 497 #define FDISPOFFSET ( FBNOFFSET + sizeof frtn.fbn ) 498 499 /* 500 * The following structure is used 501 * to keep track of the amount of variable 502 * storage required by each block. 503 * "Max" is the high water mark, "off" 504 * the current need. Temporaries for "for" 505 * loops and "with" statements are allocated 506 * in the local variable area and these 507 * numbers are thereby changed if necessary. 508 */ 509 struct om { 510 long om_max; 511 long reg_max; 512 struct tmps { 513 long om_off; 514 long reg_off; 515 } curtmps; 516 } sizes[DSPLYSZ]; 517 #define NOREG 0 518 #define REGOK 1 519 520 /* 521 * the following structure records whether a level declares 522 * any variables which are (or contain) files. 523 * this so that the runtime routines for file cleanup can be invoked. 524 */ 525 bool dfiles[ DSPLYSZ ]; 526 527 /* 528 * Structure recording information about a constant 529 * declaration. It is actually the return value from 530 * the routine "gconst", but since C doesn't support 531 * record valued functions, this is more convenient. 532 */ 533 struct { 534 struct nl *ctype; 535 short cival; 536 double crval; 537 int *cpval; 538 } con; 539 540 /* 541 * The set structure records the lower bound 542 * and upper bound with the lower bound normalized 543 * to zero when working with a set. It is set by 544 * the routine setran in var.c. 545 */ 546 struct { 547 short lwrb, uprbp; 548 } set; 549 550 /* 551 * structures of this kind are filled in by precset and used by postcset 552 * to indicate things about constant sets. 553 */ 554 struct csetstr { 555 struct nl *csettype; 556 long paircnt; 557 long singcnt; 558 bool comptime; 559 }; 560 /* 561 * The following flags are passed on calls to lvalue 562 * to indicate how the reference is to affect the usage 563 * information for the variable being referenced. 564 * MOD is used to set the NMOD flag in the namelist 565 * entry for the variable, ASGN permits diagnostics 566 * to be formed when a for variable is assigned to in 567 * the range of the loop. 568 */ 569 #define NOFLAGS 0 570 #define MOD 01 571 #define ASGN 02 572 #define NOUSE 04 573 574 /* 575 * the following flags are passed to lvalue and rvalue 576 * to tell them whether an lvalue or rvalue is required. 577 * the semantics checking is done according to the function called, 578 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 579 * and rvalue may stop short of putting out the indirection. 580 */ 581 #define LREQ 01 582 #define RREQ 02 583 584 double MAXINT; 585 double MININT; 586 587 /* 588 * Variables for generation of profile information. 589 * Monflg is set when we want to generate a profile. 590 * Gocnt record the total number of goto's and 591 * cnts records the current counter for generating 592 * COUNT operators. 593 */ 594 short gocnt; 595 short cnts; 596 597 /* 598 * Most routines call "incompat" rather than asking "!compat" 599 * for historical reasons. 600 */ 601 #define incompat !compat 602 603 /* 604 * Parts records which declaration parts have been seen. 605 * The grammar allows the "label" "const" "type" "var" and routine 606 * parts to be repeated and to be in any order, so that 607 * they can be detected semantically to give better 608 * error diagnostics. 609 */ 610 int parts[ DSPLYSZ ]; 611 612 #define LPRT 1 613 #define CPRT 2 614 #define TPRT 4 615 #define VPRT 8 616 #define RPRT 16 617 618 /* 619 * Flags for the "you used / instead of div" diagnostic 620 */ 621 bool divchk; 622 bool divflg; 623 624 bool errcnt[DSPLYSZ]; 625 626 /* 627 * Forechain links those types which are 628 * ^ sometype 629 * so that they can be evaluated later, permitting 630 * circular, recursive list structures to be defined. 631 */ 632 struct nl *forechain; 633 634 /* 635 * Withlist links all the records which are currently 636 * opened scopes because of with statements. 637 */ 638 struct nl *withlist; 639 640 struct nl *intset; 641 struct nl *input, *output; 642 struct nl *program; 643 644 /* progseen flag used by PC to determine if 645 * a routine segment is being compiled (and 646 * therefore no program statement seen) 647 */ 648 bool progseen; 649 650 651 /* 652 * STRUCTURED STATEMENT GOTO CHECKING 653 * 654 * The variable level keeps track of the current 655 * "structured statement level" when processing the statement 656 * body of blocks. This is used in the detection of goto's into 657 * structured statements in a block. 658 * 659 * Each label's namelist entry contains two pieces of information 660 * related to this check. The first `NL_GOLEV' either contains 661 * the level at which the label was declared, `NOTYET' if the label 662 * has not yet been declared, or `DEAD' if the label is dead, i.e. 663 * if we have exited the level in which the label was defined. 664 * 665 * When we discover a "goto" statement, if the label has not 666 * been defined yet, then we record the current level and the current line 667 * for a later error check. If the label has been already become "DEAD" 668 * then a reference to it is an error. Now the compiler maintains, 669 * for each block, a linked list of the labels headed by "gotos[bn]". 670 * When we exit a structured level, we perform the routine 671 * ungoto in stat.c. It notices labels whose definition levels have been 672 * exited and makes them be dead. For labels which have not yet been 673 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 674 * since the first usage of the label. It is not hard to see that the label 675 * must eventually be declared at this level or an outer level to this 676 * one or a goto into a structured statement will exist. 677 */ 678 short level; 679 struct nl *gotos[DSPLYSZ]; 680 681 #define NOTYET 10000 682 #define DEAD 10000 683 684 /* 685 * Noreach is true when the next statement will 686 * be unreachable unless something happens along 687 * (like exiting a looping construct) to save 688 * the day. 689 */ 690 bool noreach; 691 692 /* 693 * UNDEFINED VARIABLE REFERENCE STRUCTURES 694 */ 695 struct udinfo { 696 int ud_line; 697 struct udinfo *ud_next; 698 char nullch; 699 }; 700 701 /* 702 * CODE GENERATION DEFINITIONS 703 */ 704 705 /* 706 * NSTAND is or'ed onto the abstract machine opcode 707 * for non-standard built-in procedures and functions. 708 */ 709 #define NSTAND 0400 710 711 #define codeon() cgenflg++ 712 #define codeoff() --cgenflg 713 #define CGENNING ( cgenflg >= 0 ) 714 715 /* 716 * Codeline is the last lino output in the code generator. 717 * It used to be used to suppress LINO operators but no 718 * more since we now count statements. 719 * Lc is the intepreter code location counter. 720 * 721 short codeline; 722 */ 723 char *lc; 724 725 726 /* 727 * Routines which need types 728 * other than "integer" to be 729 * assumed by the compiler. 730 */ 731 double atof(); 732 long lwidth(); 733 long leven(); 734 long aryconst(); 735 long a8tol(); 736 long roundup(); 737 long tmpalloc(); 738 struct nl *lookup(); 739 double atof(); 740 int *tree(); 741 int *hash(); 742 char *alloc(); 743 int *calloc(); 744 char *savestr(); 745 char *parnam(); 746 bool fcompat(); 747 struct nl *lookup1(); 748 struct nl *hdefnl(); 749 struct nl *defnl(); 750 struct nl *enter(); 751 struct nl *nlcopy(); 752 struct nl *tyrecl(); 753 struct nl *tyary(); 754 struct nl *fields(); 755 struct nl *variants(); 756 struct nl *deffld(); 757 struct nl *defvnt(); 758 struct nl *tyrec1(); 759 struct nl *reclook(); 760 struct nl *asgnop1(); 761 struct nl *gtype(); 762 struct nl *call(); 763 struct nl *lvalue(); 764 struct nl *rvalue(); 765 struct nl *cset(); 766 767 /* 768 * type cast NIL to keep lint happy (which is not so bad) 769 */ 770 #define NLNIL ( (struct nl *) NIL ) 771 772 /* 773 * Funny structures to use 774 * pointers in wild and wooly ways 775 */ 776 struct { 777 char pchar; 778 }; 779 struct { 780 short pint; 781 short pint2; 782 }; 783 struct { 784 long plong; 785 }; 786 struct { 787 double pdouble; 788 }; 789 790 #define OCT 1 791 #define HEX 2 792 793 /* 794 * MAIN PROGRAM VARIABLES, MISCELLANY 795 */ 796 797 /* 798 * Variables forming a data base referencing 799 * the command line arguments with the "i" option, e.g. 800 * in "pi -i scanner.i compiler.p". 801 */ 802 char **pflist; 803 short pflstc; 804 short pfcnt; 805 806 char *filename; /* current source file name */ 807 long tvec; 808 extern char *snark; /* SNARK */ 809 extern char *classes[ ]; /* maps namelist classes to string names */ 810 811 #define derror error 812 813 #ifdef PC 814 815 /* 816 * the current function number, for [ lines 817 */ 818 int ftnno; 819 820 /* 821 * the pc output stream 822 */ 823 FILE *pcstream; 824 825 #endif PC 826