1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.2 09/02/80"; */ 4 5 #define DEBUG 6 #define CHAR 7 #define STATIC 8 #define hp21mx 0 9 10 #include <stdio.h> 11 #include <sys/types.h> 12 13 #define bool short 14 #define TRUE 1 15 #define FALSE 0 16 17 /* 18 * Option flags 19 * 20 * The following options are recognized in the text of the program 21 * and also on the command line: 22 * 23 * b block buffer the file output 24 * 25 * i make a listing of the procedures and functions in 26 * the following include files 27 * 28 * l make a listing of the program 29 * 30 * n place each include file on a new page with a header 31 * 32 * p disable post mortem and statement limit counting 33 * 34 * t disable run-time tests 35 * 36 * u card image mode; only first 72 chars of input count 37 * 38 * w suppress special diagnostic warnings 39 * 40 * z generate counters for an execution profile 41 */ 42 #ifdef DEBUG 43 bool fulltrace, errtrace, testtrace, yyunique; 44 #endif DEBUG 45 46 /* 47 * Each option has a stack of 17 option values, with opts giving 48 * the current, top value, and optstk the value beneath it. 49 * One refers to option `l' as, e.g., opt('l') in the text for clarity. 50 */ 51 char opts[ 'z' - 'A' + 1]; 52 short optstk[ 'z' - 'A' + 1]; 53 54 #define opt(c) opts[c-'A'] 55 56 /* 57 * Monflg is set when we are generating 58 * a pxp profile. this is set by the -z command line option. 59 */ 60 bool monflg; 61 62 /* 63 * profflag is set when we are generating a prof profile. 64 * this is set by the -p command line option. 65 */ 66 bool profflag; 67 68 69 /* 70 * NOTES ON THE DYNAMIC NATURE OF THE DATA STRUCTURES 71 * 72 * Pi uses expandable tables for 73 * its namelist (symbol table), string table 74 * hash table, and parse tree space. The following 75 * definitions specify the size of the increments 76 * for these items in fundamental units so that 77 * each uses approximately 1024 bytes. 78 */ 79 80 #define STRINC 1024 /* string space increment */ 81 #define TRINC 512 /* tree space increment */ 82 #define HASHINC 509 /* hash table size in words, each increment */ 83 #define NLINC 56 /* namelist increment size in nl structs */ 84 85 /* 86 * The initial sizes of the structures. 87 * These should be large enough to compile 88 * an "average" sized program so as to minimize 89 * storage requests. 90 * On a small system or and 11/34 or 11/40 91 * these numbers can be trimmed to make the 92 * compiler smaller. 93 */ 94 #define ITREE 2000 95 #define INL 200 96 #define IHASH 509 97 98 /* 99 * The following limits on hash and tree tables currently 100 * allow approximately 1200 symbols and 20k words of tree 101 * space. The fundamental limit of 64k total data space 102 * should be exceeded well before these are full. 103 */ 104 /* 105 * TABLE_MULTIPLIER is for uniformly increasing the sizes of the tables 106 */ 107 #define TABLE_MULTIPLIER 8 108 #define MAXHASH (4 * TABLE_MULTIPLIER) 109 #define MAXNL (12 * TABLE_MULTIPLIER) 110 #define MAXTREE (30 * TABLE_MULTIPLIER) 111 /* 112 * MAXDEPTH is the depth of the parse stack. 113 * STACK_MULTIPLIER is for increasing its size. 114 */ 115 #define STACK_MULTIPLIER 8 116 #define MAXDEPTH ( 150 * STACK_MULTIPLIER ) 117 118 /* 119 * ERROR RELATED DEFINITIONS 120 */ 121 122 /* 123 * Exit statuses to pexit 124 * 125 * AOK 126 * ERRS Compilation errors inhibit obj productin 127 * NOSTART Errors before we ever got started 128 * DIED We ran out of memory or some such 129 */ 130 #define AOK 0 131 #define ERRS 1 132 #define NOSTART 2 133 #define DIED 3 134 135 bool Recovery; 136 137 #define eholdnl() Eholdnl = 1 138 #define nocascade() Enocascade = 1 139 140 bool Eholdnl, Enocascade; 141 142 143 /* 144 * The flag eflg is set whenever we have a hard error. 145 * The character in errpfx will precede the next error message. 146 * When cgenflg is set code generation is suppressed. 147 * This happens whenver we have an error (i.e. if eflg is set) 148 * and when we are walking the tree to determine types only. 149 */ 150 bool eflg; 151 char errpfx; 152 153 #define setpfx(x) errpfx = x 154 155 #define standard() setpfx('s') 156 #define warning() setpfx('w') 157 #define recovered() setpfx('e') 158 159 bool cgenflg; 160 161 162 /* 163 * The flag syneflg is used to suppress the diagnostics of the form 164 * E 10 a, defined in someprocedure, is neither used nor set 165 * when there were syntax errors in "someprocedure". 166 * In this case, it is likely that these warinings would be spurious. 167 */ 168 bool syneflg; 169 170 /* 171 * The compiler keeps its error messages in a file. 172 * The variable efil is the unit number on which 173 * this file is open for reading of error message text. 174 * Similarly, the file ofil is the unit of the file 175 * "obj" where we write the interpreter code. 176 */ 177 short efil; 178 short ofil; 179 short obuf[518]; 180 181 #define elineoff() Enoline++ 182 #define elineon() Enoline = 0 183 184 bool Enoline; 185 186 /* 187 * SYMBOL TABLE STRUCTURE DEFINITIONS 188 * 189 * The symbol table is henceforth referred to as the "namelist". 190 * It consists of a number of structures of the form "nl" below. 191 * These are contained in a number of segments of the symbol 192 * table which are dynamically allocated as needed. 193 * The major namelist manipulation routines are contained in the 194 * file "nl.c". 195 * 196 * The major components of a namelist entry are the "symbol", giving 197 * a pointer into the string table for the string associated with this 198 * entry and the "class" which tells which of the (currently 19) 199 * possible types of structure this is. 200 * 201 * Many of the classes use the "type" field for a pointer to the type 202 * which the entry has. 203 * 204 * Other pieces of information in more than one class include the block 205 * in which the symbol is defined, flags indicating whether the symbol 206 * has been used and whether it has been assigned to, etc. 207 * 208 * A more complete discussion of the features of the namelist is impossible 209 * here as it would be too voluminous. Refer to the "PI 1.0 Implementation 210 * Notes" for more details. 211 */ 212 213 /* 214 * The basic namelist structure. 215 * There are also two other variants, defining the real 216 * field as longs or integers given below. 217 * 218 * The array disptab defines the hash header for the symbol table. 219 * Symbols are hashed based on the low 6 bits of their pointer into 220 * the string table; see the routines in the file "lookup.c" and also "fdec.c" 221 * especially "funcend". 222 */ 223 #ifdef PTREE 224 # include "pTree.h" 225 #endif PTREE 226 struct nl { 227 char *symbol; 228 char class, nl_flags; 229 #ifdef PC 230 char ext_flags; /* an extra flag is used for externals */ 231 #endif PC 232 struct nl *type; 233 struct nl *chain, *nl_next; 234 int *ptr[4]; 235 #ifdef PI 236 int entloc; 237 #endif PI 238 # ifdef PTREE 239 pPointer inTree; 240 # endif PTREE 241 } *nlp, *disptab[077+1]; 242 243 extern struct nl nl[INL]; 244 245 struct { 246 char *symbol; 247 char class, nl_flags; 248 #ifdef PC 249 char ext_flags; 250 #endif 251 struct nl *type; 252 struct nl *chain, *nl_next; 253 double real; 254 }; 255 256 struct { 257 char *symbol; 258 char class, nl_block; 259 #ifdef PC 260 char ext_flags; 261 #endif 262 struct nl *type; 263 struct nl *chain, *nl_next; 264 long range[2]; 265 }; 266 267 struct { 268 char *symbol; 269 char class, nl_flags; 270 #ifdef PC 271 char ext_flags; 272 #endif 273 struct nl *type; 274 struct nl *chain, *nl_next; 275 long value[4]; 276 }; 277 278 /* 279 * NL FLAGS BITS 280 * 281 * Definitions of the usage of the bits in 282 * the nl_flags byte. Note that the low 5 bits of the 283 * byte are the "nl_block" and that some classes make use 284 * of this byte as a "width". 285 * 286 * The only non-obvious bit definition here is "NFILES" 287 * which records whether a structure contains any files. 288 * Such structures are not allowed to be dynamically allocated. 289 */ 290 #define NUSED 0100 291 #define NMOD 0040 292 #define NFORWD 0200 293 #define NFILES 0200 294 295 #ifdef PC 296 #define NEXTERN 0001 /* flag used to mark external funcs and procs */ 297 #endif 298 299 /* 300 * Definition of the commonly used "value" fields. 301 * The most important one is NL_OFFS which gives 302 * the offset of a variable in its stack mark. 303 */ 304 #define NL_OFFS 0 305 306 #define NL_CNTR 1 307 #define NL_FVAR 3 308 309 #define NL_GOLEV 2 310 #define NL_GOLINE 3 311 #define NL_FORV 1 312 313 #define NL_FLDSZ 1 314 #define NL_VARNT 2 315 #define NL_VTOREC 2 316 #define NL_TAG 3 317 318 #define NL_ELABEL 3 319 320 /* 321 * For BADUSE nl structures, NL_KINDS is a bit vector 322 * indicating the kinds of illegal usages complained about 323 * so far. For kind of bad use "kind", "1 << kind" is set. 324 * The low bit is reserved as ISUNDEF to indicate whether 325 * this identifier is totally undefined. 326 */ 327 #define NL_KINDS 0 328 329 #define ISUNDEF 1 330 331 /* 332 * NAMELIST CLASSES 333 * 334 * The following are the namelist classes. 335 * Different classes make use of the value fields 336 * of the namelist in different ways. 337 * 338 * The namelist should be redesigned by providing 339 * a number of structure definitions with one corresponding 340 * to each namelist class, ala a variant record in Pascal. 341 */ 342 #define BADUSE 0 343 #define CONST 1 344 #define TYPE 2 345 #define VAR 3 346 #define ARRAY 4 347 #define PTRFILE 5 348 #define RECORD 6 349 #define FIELD 7 350 #define PROC 8 351 #define FUNC 9 352 #define FVAR 10 353 #define REF 11 354 #define PTR 12 355 #define FILET 13 356 #define SET 14 357 #define RANGE 15 358 #define LABEL 16 359 #define WITHPTR 17 360 #define SCAL 18 361 #define STR 19 362 #define PROG 20 363 #define IMPROPER 21 364 #define VARNT 22 365 366 /* 367 * Clnames points to an array of names for the 368 * namelist classes. 369 */ 370 char **clnames; 371 372 /* 373 * PRE-DEFINED NAMELIST OFFSETS 374 * 375 * The following are the namelist offsets for the 376 * primitive types. The ones which are negative 377 * don't actually exist, but are generated and tested 378 * internally. These definitions are sensitive to the 379 * initializations in nl.c. 380 */ 381 #define TFIRST -7 382 #define TFILE -7 383 #define TREC -6 384 #define TARY -5 385 #define TSCAL -4 386 #define TPTR -3 387 #define TSET -2 388 #define TSTR -1 389 #define NIL 0 390 #define TBOOL 1 391 #define TCHAR 2 392 #define TINT 3 393 #define TDOUBLE 4 394 #define TNIL 5 395 #define T1INT 6 396 #define T2INT 7 397 #define T4INT 8 398 #define T1CHAR 9 399 #define T1BOOL 10 400 #define T8REAL 11 401 #define TLAST 11 402 403 /* 404 * SEMANTIC DEFINITIONS 405 */ 406 407 /* 408 * NOCON and SAWCON are flags in the tree telling whether 409 * a constant set is part of an expression. 410 */ 411 #define NOCON 0 412 #define SAWCON 1 413 414 /* 415 * The variable cbn gives the current block number, 416 * the variable bn is set as a side effect of a call to 417 * lookup, and is the block number of the variable which 418 * was found. 419 */ 420 short bn, cbn; 421 422 /* 423 * The variable line is the current semantic 424 * line and is set in stat.c from the numbers 425 * embedded in statement type tree nodes. 426 */ 427 short line; 428 429 /* 430 * The size of the display 431 * which defines the maximum nesting 432 * of procedures and functions allowed. 433 * Because of the flags in the current namelist 434 * this must be no greater than 32. 435 */ 436 #define DSPLYSZ 20 437 438 /* 439 * The following structure is used 440 * to keep track of the amount of variable 441 * storage required by each block. 442 * "Max" is the high water mark, "off" 443 * the current need. Temporaries for "for" 444 * loops and "with" statements are allocated 445 * in the local variable area and these 446 * numbers are thereby changed if necessary. 447 */ 448 struct om { 449 long om_off; 450 long om_max; 451 } sizes[DSPLYSZ]; 452 453 /* 454 * the following structure records whether a level declares 455 * any variables which are (or contain) files. 456 * this so that the runtime routines for file cleanup can be invoked. 457 */ 458 bool dfiles[ DSPLYSZ ]; 459 460 /* 461 * Structure recording information about a constant 462 * declaration. It is actually the return value from 463 * the routine "gconst", but since C doesn't support 464 * record valued functions, this is more convenient. 465 */ 466 struct { 467 struct nl *ctype; 468 short cival; 469 double crval; 470 int *cpval; 471 } con; 472 473 /* 474 * The set structure records the lower bound 475 * and upper bound with the lower bound normalized 476 * to zero when working with a set. It is set by 477 * the routine setran in var.c. 478 */ 479 struct { 480 short lwrb, uprbp; 481 } set; 482 483 /* 484 * structures of this kind are filled in by precset and used by postcset 485 * to indicate things about constant sets. 486 */ 487 struct csetstr { 488 struct nl *csettype; 489 long paircnt; 490 long singcnt; 491 bool comptime; 492 }; 493 /* 494 * The following flags are passed on calls to lvalue 495 * to indicate how the reference is to affect the usage 496 * information for the variable being referenced. 497 * MOD is used to set the NMOD flag in the namelist 498 * entry for the variable, ASGN permits diagnostics 499 * to be formed when a for variable is assigned to in 500 * the range of the loop. 501 */ 502 #define NOFLAGS 0 503 #define MOD 01 504 #define ASGN 02 505 #define NOUSE 04 506 507 /* 508 * the following flags are passed to lvalue and rvalue 509 * to tell them whether an lvalue or rvalue is required. 510 * the semantics checking is done according to the function called, 511 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 512 * and rvalue may stop short of putting out the indirection. 513 */ 514 #define LREQ 01 515 #define RREQ 02 516 517 double MAXINT; 518 double MININT; 519 520 /* 521 * Variables for generation of profile information. 522 * Monflg is set when we want to generate a profile. 523 * Gocnt record the total number of goto's and 524 * cnts records the current counter for generating 525 * COUNT operators. 526 */ 527 short gocnt; 528 short cnts; 529 530 /* 531 * Most routines call "incompat" rather than asking "!compat" 532 * for historical reasons. 533 */ 534 #define incompat !compat 535 536 /* 537 * Parts records which declaration parts have been seen. 538 * The grammar allows the "label" "const" "type" "var" and routine 539 * parts to be repeated and to be in any order, so that 540 * they can be detected semantically to give better 541 * error diagnostics. 542 */ 543 int parts[ DSPLYSZ ]; 544 545 #define LPRT 1 546 #define CPRT 2 547 #define TPRT 4 548 #define VPRT 8 549 #define RPRT 16 550 551 /* 552 * Flags for the "you used / instead of div" diagnostic 553 */ 554 bool divchk; 555 bool divflg; 556 557 short errcnt[DSPLYSZ]; 558 559 /* 560 * Forechain links those types which are 561 * ^ sometype 562 * so that they can be evaluated later, permitting 563 * circular, recursive list structures to be defined. 564 */ 565 struct nl *forechain; 566 567 /* 568 * Withlist links all the records which are currently 569 * opened scopes because of with statements. 570 */ 571 struct nl *withlist; 572 573 struct nl *intset; 574 struct nl *input, *output; 575 struct nl *program; 576 577 /* progseen flag used by PC to determine if 578 * a routine segment is being compiled (and 579 * therefore no program statement seen) 580 */ 581 bool progseen; 582 583 584 /* 585 * STRUCTURED STATEMENT GOTO CHECKING 586 * 587 * The variable level keeps track of the current 588 * "structured statement level" when processing the statement 589 * body of blocks. This is used in the detection of goto's into 590 * structured statements in a block. 591 * 592 * Each label's namelist entry contains two pieces of information 593 * related to this check. The first `NL_GOLEV' either contains 594 * the level at which the label was declared, `NOTYET' if the label 595 * has not yet been declared, or `DEAD' if the label is dead, i.e. 596 * if we have exited the level in which the label was defined. 597 * 598 * When we discover a "goto" statement, if the label has not 599 * been defined yet, then we record the current level and the current line 600 * for a later error check. If the label has been already become "DEAD" 601 * then a reference to it is an error. Now the compiler maintains, 602 * for each block, a linked list of the labels headed by "gotos[bn]". 603 * When we exit a structured level, we perform the routine 604 * ungoto in stat.c. It notices labels whose definition levels have been 605 * exited and makes them be dead. For labels which have not yet been 606 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 607 * since the first usage of the label. It is not hard to see that the label 608 * must eventually be declared at this level or an outer level to this 609 * one or a goto into a structured statement will exist. 610 */ 611 short level; 612 struct nl *gotos[DSPLYSZ]; 613 614 #define NOTYET 10000 615 #define DEAD 10000 616 617 /* 618 * Noreach is true when the next statement will 619 * be unreachable unless something happens along 620 * (like exiting a looping construct) to save 621 * the day. 622 */ 623 bool noreach; 624 625 /* 626 * UNDEFINED VARIABLE REFERENCE STRUCTURES 627 */ 628 struct udinfo { 629 int ud_line; 630 struct udinfo *ud_next; 631 char nullch; 632 }; 633 634 /* 635 * CODE GENERATION DEFINITIONS 636 */ 637 638 /* 639 * NSTAND is or'ed onto the abstract machine opcode 640 * for non-standard built-in procedures and functions. 641 */ 642 #define NSTAND 0400 643 644 #define codeon() cgenflg++ 645 #define codeoff() --cgenflg 646 647 /* 648 * Codeline is the last lino output in the code generator. 649 * It used to be used to suppress LINO operators but no 650 * more since we now count statements. 651 * Lc is the intepreter code location counter. 652 * 653 short codeline; 654 */ 655 char *lc; 656 657 658 /* 659 * Routines which need types 660 * other than "integer" to be 661 * assumed by the compiler. 662 */ 663 double atof(); 664 long lwidth(); 665 long aryconst(); 666 long a8tol(); 667 struct nl *lookup(); 668 double atof(); 669 int *tree(); 670 int *hash(); 671 char *alloc(); 672 int *calloc(); 673 char *savestr(); 674 struct nl *lookup1(); 675 struct nl *hdefnl(); 676 struct nl *defnl(); 677 struct nl *enter(); 678 struct nl *nlcopy(); 679 struct nl *tyrecl(); 680 struct nl *tyary(); 681 struct nl *fields(); 682 struct nl *variants(); 683 struct nl *deffld(); 684 struct nl *defvnt(); 685 struct nl *tyrec1(); 686 struct nl *reclook(); 687 struct nl *asgnop1(); 688 struct nl *gtype(); 689 struct nl *call(); 690 struct nl *lvalue(); 691 struct nl *rvalue(); 692 struct nl *cset(); 693 694 /* 695 * type cast NIL to keep lint happy (which is not so bad) 696 */ 697 #define NLNIL ( (struct nl *) NIL ) 698 699 /* 700 * Funny structures to use 701 * pointers in wild and wooly ways 702 */ 703 struct { 704 char pchar; 705 }; 706 struct { 707 short pint; 708 short pint2; 709 }; 710 struct { 711 long plong; 712 }; 713 struct { 714 double pdouble; 715 }; 716 717 #define OCT 1 718 #define HEX 2 719 720 /* 721 * MAIN PROGRAM VARIABLES, MISCELLANY 722 */ 723 724 /* 725 * Variables forming a data base referencing 726 * the command line arguments with the "i" option, e.g. 727 * in "pi -i scanner.i compiler.p". 728 */ 729 char **pflist; 730 short pflstc; 731 short pfcnt; 732 733 char *filename; /* current source file name */ 734 long tvec; 735 extern char *snark; /* SNARK */ 736 extern char *classes[ ]; /* maps namelist classes to string names */ 737 738 #define derror error 739 740 #ifdef PC 741 742 /* 743 * the current function number, for [ lines 744 */ 745 int ftnno; 746 747 /* 748 * the pc output stream 749 */ 750 FILE *pcstream; 751 752 #endif PC 753