1 /* Copyright (c) 1979 Regents of the University of California */ 2 3 /* static char sccsid[] = "@(#)0.h 1.20 02/28/83"; */ 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 ADDR32 107 #define TABLE_MULTIPLIER 8 108 #endif ADDR32 109 #ifdef ADDR16 110 #define TABLE_MULTIPLIER 1 111 #endif ADDR16 112 #define MAXHASH (4 * TABLE_MULTIPLIER) 113 #define MAXNL (12 * TABLE_MULTIPLIER) 114 #define MAXTREE (30 * TABLE_MULTIPLIER) 115 /* 116 * MAXDEPTH is the depth of the parse stack. 117 * STACK_MULTIPLIER is for increasing its size. 118 */ 119 #ifdef ADDR32 120 #define STACK_MULTIPLIER 8 121 #endif ADDR32 122 #ifdef ADDR16 123 #define STACK_MULTIPLIER 1 124 #endif ADDR16 125 #define MAXDEPTH ( 150 * STACK_MULTIPLIER ) 126 127 /* 128 * ERROR RELATED DEFINITIONS 129 */ 130 131 /* 132 * Exit statuses to pexit 133 * 134 * AOK 135 * ERRS Compilation errors inhibit obj productin 136 * NOSTART Errors before we ever got started 137 * DIED We ran out of memory or some such 138 */ 139 #define AOK 0 140 #define ERRS 1 141 #define NOSTART 2 142 #define DIED 3 143 144 bool Recovery; 145 146 #define eholdnl() Eholdnl = 1 147 #define nocascade() Enocascade = 1 148 149 bool Eholdnl, Enocascade; 150 151 152 /* 153 * The flag eflg is set whenever we have a hard error. 154 * The character in errpfx will precede the next error message. 155 * When cgenflg is set code generation is suppressed. 156 * This happens whenver we have an error (i.e. if eflg is set) 157 * and when we are walking the tree to determine types only. 158 */ 159 bool eflg; 160 char errpfx; 161 162 #define setpfx(x) errpfx = x 163 164 #define standard() setpfx('s') 165 #define warning() setpfx('w') 166 #define recovered() setpfx('e') 167 #define continuation() setpfx(' ') 168 169 int cgenflg; 170 171 172 /* 173 * The flag syneflg is used to suppress the diagnostics of the form 174 * E 10 a, defined in someprocedure, is neither used nor set 175 * when there were syntax errors in "someprocedure". 176 * In this case, it is likely that these warinings would be spurious. 177 */ 178 bool syneflg; 179 180 /* 181 * The compiler keeps its error messages in a file. 182 * The variable efil is the unit number on which 183 * this file is open for reading of error message text. 184 * Similarly, the file ofil is the unit of the file 185 * "obj" where we write the interpreter code. 186 */ 187 short efil; 188 short ofil; 189 short obuf[518]; 190 191 bool Enoline; 192 #define elineoff() Enoline = TRUE 193 #define elineon() Enoline = FALSE 194 195 196 /* 197 * SYMBOL TABLE STRUCTURE DEFINITIONS 198 * 199 * The symbol table is henceforth referred to as the "namelist". 200 * It consists of a number of structures of the form "nl" below. 201 * These are contained in a number of segments of the symbol 202 * table which are dynamically allocated as needed. 203 * The major namelist manipulation routines are contained in the 204 * file "nl.c". 205 * 206 * The major components of a namelist entry are the "symbol", giving 207 * a pointer into the string table for the string associated with this 208 * entry and the "class" which tells which of the (currently 19) 209 * possible types of structure this is. 210 * 211 * Many of the classes use the "type" field for a pointer to the type 212 * which the entry has. 213 * 214 * Other pieces of information in more than one class include the block 215 * in which the symbol is defined, flags indicating whether the symbol 216 * has been used and whether it has been assigned to, etc. 217 * 218 * A more complete discussion of the features of the namelist is impossible 219 * here as it would be too voluminous. Refer to the "PI 1.0 Implementation 220 * Notes" for more details. 221 */ 222 223 /* 224 * The basic namelist structure. 225 * There is a union of data types defining the stored information 226 * as pointers, integers, longs, or a double. 227 * 228 * The array disptab defines the hash header for the symbol table. 229 * Symbols are hashed based on the low 6 bits of their pointer into 230 * the string table; see the routines in the file "lookup.c" and also "fdec.c" 231 * especially "funcend". 232 */ 233 extern int pnumcnt; 234 235 struct nl { 236 char *symbol; 237 char info[4]; 238 struct nl *type; 239 struct nl *chain, *nl_next; 240 union { 241 int *un_ptr[5]; 242 int un_value[5]; 243 long un_range[2]; 244 double un_real; 245 } nl_un; 246 # ifdef PTREE 247 pPointer inTree; 248 # endif PTREE 249 }; 250 251 #define class info[0] 252 #define nl_flags info[1] 253 #define nl_block info[1] 254 #define extra_flags info[2] 255 #define align_info info[3] 256 257 #define range nl_un.un_range 258 #define value nl_un.un_value 259 #define ptr nl_un.un_ptr 260 #define real nl_un.un_real 261 262 extern struct nl *nlp, *disptab[077+1], *Fp; 263 extern struct nl nl[INL]; 264 265 266 /* 267 * NL FLAGS BITS 268 * 269 * Definitions of the usage of the bits in 270 * the nl_flags byte. Note that the low 5 bits of the 271 * byte are the "nl_block" and that some classes make use 272 * of this byte as a "width". 273 * 274 * The only non-obvious bit definition here is "NFILES" 275 * which records whether a structure contains any files. 276 * Such structures are not allowed to be dynamically allocated. 277 */ 278 279 #define BLOCKNO( flag ) ( flag & 037 ) 280 #define NLFLAGS( flag ) ( flag &~ 037 ) 281 282 #define NUSED 0100 283 #define NMOD 0040 284 #define NFORWD 0200 285 #define NFILES 0200 286 #ifdef PC 287 #define NEXTERN 0001 /* flag used to mark external funcs and procs */ 288 #define NLOCAL 0002 /* variable is a local */ 289 #define NPARAM 0004 /* variable is a parameter */ 290 #define NGLOBAL 0010 /* variable is a global */ 291 #define NREGVAR 0020 /* or'ed in if variable is in a register */ 292 #define NNLOCAL 0040 /* named local variable, not used in symbol table */ 293 #endif PC 294 295 /* 296 * used to mark value[ NL_FORV ] for loop variables 297 */ 298 #define FORVAR 1 299 300 /* 301 * Definition of the commonly used "value" fields. 302 * The most important one is NL_OFFS which gives 303 * the offset of a variable in its stack mark. 304 */ 305 #define NL_OFFS 0 306 307 #define NL_CNTR 1 308 #define NL_NLSTRT 2 309 #define NL_LINENO 3 310 #define NL_FVAR 3 311 #define NL_ENTLOC 4 /* FUNC, PROC - entry point */ 312 #define NL_FCHAIN 4 /* FFUNC, FPROC - ptr to formals */ 313 314 #define NL_GOLEV 2 315 #define NL_GOLINE 3 316 #define NL_FORV 1 317 318 /* 319 * nlp -> nl_un.un_ptr[] subscripts for records 320 * NL_FIELDLIST the chain of fixed fields of a record, in order. 321 * the fields are also chained through ptr[NL_FIELDLIST]. 322 * this does not include the tag, or fields of variants. 323 * NL_VARNT pointer to the variants of a record, 324 * these are then chained through the .chain field. 325 * NL_VTOREC pointer from a VARNT to the RECORD that is the variant. 326 * NL_TAG pointer from a RECORD to the tagfield 327 * if there are any variants. 328 * align_info the alignment of a RECORD is in info[3]. 329 */ 330 #define NL_FIELDLIST 1 331 #define NL_VARNT 2 332 #define NL_VTOREC 2 333 #define NL_TAG 3 334 /* and align_info is info[3]. #defined above */ 335 336 #define NL_ELABEL 4 /* SCAL - ptr to definition of enums */ 337 338 /* 339 * For BADUSE nl structures, NL_KINDS is a bit vector 340 * indicating the kinds of illegal usages complained about 341 * so far. For kind of bad use "kind", "1 << kind" is set. 342 * The low bit is reserved as ISUNDEF to indicate whether 343 * this identifier is totally undefined. 344 */ 345 #define NL_KINDS 0 346 347 #define ISUNDEF 1 348 349 /* 350 * variables come in three flavors: globals, parameters, locals; 351 * they can also hide in registers, but that's a different flag 352 */ 353 #define PARAMVAR 1 354 #define LOCALVAR 2 355 #define GLOBALVAR 3 356 #define NAMEDLOCALVAR 4 357 358 /* 359 * NAMELIST CLASSES 360 * 361 * The following are the namelist classes. 362 * Different classes make use of the value fields 363 * of the namelist in different ways. 364 * 365 * The namelist should be redesigned by providing 366 * a number of structure definitions with one corresponding 367 * to each namelist class, ala a variant record in Pascal. 368 */ 369 #define BADUSE 0 370 #define CONST 1 371 #define TYPE 2 372 #define VAR 3 373 #define ARRAY 4 374 #define PTRFILE 5 375 #define RECORD 6 376 #define FIELD 7 377 #define PROC 8 378 #define FUNC 9 379 #define FVAR 10 380 #define REF 11 381 #define PTR 12 382 #define FILET 13 383 #define SET 14 384 #define RANGE 15 385 #define LABEL 16 386 #define WITHPTR 17 387 #define SCAL 18 388 #define STR 19 389 #define PROG 20 390 #define IMPROPER 21 391 #define VARNT 22 392 #define FPROC 23 393 #define FFUNC 24 394 395 /* 396 * Clnames points to an array of names for the 397 * namelist classes. 398 */ 399 char **clnames; 400 401 /* 402 * PRE-DEFINED NAMELIST OFFSETS 403 * 404 * The following are the namelist offsets for the 405 * primitive types. The ones which are negative 406 * don't actually exist, but are generated and tested 407 * internally. These definitions are sensitive to the 408 * initializations in nl.c. 409 */ 410 #define TFIRST -7 411 #define TFILE -7 412 #define TREC -6 413 #define TARY -5 414 #define TSCAL -4 415 #define TPTR -3 416 #define TSET -2 417 #define TSTR -1 418 #define NIL 0 419 #define TBOOL 1 420 #define TCHAR 2 421 #define TINT 3 422 #define TDOUBLE 4 423 #define TNIL 5 424 #define T1INT 6 425 #define T2INT 7 426 #define T4INT 8 427 #define T1CHAR 9 428 #define T1BOOL 10 429 #define T8REAL 11 430 #define TLAST 11 431 432 /* 433 * SEMANTIC DEFINITIONS 434 */ 435 436 /* 437 * NOCON and SAWCON are flags in the tree telling whether 438 * a constant set is part of an expression. 439 * these are no longer used, 440 * since we now do constant sets at compile time. 441 */ 442 #define NOCON 0 443 #define SAWCON 1 444 445 /* 446 * The variable cbn gives the current block number, 447 * the variable bn is set as a side effect of a call to 448 * lookup, and is the block number of the variable which 449 * was found. 450 */ 451 short bn, cbn; 452 453 /* 454 * The variable line is the current semantic 455 * line and is set in stat.c from the numbers 456 * embedded in statement type tree nodes. 457 */ 458 short line; 459 460 /* 461 * The size of the display 462 * which defines the maximum nesting 463 * of procedures and functions allowed. 464 * Because of the flags in the current namelist 465 * this must be no greater than 32. 466 */ 467 #define DSPLYSZ 20 468 469 /* 470 * the following structure records whether a level declares 471 * any variables which are (or contain) files. 472 * this so that the runtime routines for file cleanup can be invoked. 473 */ 474 bool dfiles[ DSPLYSZ ]; 475 476 /* 477 * Structure recording information about a constant 478 * declaration. It is actually the return value from 479 * the routine "gconst", but since C doesn't support 480 * record valued functions, this is more convenient. 481 */ 482 struct { 483 struct nl *ctype; 484 short cival; 485 double crval; 486 int *cpval; 487 } con; 488 489 /* 490 * The set structure records the lower bound 491 * and upper bound with the lower bound normalized 492 * to zero when working with a set. It is set by 493 * the routine setran in var.c. 494 */ 495 struct { 496 short lwrb, uprbp; 497 } set; 498 499 /* 500 * structures of this kind are filled in by precset and used by postcset 501 * to indicate things about constant sets. 502 */ 503 struct csetstr { 504 struct nl *csettype; 505 long paircnt; 506 long singcnt; 507 bool comptime; 508 }; 509 /* 510 * The following flags are passed on calls to lvalue 511 * to indicate how the reference is to affect the usage 512 * information for the variable being referenced. 513 * MOD is used to set the NMOD flag in the namelist 514 * entry for the variable, ASGN permits diagnostics 515 * to be formed when a for variable is assigned to in 516 * the range of the loop. 517 */ 518 #define NOFLAGS 0 519 #define MOD 01 520 #define ASGN 02 521 #define NOUSE 04 522 523 /* 524 * the following flags are passed to lvalue and rvalue 525 * to tell them whether an lvalue or rvalue is required. 526 * the semantics checking is done according to the function called, 527 * but for pc, lvalue may put out an rvalue by indirecting afterwards, 528 * and rvalue may stop short of putting out the indirection. 529 */ 530 #define LREQ 01 531 #define RREQ 02 532 533 double MAXINT; 534 double MININT; 535 536 /* 537 * Variables for generation of profile information. 538 * Monflg is set when we want to generate a profile. 539 * Gocnt record the total number of goto's and 540 * cnts records the current counter for generating 541 * COUNT operators. 542 */ 543 short gocnt; 544 short cnts; 545 546 /* 547 * Most routines call "incompat" rather than asking "!compat" 548 * for historical reasons. 549 */ 550 #define incompat !compat 551 552 /* 553 * Parts records which declaration parts have been seen. 554 * The grammar allows the "label" "const" "type" "var" and routine 555 * parts to be repeated and to be in any order, so that 556 * they can be detected semantically to give better 557 * error diagnostics. 558 * 559 * The flag NONLOCALVAR indicates that a non-local var has actually 560 * been used hence the display must be saved; NONLOCALGOTO indicates 561 * that a non-local goto has been done hence that a setjmp must be done. 562 */ 563 int parts[ DSPLYSZ ]; 564 565 #define LPRT 0x0001 566 #define CPRT 0x0002 567 #define TPRT 0x0004 568 #define VPRT 0x0008 569 #define RPRT 0x0010 570 571 #define NONLOCALVAR 0x0020 572 #define NONLOCALGOTO 0x0040 573 574 /* 575 * Flags for the "you used / instead of div" diagnostic 576 */ 577 bool divchk; 578 bool divflg; 579 580 bool errcnt[DSPLYSZ]; 581 582 /* 583 * Forechain links those types which are 584 * ^ sometype 585 * so that they can be evaluated later, permitting 586 * circular, recursive list structures to be defined. 587 */ 588 struct nl *forechain; 589 590 /* 591 * Withlist links all the records which are currently 592 * opened scopes because of with statements. 593 */ 594 struct nl *withlist; 595 596 struct nl *intset; 597 struct nl *input, *output; 598 struct nl *program; 599 600 /* progseen flag used by PC to determine if 601 * a routine segment is being compiled (and 602 * therefore no program statement seen) 603 */ 604 bool progseen; 605 606 607 /* 608 * STRUCTURED STATEMENT GOTO CHECKING 609 * 610 * The variable level keeps track of the current 611 * "structured statement level" when processing the statement 612 * body of blocks. This is used in the detection of goto's into 613 * structured statements in a block. 614 * 615 * Each label's namelist entry contains two pieces of information 616 * related to this check. The first `NL_GOLEV' either contains 617 * the level at which the label was declared, `NOTYET' if the label 618 * has not yet been declared, or `DEAD' if the label is dead, i.e. 619 * if we have exited the level in which the label was defined. 620 * 621 * When we discover a "goto" statement, if the label has not 622 * been defined yet, then we record the current level and the current line 623 * for a later error check. If the label has been already become "DEAD" 624 * then a reference to it is an error. Now the compiler maintains, 625 * for each block, a linked list of the labels headed by "gotos[bn]". 626 * When we exit a structured level, we perform the routine 627 * ungoto in stat.c. It notices labels whose definition levels have been 628 * exited and makes them be dead. For labels which have not yet been 629 * defined, ungoto will maintain NL_GOLEV as the minimum structured level 630 * since the first usage of the label. It is not hard to see that the label 631 * must eventually be declared at this level or an outer level to this 632 * one or a goto into a structured statement will exist. 633 */ 634 short level; 635 struct nl *gotos[DSPLYSZ]; 636 637 #define NOTYET 10000 638 #define DEAD 10000 639 640 /* 641 * Noreach is true when the next statement will 642 * be unreachable unless something happens along 643 * (like exiting a looping construct) to save 644 * the day. 645 */ 646 bool noreach; 647 648 /* 649 * UNDEFINED VARIABLE REFERENCE STRUCTURES 650 */ 651 struct udinfo { 652 int ud_line; 653 struct udinfo *ud_next; 654 char nullch; 655 }; 656 657 /* 658 * CODE GENERATION DEFINITIONS 659 */ 660 661 /* 662 * NSTAND is or'ed onto the abstract machine opcode 663 * for non-standard built-in procedures and functions. 664 */ 665 #define NSTAND 0400 666 667 #define codeon() cgenflg++ 668 #define codeoff() --cgenflg 669 #define CGENNING ( cgenflg >= 0 ) 670 671 /* 672 * Codeline is the last lino output in the code generator. 673 * It used to be used to suppress LINO operators but no 674 * more since we now count statements. 675 * Lc is the intepreter code location counter. 676 * 677 short codeline; 678 */ 679 char *lc; 680 681 682 /* 683 * Routines which need types 684 * other than "integer" to be 685 * assumed by the compiler. 686 */ 687 double atof(); 688 long lwidth(); 689 long leven(); 690 long aryconst(); 691 long a8tol(); 692 long roundup(); 693 struct nl *tmpalloc(); 694 struct nl *lookup(); 695 double atof(); 696 int *tree(); 697 int *hash(); 698 char *alloc(); 699 int *calloc(); 700 char *savestr(); 701 char *parnam(); 702 bool fcompat(); 703 struct nl *lookup1(); 704 struct nl *hdefnl(); 705 struct nl *defnl(); 706 struct nl *enter(); 707 struct nl *nlcopy(); 708 struct nl *tyrec(); 709 struct nl *tyary(); 710 struct nl *deffld(); 711 struct nl *defvnt(); 712 struct nl *tyrec1(); 713 struct nl *reclook(); 714 struct nl *asgnop1(); 715 struct nl *gtype(); 716 struct nl *call(); 717 struct nl *lvalue(); 718 struct nl *rvalue(); 719 struct nl *cset(); 720 721 /* 722 * type cast NIL to keep lint happy (which is not so bad) 723 */ 724 #define NLNIL ( (struct nl *) NIL ) 725 726 /* 727 * Funny structures to use 728 * pointers in wild and wooly ways 729 */ 730 struct { 731 char pchar; 732 }; 733 struct { 734 short pint; 735 short pint2; 736 }; 737 struct { 738 long plong; 739 }; 740 struct { 741 double pdouble; 742 }; 743 744 #define OCT 1 745 #define HEX 2 746 747 /* 748 * MAIN PROGRAM VARIABLES, MISCELLANY 749 */ 750 751 /* 752 * Variables forming a data base referencing 753 * the command line arguments with the "i" option, e.g. 754 * in "pi -i scanner.i compiler.p". 755 */ 756 char **pflist; 757 short pflstc; 758 short pfcnt; 759 760 char *filename; /* current source file name */ 761 long tvec; 762 extern char *snark; /* SNARK */ 763 extern char *classes[ ]; /* maps namelist classes to string names */ 764 765 #define derror error 766 767 #ifdef PC 768 769 /* 770 * the current function number, for [ lines 771 */ 772 int ftnno; 773 774 /* 775 * the pc output stream 776 */ 777 FILE *pcstream; 778 779 #endif PC 780