Quiet a smoke warning in toke.c and bump the copyright year.

[perl5.git] / toke.c

/*    toke.c
 *
 *    Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
 *    2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, by Larry Wall and others
 *
 *    You may distribute under the terms of either the GNU General Public
 *    License or the Artistic License, as specified in the README file.
 *
 */

/*
 *   "It all comes from here, the stench and the peril."  --Frodo
 */

/*
 * This file is the lexer for Perl.  It's closely linked to the
 * parser, perly.y.
 *
 * The main routine is yylex(), which returns the next token.
 */

#include "EXTERN.h"
#define PERL_IN_TOKE_C
#include "perl.h"

#define yylval	(PL_parser->yylval)

/* YYINITDEPTH -- initial size of the parser's stacks.  */
#define YYINITDEPTH 200

static const char ident_too_long[] = "Identifier too long";
static const char commaless_variable_list[] = "comma-less variable list";

static void restore_rsfp(pTHX_ void *f);
#ifndef PERL_NO_UTF16_FILTER
static I32 utf16_textfilter(pTHX_ int idx, SV *sv, int maxlen);
static I32 utf16rev_textfilter(pTHX_ int idx, SV *sv, int maxlen);
#endif

#ifdef PERL_MAD
#  define CURMAD(slot,sv) if (PL_madskills) { curmad(slot,sv); sv = 0; }
#  define NEXTVAL_NEXTTOKE PL_nexttoke[PL_curforce].next_val
#else
#  define CURMAD(slot,sv)
#  define NEXTVAL_NEXTTOKE PL_nextval[PL_nexttoke]
#endif

#define XFAKEBRACK 128
#define XENUMMASK 127

#ifdef USE_UTF8_SCRIPTS
#   define UTF (!IN_BYTES)
#else
#   define UTF ((PL_linestr && DO_UTF8(PL_linestr)) || (PL_hints & HINT_UTF8))
#endif

/* In variables named $^X, these are the legal values for X.
 * 1999-02-27 mjd-perl-patch@plover.com */
#define isCONTROLVAR(x) (isUPPER(x) || strchr("[\\]^_?", (x)))

/* On MacOS, respect nonbreaking spaces */
#ifdef MACOS_TRADITIONAL
#define SPACE_OR_TAB(c) ((c)==' '||(c)=='\312'||(c)=='\t')
#else
#define SPACE_OR_TAB(c) ((c)==' '||(c)=='\t')
#endif

/* LEX_* are values for PL_lex_state, the state of the lexer.
 * They are arranged oddly so that the guard on the switch statement
 * can get by with a single comparison (if the compiler is smart enough).
 */

/* #define LEX_NOTPARSING		11 is done in perl.h. */

#define LEX_NORMAL		10 /* normal code (ie not within "...")     */
#define LEX_INTERPNORMAL	 9 /* code within a string, eg "$foo[$x+1]" */
#define LEX_INTERPCASEMOD	 8 /* expecting a \U, \Q or \E etc          */
#define LEX_INTERPPUSH		 7 /* starting a new sublex parse level     */
#define LEX_INTERPSTART		 6 /* expecting the start of a $var         */

				   /* at end of code, eg "$x" followed by:  */
#define LEX_INTERPEND		 5 /* ... eg not one of [, { or ->          */
#define LEX_INTERPENDMAYBE	 4 /* ... eg one of [, { or ->              */

#define LEX_INTERPCONCAT	 3 /* expecting anything, eg at start of
				        string or after \E, $foo, etc       */
#define LEX_INTERPCONST		 2 /* NOT USED */
#define LEX_FORMLINE		 1 /* expecting a format line               */
#define LEX_KNOWNEXT		 0 /* next token known; just return it      */


#ifdef DEBUGGING
static const char* const lex_state_names[] = {
    "KNOWNEXT",
    "FORMLINE",
    "INTERPCONST",
    "INTERPCONCAT",
    "INTERPENDMAYBE",
    "INTERPEND",
    "INTERPSTART",
    "INTERPPUSH",
    "INTERPCASEMOD",
    "INTERPNORMAL",
    "NORMAL"
};
#endif

#ifdef ff_next
#undef ff_next
#endif

#include "keywords.h"

/* CLINE is a macro that ensures PL_copline has a sane value */

#ifdef CLINE
#undef CLINE
#endif
#define CLINE (PL_copline = (CopLINE(PL_curcop) < PL_copline ? CopLINE(PL_curcop) : PL_copline))

#ifdef PERL_MAD
#  define SKIPSPACE0(s) skipspace0(s)
#  define SKIPSPACE1(s) skipspace1(s)
#  define SKIPSPACE2(s,tsv) skipspace2(s,&tsv)
#  define PEEKSPACE(s) skipspace2(s,0)
#else
#  define SKIPSPACE0(s) skipspace(s)
#  define SKIPSPACE1(s) skipspace(s)
#  define SKIPSPACE2(s,tsv) skipspace(s)
#  define PEEKSPACE(s) skipspace(s)
#endif

/*
 * Convenience functions to return different tokens and prime the
 * lexer for the next token.  They all take an argument.
 *
 * TOKEN        : generic token (used for '(', DOLSHARP, etc)
 * OPERATOR     : generic operator
 * AOPERATOR    : assignment operator
 * PREBLOCK     : beginning the block after an if, while, foreach, ...
 * PRETERMBLOCK : beginning a non-code-defining {} block (eg, hash ref)
 * PREREF       : *EXPR where EXPR is not a simple identifier
 * TERM         : expression term
 * LOOPX        : loop exiting command (goto, last, dump, etc)
 * FTST         : file test operator
 * FUN0         : zero-argument function
 * FUN1         : not used, except for not, which isn't a UNIOP
 * BOop         : bitwise or or xor
 * BAop         : bitwise and
 * SHop         : shift operator
 * PWop         : power operator
 * PMop         : pattern-matching operator
 * Aop          : addition-level operator
 * Mop          : multiplication-level operator
 * Eop          : equality-testing operator
 * Rop          : relational operator <= != gt
 *
 * Also see LOP and lop() below.
 */

#ifdef DEBUGGING /* Serve -DT. */
#   define REPORT(retval) tokereport((I32)retval)
#else
#   define REPORT(retval) (retval)
#endif

#define TOKEN(retval) return ( PL_bufptr = s, REPORT(retval))
#define OPERATOR(retval) return (PL_expect = XTERM, PL_bufptr = s, REPORT(retval))
#define AOPERATOR(retval) return ao((PL_expect = XTERM, PL_bufptr = s, REPORT(retval)))
#define PREBLOCK(retval) return (PL_expect = XBLOCK,PL_bufptr = s, REPORT(retval))
#define PRETERMBLOCK(retval) return (PL_expect = XTERMBLOCK,PL_bufptr = s, REPORT(retval))
#define PREREF(retval) return (PL_expect = XREF,PL_bufptr = s, REPORT(retval))
#define TERM(retval) return (CLINE, PL_expect = XOPERATOR, PL_bufptr = s, REPORT(retval))
#define LOOPX(f) return (yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)LOOPEX))
#define FTST(f)  return (yylval.ival=f, PL_expect=XTERMORDORDOR, PL_bufptr=s, REPORT((int)UNIOP))
#define FUN0(f)  return (yylval.ival=f, PL_expect=XOPERATOR, PL_bufptr=s, REPORT((int)FUNC0))
#define FUN1(f)  return (yylval.ival=f, PL_expect=XOPERATOR, PL_bufptr=s, REPORT((int)FUNC1))
#define BOop(f)  return ao((yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)BITOROP)))
#define BAop(f)  return ao((yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)BITANDOP)))
#define SHop(f)  return ao((yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)SHIFTOP)))
#define PWop(f)  return ao((yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)POWOP)))
#define PMop(f)  return(yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)MATCHOP))
#define Aop(f)   return ao((yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)ADDOP)))
#define Mop(f)   return ao((yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)MULOP)))
#define Eop(f)   return (yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)EQOP))
#define Rop(f)   return (yylval.ival=f, PL_expect=XTERM, PL_bufptr=s, REPORT((int)RELOP))

/* This bit of chicanery makes a unary function followed by
 * a parenthesis into a function with one argument, highest precedence.
 * The UNIDOR macro is for unary functions that can be followed by the //
 * operator (such as C<shift // 0>).
 */
#define UNI2(f,x) { \
	yylval.ival = f; \
	PL_expect = x; \
	PL_bufptr = s; \
	PL_last_uni = PL_oldbufptr; \
	PL_last_lop_op = f; \
	if (*s == '(') \
	    return REPORT( (int)FUNC1 ); \
	s = PEEKSPACE(s); \
	return REPORT( *s=='(' ? (int)FUNC1 : (int)UNIOP ); \
	}
#define UNI(f)    UNI2(f,XTERM)
#define UNIDOR(f) UNI2(f,XTERMORDORDOR)

#define UNIBRACK(f) { \
	yylval.ival = f; \
	PL_bufptr = s; \
	PL_last_uni = PL_oldbufptr; \
	if (*s == '(') \
	    return REPORT( (int)FUNC1 ); \
	s = PEEKSPACE(s); \
	return REPORT( (*s == '(') ? (int)FUNC1 : (int)UNIOP ); \
	}

/* grandfather return to old style */
#define OLDLOP(f) return(yylval.ival=f,PL_expect = XTERM,PL_bufptr = s,(int)LSTOP)

#ifdef DEBUGGING

/* how to interpret the yylval associated with the token */
enum token_type {
    TOKENTYPE_NONE,
    TOKENTYPE_IVAL,
    TOKENTYPE_OPNUM, /* yylval.ival contains an opcode number */
    TOKENTYPE_PVAL,
    TOKENTYPE_OPVAL,
    TOKENTYPE_GVVAL
};

static struct debug_tokens {
    const int token;
    enum token_type type;
    const char *name;
} const debug_tokens[] =
{
    { ADDOP,		TOKENTYPE_OPNUM,	"ADDOP" },
    { ANDAND,		TOKENTYPE_NONE,		"ANDAND" },
    { ANDOP,		TOKENTYPE_NONE,		"ANDOP" },
    { ANONSUB,		TOKENTYPE_IVAL,		"ANONSUB" },
    { ARROW,		TOKENTYPE_NONE,		"ARROW" },
    { ASSIGNOP,		TOKENTYPE_OPNUM,	"ASSIGNOP" },
    { BITANDOP,		TOKENTYPE_OPNUM,	"BITANDOP" },
    { BITOROP,		TOKENTYPE_OPNUM,	"BITOROP" },
    { COLONATTR,	TOKENTYPE_NONE,		"COLONATTR" },
    { CONTINUE,		TOKENTYPE_NONE,		"CONTINUE" },
    { DEFAULT,		TOKENTYPE_NONE,		"DEFAULT" },
    { DO,		TOKENTYPE_NONE,		"DO" },
    { DOLSHARP,		TOKENTYPE_NONE,		"DOLSHARP" },
    { DORDOR,		TOKENTYPE_NONE,		"DORDOR" },
    { DOROP,		TOKENTYPE_OPNUM,	"DOROP" },
    { DOTDOT,		TOKENTYPE_IVAL,		"DOTDOT" },
    { ELSE,		TOKENTYPE_NONE,		"ELSE" },
    { ELSIF,		TOKENTYPE_IVAL,		"ELSIF" },
    { EQOP,		TOKENTYPE_OPNUM,	"EQOP" },
    { FOR,		TOKENTYPE_IVAL,		"FOR" },
    { FORMAT,		TOKENTYPE_NONE,		"FORMAT" },
    { FUNC,		TOKENTYPE_OPNUM,	"FUNC" },
    { FUNC0,		TOKENTYPE_OPNUM,	"FUNC0" },
    { FUNC0SUB,		TOKENTYPE_OPVAL,	"FUNC0SUB" },
    { FUNC1,		TOKENTYPE_OPNUM,	"FUNC1" },
    { FUNCMETH,		TOKENTYPE_OPVAL,	"FUNCMETH" },
    { GIVEN,		TOKENTYPE_IVAL,		"GIVEN" },
    { HASHBRACK,	TOKENTYPE_NONE,		"HASHBRACK" },
    { IF,		TOKENTYPE_IVAL,		"IF" },
    { LABEL,		TOKENTYPE_PVAL,		"LABEL" },
    { LOCAL,		TOKENTYPE_IVAL,		"LOCAL" },
    { LOOPEX,		TOKENTYPE_OPNUM,	"LOOPEX" },
    { LSTOP,		TOKENTYPE_OPNUM,	"LSTOP" },
    { LSTOPSUB,		TOKENTYPE_OPVAL,	"LSTOPSUB" },
    { MATCHOP,		TOKENTYPE_OPNUM,	"MATCHOP" },
    { METHOD,		TOKENTYPE_OPVAL,	"METHOD" },
    { MULOP,		TOKENTYPE_OPNUM,	"MULOP" },
    { MY,		TOKENTYPE_IVAL,		"MY" },
    { MYSUB,		TOKENTYPE_NONE,		"MYSUB" },
    { NOAMP,		TOKENTYPE_NONE,		"NOAMP" },
    { NOTOP,		TOKENTYPE_NONE,		"NOTOP" },
    { OROP,		TOKENTYPE_IVAL,		"OROP" },
    { OROR,		TOKENTYPE_NONE,		"OROR" },
    { PACKAGE,		TOKENTYPE_NONE,		"PACKAGE" },
    { PMFUNC,		TOKENTYPE_OPVAL,	"PMFUNC" },
    { POSTDEC,		TOKENTYPE_NONE,		"POSTDEC" },
    { POSTINC,		TOKENTYPE_NONE,		"POSTINC" },
    { POWOP,		TOKENTYPE_OPNUM,	"POWOP" },
    { PREDEC,		TOKENTYPE_NONE,		"PREDEC" },
    { PREINC,		TOKENTYPE_NONE,		"PREINC" },
    { PRIVATEREF,	TOKENTYPE_OPVAL,	"PRIVATEREF" },
    { REFGEN,		TOKENTYPE_NONE,		"REFGEN" },
    { RELOP,		TOKENTYPE_OPNUM,	"RELOP" },
    { SHIFTOP,		TOKENTYPE_OPNUM,	"SHIFTOP" },
    { SUB,		TOKENTYPE_NONE,		"SUB" },
    { THING,		TOKENTYPE_OPVAL,	"THING" },
    { UMINUS,		TOKENTYPE_NONE,		"UMINUS" },
    { UNIOP,		TOKENTYPE_OPNUM,	"UNIOP" },
    { UNIOPSUB,		TOKENTYPE_OPVAL,	"UNIOPSUB" },
    { UNLESS,		TOKENTYPE_IVAL,		"UNLESS" },
    { UNTIL,		TOKENTYPE_IVAL,		"UNTIL" },
    { USE,		TOKENTYPE_IVAL,		"USE" },
    { WHEN,		TOKENTYPE_IVAL,		"WHEN" },
    { WHILE,		TOKENTYPE_IVAL,		"WHILE" },
    { WORD,		TOKENTYPE_OPVAL,	"WORD" },
    { 0,		TOKENTYPE_NONE,		NULL }
};

/* dump the returned token in rv, plus any optional arg in yylval */

STATIC int
S_tokereport(pTHX_ I32 rv)
{
    dVAR;
    if (DEBUG_T_TEST) {
	const char *name = NULL;
	enum token_type type = TOKENTYPE_NONE;
	const struct debug_tokens *p;
	SV* const report = newSVpvs("<== ");

	for (p = debug_tokens; p->token; p++) {
	    if (p->token == (int)rv) {
		name = p->name;
		type = p->type;
		break;
	    }
	}
	if (name)
	    Perl_sv_catpv(aTHX_ report, name);
	else if ((char)rv > ' ' && (char)rv < '~')
	    Perl_sv_catpvf(aTHX_ report, "'%c'", (char)rv);
	else if (!rv)
	    sv_catpvs(report, "EOF");
	else
	    Perl_sv_catpvf(aTHX_ report, "?? %"IVdf, (IV)rv);
	switch (type) {
	case TOKENTYPE_NONE:
	case TOKENTYPE_GVVAL: /* doesn't appear to be used */
	    break;
	case TOKENTYPE_IVAL:
	    Perl_sv_catpvf(aTHX_ report, "(ival=%"IVdf")", (IV)yylval.ival);
	    break;
	case TOKENTYPE_OPNUM:
	    Perl_sv_catpvf(aTHX_ report, "(ival=op_%s)",
				    PL_op_name[yylval.ival]);
	    break;
	case TOKENTYPE_PVAL:
	    Perl_sv_catpvf(aTHX_ report, "(pval=\"%s\")", yylval.pval);
	    break;
	case TOKENTYPE_OPVAL:
	    if (yylval.opval) {
		Perl_sv_catpvf(aTHX_ report, "(opval=op_%s)",
				    PL_op_name[yylval.opval->op_type]);
		if (yylval.opval->op_type == OP_CONST) {
		    Perl_sv_catpvf(aTHX_ report, " %s",
			SvPEEK(cSVOPx_sv(yylval.opval)));
		}

	    }
	    else
		sv_catpvs(report, "(opval=null)");
	    break;
	}
        PerlIO_printf(Perl_debug_log, "### %s\n\n", SvPV_nolen_const(report));
    };
    return (int)rv;
}


/* print the buffer with suitable escapes */

STATIC void
S_printbuf(pTHX_ const char* fmt, const char* s)
{
    SV* const tmp = newSVpvs("");
    PerlIO_printf(Perl_debug_log, fmt, pv_display(tmp, s, strlen(s), 0, 60));
    SvREFCNT_dec(tmp);
}

#endif

/*
 * S_ao
 *
 * This subroutine detects &&=, ||=, and //= and turns an ANDAND, OROR or DORDOR
 * into an OP_ANDASSIGN, OP_ORASSIGN, or OP_DORASSIGN
 */

STATIC int
S_ao(pTHX_ int toketype)
{
    dVAR;
    if (*PL_bufptr == '=') {
	PL_bufptr++;
	if (toketype == ANDAND)
	    yylval.ival = OP_ANDASSIGN;
	else if (toketype == OROR)
	    yylval.ival = OP_ORASSIGN;
	else if (toketype == DORDOR)
	    yylval.ival = OP_DORASSIGN;
	toketype = ASSIGNOP;
    }
    return toketype;
}

/*
 * S_no_op
 * When Perl expects an operator and finds something else, no_op
 * prints the warning.  It always prints "<something> found where
 * operator expected.  It prints "Missing semicolon on previous line?"
 * if the surprise occurs at the start of the line.  "do you need to
 * predeclare ..." is printed out for code like "sub bar; foo bar $x"
 * where the compiler doesn't know if foo is a method call or a function.
 * It prints "Missing operator before end of line" if there's nothing
 * after the missing operator, or "... before <...>" if there is something
 * after the missing operator.
 */

STATIC void
S_no_op(pTHX_ const char *what, char *s)
{
    dVAR;
    char * const oldbp = PL_bufptr;
    const bool is_first = (PL_oldbufptr == PL_linestart);

    if (!s)
	s = oldbp;
    else
	PL_bufptr = s;
    yywarn(Perl_form(aTHX_ "%s found where operator expected", what));
    if (ckWARN_d(WARN_SYNTAX)) {
	if (is_first)
	    Perl_warner(aTHX_ packWARN(WARN_SYNTAX),
		    "\t(Missing semicolon on previous line?)\n");
	else if (PL_oldoldbufptr && isIDFIRST_lazy_if(PL_oldoldbufptr,UTF)) {
	    const char *t;
	    for (t = PL_oldoldbufptr; (isALNUM_lazy_if(t,UTF) || *t == ':'); t++)
		NOOP;
	    if (t < PL_bufptr && isSPACE(*t))
		Perl_warner(aTHX_ packWARN(WARN_SYNTAX),
			"\t(Do you need to predeclare %.*s?)\n",
		    (int)(t - PL_oldoldbufptr), PL_oldoldbufptr);
	}
	else {
	    assert(s >= oldbp);
	    Perl_warner(aTHX_ packWARN(WARN_SYNTAX),
		    "\t(Missing operator before %.*s?)\n", (int)(s - oldbp), oldbp);
	}
    }
    PL_bufptr = oldbp;
}

/*
 * S_missingterm
 * Complain about missing quote/regexp/heredoc terminator.
 * If it's called with NULL then it cauterizes the line buffer.
 * If we're in a delimited string and the delimiter is a control
 * character, it's reformatted into a two-char sequence like ^C.
 * This is fatal.
 */

STATIC void
S_missingterm(pTHX_ char *s)
{
    dVAR;
    char tmpbuf[3];
    char q;
    if (s) {
	char * const nl = strrchr(s,'\n');
	if (nl)
	    *nl = '\0';
    }
    else if (
#ifdef EBCDIC
	iscntrl(PL_multi_close)
#else
	PL_multi_close < 32 || PL_multi_close == 127
#endif
	) {
	*tmpbuf = '^';
	tmpbuf[1] = (char)toCTRL(PL_multi_close);
	tmpbuf[2] = '\0';
	s = tmpbuf;
    }
    else {
	*tmpbuf = (char)PL_multi_close;
	tmpbuf[1] = '\0';
	s = tmpbuf;
    }
    q = strchr(s,'"') ? '\'' : '"';
    Perl_croak(aTHX_ "Can't find string terminator %c%s%c anywhere before EOF",q,s,q);
}

#define FEATURE_IS_ENABLED(name)				        \
	((0 != (PL_hints & HINT_LOCALIZE_HH))				\
	    && S_feature_is_enabled(aTHX_ STR_WITH_LEN(name)))
/*
 * S_feature_is_enabled
 * Check whether the named feature is enabled.
 */
STATIC bool
S_feature_is_enabled(pTHX_ const char *name, STRLEN namelen)
{
    dVAR;
    HV * const hinthv = GvHV(PL_hintgv);
    char he_name[32] = "feature_";
    (void) my_strlcpy(&he_name[8], name, 24);

    return (hinthv && hv_exists(hinthv, he_name, 8 + namelen));
}

/*
 * Perl_deprecate
 */

void
Perl_deprecate(pTHX_ const char *s)
{
    if (ckWARN(WARN_DEPRECATED))
	Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), "Use of %s is deprecated", s);
}

void
Perl_deprecate_old(pTHX_ const char *s)
{
    /* This function should NOT be called for any new deprecated warnings */
    /* Use Perl_deprecate instead                                         */
    /*                                                                    */
    /* It is here to maintain backward compatibility with the pre-5.8     */
    /* warnings category hierarchy. The "deprecated" category used to     */
    /* live under the "syntax" category. It is now a top-level category   */
    /* in its own right.                                                  */

    if (ckWARN2(WARN_DEPRECATED, WARN_SYNTAX))
	Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_SYNTAX),
			"Use of %s is deprecated", s);
}

/*
 * experimental text filters for win32 carriage-returns, utf16-to-utf8 and
 * utf16-to-utf8-reversed.
 */

#ifdef PERL_CR_FILTER
static void
strip_return(SV *sv)
{
    register const char *s = SvPVX_const(sv);
    register const char * const e = s + SvCUR(sv);
    /* outer loop optimized to do nothing if there are no CR-LFs */
    while (s < e) {
	if (*s++ == '\r' && *s == '\n') {
	    /* hit a CR-LF, need to copy the rest */
	    register char *d = s - 1;
	    *d++ = *s++;
	    while (s < e) {
		if (*s == '\r' && s[1] == '\n')
		    s++;
		*d++ = *s++;
	    }
	    SvCUR(sv) -= s - d;
	    return;
	}
    }
}

STATIC I32
S_cr_textfilter(pTHX_ int idx, SV *sv, int maxlen)
{
    const I32 count = FILTER_READ(idx+1, sv, maxlen);
    if (count > 0 && !maxlen)
	strip_return(sv);
    return count;
}
#endif

/*
 * Perl_lex_start
 * Initialize variables.  Uses the Perl save_stack to save its state (for
 * recursive calls to the parser).
 */

void
Perl_lex_start(pTHX_ SV *line)
{
    dVAR;
    const char *s = NULL;
    STRLEN len;
    yy_parser *parser;

    /* create and initialise a parser */

    Newx(parser, 1, yy_parser);
    parser->old_parser = PL_parser;
    PL_parser = parser;

    Newx(parser->stack, YYINITDEPTH, yy_stack_frame);
    parser->ps = parser->stack;
    parser->stack_size = YYINITDEPTH;

    parser->stack->state = 0;
    parser->yyerrstatus = 0;
    parser->yychar = YYEMPTY;		/* Cause a token to be read.  */

    /* initialise lexer state */

    SAVEI32(PL_lex_dojoin);
    SAVEI32(PL_lex_brackets);
    SAVEI32(PL_lex_casemods);
    SAVEI32(PL_lex_starts);
    SAVEI32(PL_lex_state);
    SAVEVPTR(PL_lex_inpat);
    SAVEI32(PL_lex_inwhat);
#ifdef PERL_MAD
    if (PL_lex_state == LEX_KNOWNEXT) {
	I32 toke = PL_lasttoke;
 	while (--toke >= 0) {
	    SAVEI32(PL_nexttoke[toke].next_type);
	    SAVEVPTR(PL_nexttoke[toke].next_val);
	    if (PL_madskills)
		SAVEVPTR(PL_nexttoke[toke].next_mad);
	}
	SAVEI32(PL_lasttoke);
    }
    SAVESPTR(PL_endwhite);
    SAVESPTR(PL_thistoken);
    SAVESPTR(PL_thiswhite);
    SAVESPTR(PL_nextwhite);
    SAVESPTR(PL_thisopen);
    SAVESPTR(PL_thisclose);
    SAVESPTR(PL_thisstuff);
    SAVEVPTR(PL_thismad);
    SAVEI32(PL_realtokenstart);
    SAVEI32(PL_faketokens);
    SAVESPTR(PL_skipwhite);
    SAVEI32(PL_curforce);
#else
    if (PL_lex_state == LEX_KNOWNEXT) {
	I32 toke = PL_nexttoke;
	while (--toke >= 0) {
	    SAVEI32(PL_nexttype[toke]);
	    SAVEVPTR(PL_nextval[toke]);
	}
	SAVEI32(PL_nexttoke);
    }
#endif
    SAVECOPLINE(PL_curcop);
    SAVEPPTR(PL_bufptr);
    SAVEPPTR(PL_bufend);
    SAVEPPTR(PL_oldbufptr);
    SAVEPPTR(PL_oldoldbufptr);
    SAVEPPTR(PL_last_lop);
    SAVEPPTR(PL_last_uni);
    SAVEPPTR(PL_linestart);
    SAVESPTR(PL_linestr);
    SAVEGENERICPV(PL_lex_brackstack);
    SAVEGENERICPV(PL_lex_casestack);
    SAVEDESTRUCTOR_X(restore_rsfp, PL_rsfp);
    SAVESPTR(PL_lex_stuff);
    SAVEI32(PL_lex_defer);
    SAVEI32(PL_sublex_info.sub_inwhat);
    SAVEI32(PL_sublex_info.super_state);
    SAVEVPTR(PL_sublex_info.sub_op);
    SAVEPPTR(PL_sublex_info.super_bufptr);
    SAVEPPTR(PL_sublex_info.super_bufend);
    SAVESPTR(PL_lex_repl);
    SAVEINT(PL_expect);
    SAVEINT(PL_lex_expect);
    SAVEI32(PL_lex_formbrack);
    SAVEVPTR(PL_lex_op);
    SAVEI32(PL_multi_close);
    SAVEI32(PL_multi_open);
    SAVEI32(PL_multi_start);
    SAVEI8(PL_pending_ident);
    SAVEBOOL(PL_preambled);

    PL_lex_state = LEX_NORMAL;
    PL_lex_defer = 0;
    PL_expect = XSTATE;
    PL_lex_brackets = 0;
    Newx(PL_lex_brackstack, 120, char);
    Newx(PL_lex_casestack, 12, char);
    PL_lex_casemods = 0;
    *PL_lex_casestack = '\0';
    PL_lex_dojoin = 0;
    PL_lex_starts = 0;
    PL_lex_stuff = NULL;
    PL_lex_repl = NULL;
    PL_lex_inpat = 0;
#ifdef PERL_MAD
    PL_lasttoke = 0;
    PL_endwhite = NULL;
    PL_faketokens = 0;
    PL_nextwhite = NULL;
    PL_realtokenstart = 0;
    PL_skipwhite = NULL;
    PL_thisclose = NULL;
    PL_thisopen = NULL;
    PL_thisstuff = NULL;
    PL_thistoken = NULL;
    PL_thiswhite = NULL;
    PL_thismad = NULL;
#else
    PL_nexttoke = 0;
#endif
    PL_lex_inwhat = 0;
    PL_sublex_info.sub_inwhat = 0;
    PL_sublex_info.super_state = 0;
    PL_sublex_info.sub_op = NULL;
    PL_sublex_info.super_bufptr = NULL;
    PL_sublex_info.super_bufend = NULL;
    PL_lex_expect = 0;
    PL_lex_formbrack = 0;
    PL_lex_op = NULL;
    PL_multi_close = 0;
    PL_multi_open = 0;
    PL_multi_start = 0;
    PL_pending_ident = '\0';
    PL_preambled = FALSE;

    if (line) {
	s = SvPV_const(line, len);
    } else {
	len = 0;
    }
    if (!len) {
	PL_linestr = newSVpvs("\n;");
    } else if (SvREADONLY(line) || s[len-1] != ';') {
	PL_linestr = newSVsv(line);
	if (s[len-1] != ';')
	    sv_catpvs(PL_linestr, "\n;");
    } else {
	SvTEMP_off(line);
	SvREFCNT_inc_simple_void_NN(line);
	PL_linestr = line;
    }
    /* PL_linestr needs to survive until end of scope, not just the next
       FREETMPS. See changes 17505 and 17546 which fixed the symptoms only.  */
    SAVEFREESV(PL_linestr);
    PL_oldoldbufptr = PL_oldbufptr = PL_bufptr = PL_linestart = SvPVX(PL_linestr);
    PL_bufend = PL_bufptr + SvCUR(PL_linestr);
    PL_last_lop = PL_last_uni = NULL;
    PL_rsfp = 0;
}

/*
 * Perl_lex_end
 * Finalizer for lexing operations.  Must be called when the parser is
 * done with the lexer.
 */

void
Perl_lex_end(pTHX)
{
    dVAR;
    PL_doextract = FALSE;
}

/*
 * S_incline
 * This subroutine has nothing to do with tilting, whether at windmills
 * or pinball tables.  Its name is short for "increment line".  It
 * increments the current line number in CopLINE(PL_curcop) and checks
 * to see whether the line starts with a comment of the form
 *    # line 500 "foo.pm"
 * If so, it sets the current line number and file to the values in the comment.
 */

STATIC void
S_incline(pTHX_ char *s)
{
    dVAR;
    char *t;
    char *n;
    char *e;
    char ch;

    CopLINE_inc(PL_curcop);
    if (*s++ != '#')
	return;
    while (SPACE_OR_TAB(*s))
	s++;
    if (strnEQ(s, "line", 4))
	s += 4;
    else
	return;
    if (SPACE_OR_TAB(*s))
	s++;
    else
	return;
    while (SPACE_OR_TAB(*s))
	s++;
    if (!isDIGIT(*s))
	return;

    n = s;
    while (isDIGIT(*s))
	s++;
    while (SPACE_OR_TAB(*s))
	s++;
    if (*s == '"' && (t = strchr(s+1, '"'))) {
	s++;
	e = t + 1;
    }
    else {
	t = s;
	while (!isSPACE(*t))
	    t++;
	e = t;
    }
    while (SPACE_OR_TAB(*e) || *e == '\r' || *e == '\f')
	e++;
    if (*e != '\n' && *e != '\0')
	return;		/* false alarm */

    ch = *t;
    *t = '\0';
    if (t - s > 0) {
#ifndef USE_ITHREADS
	const char * const cf = CopFILE(PL_curcop);
	STRLEN tmplen = cf ? strlen(cf) : 0;
	if (tmplen > 7 && strnEQ(cf, "(eval ", 6)) {
	    /* must copy *{"::_<(eval N)[oldfilename:L]"}
	     * to *{"::_<newfilename"} */
	    char smallbuf[256], smallbuf2[256];
	    char *tmpbuf, *tmpbuf2;
	    GV **gvp, *gv2;
	    STRLEN tmplen2 = strlen(s);
	    if (tmplen + 3 < sizeof smallbuf)
		tmpbuf = smallbuf;
	    else
		Newx(tmpbuf, tmplen + 3, char);
	    if (tmplen2 + 3 < sizeof smallbuf2)
		tmpbuf2 = smallbuf2;
	    else
		Newx(tmpbuf2, tmplen2 + 3, char);
	    tmpbuf[0] = tmpbuf2[0] = '_';
	    tmpbuf[1] = tmpbuf2[1] = '<';
	    memcpy(tmpbuf + 2, cf, ++tmplen);
	    memcpy(tmpbuf2 + 2, s, ++tmplen2);
	    ++tmplen; ++tmplen2;
	    gvp = (GV**)hv_fetch(PL_defstash, tmpbuf, tmplen, FALSE);
	    if (gvp) {
		gv2 = *(GV**)hv_fetch(PL_defstash, tmpbuf2, tmplen2, TRUE);
		if (!isGV(gv2)) {
		    gv_init(gv2, PL_defstash, tmpbuf2, tmplen2, FALSE);
		    /* adjust ${"::_<newfilename"} to store the new file name */
		    GvSV(gv2) = newSVpvn(tmpbuf2 + 2, tmplen2 - 2);
		    GvHV(gv2) = (HV*)SvREFCNT_inc(GvHV(*gvp));
		    GvAV(gv2) = (AV*)SvREFCNT_inc(GvAV(*gvp));
		}
	    }
	    if (tmpbuf != smallbuf) Safefree(tmpbuf);
	    if (tmpbuf2 != smallbuf2) Safefree(tmpbuf2);
	}
#endif
	CopFILE_free(PL_curcop);
	CopFILE_set(PL_curcop, s);
    }
    *t = ch;
    CopLINE_set(PL_curcop, atoi(n)-1);
}

#ifdef PERL_MAD
/* skip space before PL_thistoken */

STATIC char *
S_skipspace0(pTHX_ register char *s)
{
    s = skipspace(s);
    if (!PL_madskills)
	return s;
    if (PL_skipwhite) {
	if (!PL_thiswhite)
	    PL_thiswhite = newSVpvs("");
	sv_catsv(PL_thiswhite, PL_skipwhite);
	sv_free(PL_skipwhite);
	PL_skipwhite = 0;
    }
    PL_realtokenstart = s - SvPVX(PL_linestr);
    return s;
}

/* skip space after PL_thistoken */

STATIC char *
S_skipspace1(pTHX_ register char *s)
{
    const char *start = s;
    I32 startoff = start - SvPVX(PL_linestr);

    s = skipspace(s);
    if (!PL_madskills)
	return s;
    start = SvPVX(PL_linestr) + startoff;
    if (!PL_thistoken && PL_realtokenstart >= 0) {
	const char * const tstart = SvPVX(PL_linestr) + PL_realtokenstart;
	PL_thistoken = newSVpvn(tstart, start - tstart);
    }
    PL_realtokenstart = -1;
    if (PL_skipwhite) {
	if (!PL_nextwhite)
	    PL_nextwhite = newSVpvs("");
	sv_catsv(PL_nextwhite, PL_skipwhite);
	sv_free(PL_skipwhite);
	PL_skipwhite = 0;
    }
    return s;
}

STATIC char *
S_skipspace2(pTHX_ register char *s, SV **svp)
{
    char *start;
    const I32 bufptroff = PL_bufptr - SvPVX(PL_linestr);
    const I32 startoff = s - SvPVX(PL_linestr);

    s = skipspace(s);
    PL_bufptr = SvPVX(PL_linestr) + bufptroff;
    if (!PL_madskills || !svp)
	return s;
    start = SvPVX(PL_linestr) + startoff;
    if (!PL_thistoken && PL_realtokenstart >= 0) {
	char * const tstart = SvPVX(PL_linestr) + PL_realtokenstart;
	PL_thistoken = newSVpvn(tstart, start - tstart);
	PL_realtokenstart = -1;
    }
    if (PL_skipwhite) {
	if (!*svp)
	    *svp = newSVpvs("");
	sv_setsv(*svp, PL_skipwhite);
	sv_free(PL_skipwhite);
	PL_skipwhite = 0;
    }
    
    return s;
}
#endif

STATIC void
S_update_debugger_info_pv(pTHX_ const char *buf, STRLEN len)
{
    AV *av = CopFILEAVx(PL_curcop);
    if (av) {
	SV * const sv = newSV(0);
	sv_upgrade(sv, SVt_PVMG);
	sv_setpvn(sv, buf, len);
	(void)SvIOK_on(sv);
	SvIV_set(sv, 0);
	av_store(av, (I32)CopLINE(PL_curcop), sv);
    }
}

STATIC void
S_update_debugger_info_sv(pTHX_ SV *orig_sv)
{
    AV *av = CopFILEAVx(PL_curcop);
    if (av) {
	SV * const sv = newSV(0);
	sv_upgrade(sv, SVt_PVMG);
	sv_setsv(sv, orig_sv);
	(void)SvIOK_on(sv);
	SvIV_set(sv, 0);
	av_store(av, (I32)CopLINE(PL_curcop), sv);
    }
}

/*
 * S_skipspace
 * Called to gobble the appropriate amount and type of whitespace.
 * Skips comments as well.
 */

STATIC char *
S_skipspace(pTHX_ register char *s)
{
    dVAR;
#ifdef PERL_MAD
    int curoff;
    int startoff = s - SvPVX(PL_linestr);

    if (PL_skipwhite) {
	sv_free(PL_skipwhite);
	PL_skipwhite = 0;
    }
#endif

    if (PL_lex_formbrack && PL_lex_brackets <= PL_lex_formbrack) {
	while (s < PL_bufend && SPACE_OR_TAB(*s))
	    s++;
#ifdef PERL_MAD
	goto done;
#else
	return s;
#endif
    }
    for (;;) {
	STRLEN prevlen;
	SSize_t oldprevlen, oldoldprevlen;
	SSize_t oldloplen = 0, oldunilen = 0;
	while (s < PL_bufend && isSPACE(*s)) {
	    if (*s++ == '\n' && PL_in_eval && !PL_rsfp)
		incline(s);
	}

	/* comment */
	if (s < PL_bufend && *s == '#') {
	    while (s < PL_bufend && *s != '\n')
		s++;
	    if (s < PL_bufend) {
		s++;
		if (PL_in_eval && !PL_rsfp) {
		    incline(s);
		    continue;
		}
	    }
	}

	/* only continue to recharge the buffer if we're at the end
	 * of the buffer, we're not reading from a source filter, and
	 * we're in normal lexing mode
	 */
	if (s < PL_bufend || !PL_rsfp || PL_sublex_info.sub_inwhat ||
		PL_lex_state == LEX_FORMLINE)
#ifdef PERL_MAD
	    goto done;
#else
	    return s;
#endif

	/* try to recharge the buffer */
#ifdef PERL_MAD
	curoff = s - SvPVX(PL_linestr);
#endif

	if ((s = filter_gets(PL_linestr, PL_rsfp,
			     (prevlen = SvCUR(PL_linestr)))) == NULL)
	{
#ifdef PERL_MAD
	    if (PL_madskills && curoff != startoff) {
		if (!PL_skipwhite)
		    PL_skipwhite = newSVpvs("");
		sv_catpvn(PL_skipwhite, SvPVX(PL_linestr) + startoff,
					curoff - startoff);
	    }

	    /* mustn't throw out old stuff yet if madpropping */
	    SvCUR(PL_linestr) = curoff;
	    s = SvPVX(PL_linestr) + curoff;
	    *s = 0;
	    if (curoff && s[-1] == '\n')
		s[-1] = ' ';
#endif

	    /* end of file.  Add on the -p or -n magic */
	    /* XXX these shouldn't really be added here, can't set PL_faketokens */
	    if (PL_minus_p) {
#ifdef PERL_MAD
		sv_catpv(PL_linestr,
			 ";}continue{print or die qq(-p destination: $!\\n);}");
#else
		sv_setpv(PL_linestr,
			 ";}continue{print or die qq(-p destination: $!\\n);}");
#endif
		PL_minus_n = PL_minus_p = 0;
	    }
	    else if (PL_minus_n) {
#ifdef PERL_MAD
		sv_catpvn(PL_linestr, ";}", 2);
#else
		sv_setpvn(PL_linestr, ";}", 2);
#endif
		PL_minus_n = 0;
	    }
	    else
#ifdef PERL_MAD
		sv_catpvn(PL_linestr,";", 1);
#else
		sv_setpvn(PL_linestr,";", 1);
#endif

	    /* reset variables for next time we lex */
	    PL_oldoldbufptr = PL_oldbufptr = PL_bufptr = s = PL_linestart
		= SvPVX(PL_linestr)
#ifdef PERL_MAD
		+ curoff
#endif
		;
	    PL_bufend = SvPVX(PL_linestr) + SvCUR(PL_linestr);
	    PL_last_lop = PL_last_uni = NULL;

	    /* Close the filehandle.  Could be from -P preprocessor,
	     * STDIN, or a regular file.  If we were reading code from
	     * STDIN (because the commandline held no -e or filename)
	     * then we don't close it, we reset it so the code can
	     * read from STDIN too.
	     */

	    if (PL_preprocess && !PL_in_eval)
		(void)PerlProc_pclose(PL_rsfp);
	    else if ((PerlIO*)PL_rsfp == PerlIO_stdin())
		PerlIO_clearerr(PL_rsfp);
	    else
		(void)PerlIO_close(PL_rsfp);
	    PL_rsfp = NULL;
	    return s;
	}

	/* not at end of file, so we only read another line */
	/* make corresponding updates to old pointers, for yyerror() */
	oldprevlen = PL_oldbufptr - PL_bufend;
	oldoldprevlen = PL_oldoldbufptr - PL_bufend;
	if (PL_last_uni)
	    oldunilen = PL_last_uni - PL_bufend;
	if (PL_last_lop)
	    oldloplen = PL_last_lop - PL_bufend;
	PL_linestart = PL_bufptr = s + prevlen;
	PL_bufend = s + SvCUR(PL_linestr);
	s = PL_bufptr;
	PL_oldbufptr = s + oldprevlen;
	PL_oldoldbufptr = s + oldoldprevlen;
	if (PL_last_uni)
	    PL_last_uni = s + oldunilen;
	if (PL_last_lop)
	    PL_last_lop = s + oldloplen;
	incline(s);

	/* debugger active and we're not compiling the debugger code,
	 * so store the line into the debugger's array of lines
	 */
	if (PERLDB_LINE && PL_curstash != PL_debstash)
	    update_debugger_info_pv(PL_bufptr, PL_bufend - PL_bufptr);
    }

#ifdef PERL_MAD
  done:
    if (PL_madskills) {
	if (!PL_skipwhite)
	    PL_skipwhite = newSVpvs("");
	curoff = s - SvPVX(PL_linestr);
	if (curoff - startoff)
	    sv_catpvn(PL_skipwhite, SvPVX(PL_linestr) + startoff,
				curoff - startoff);
    }
    return s;
#endif
}

/*
 * S_check_uni
 * Check the unary operators to ensure there's no ambiguity in how they're
 * used.  An ambiguous piece of code would be:
 *     rand + 5
 * This doesn't mean rand() + 5.  Because rand() is a unary operator,
 * the +5 is its argument.
 */

STATIC void
S_check_uni(pTHX)
{
    dVAR;
    const char *s;
    const char *t;

    if (PL_oldoldbufptr != PL_last_uni)
	return;
    while (isSPACE(*PL_last_uni))
	PL_last_uni++;
    s = PL_last_uni;
    while (isALNUM_lazy_if(s,UTF) || *s == '-')
	s++;
    if ((t = strchr(s, '(')) && t < PL_bufptr)
	return;

    if (ckWARN_d(WARN_AMBIGUOUS)){
        Perl_warner(aTHX_ packWARN(WARN_AMBIGUOUS),
		   "Warning: Use of \"%.*s\" without parentheses is ambiguous",
		   (int)(s - PL_last_uni), PL_last_uni);
    }
}

/*
 * LOP : macro to build a list operator.  Its behaviour has been replaced
 * with a subroutine, S_lop() for which LOP is just another name.
 */

#define LOP(f,x) return lop(f,x,s)

/*
 * S_lop
 * Build a list operator (or something that might be one).  The rules:
 *  - if we have a next token, then it's a list operator [why?]
 *  - if the next thing is an opening paren, then it's a function
 *  - else it's a list operator
 */

STATIC I32
S_lop(pTHX_ I32 f, int x, char *s)
{
    dVAR;
    yylval.ival = f;
    CLINE;
    PL_expect = x;
    PL_bufptr = s;
    PL_last_lop = PL_oldbufptr;
    PL_last_lop_op = (OPCODE)f;
#ifdef PERL_MAD
    if (PL_lasttoke)
 	return REPORT(LSTOP);
#else
    if (PL_nexttoke)
	return REPORT(LSTOP);
#endif
    if (*s == '(')
	return REPORT(FUNC);
    s = PEEKSPACE(s);
    if (*s == '(')
	return REPORT(FUNC);
    else
	return REPORT(LSTOP);
}

#ifdef PERL_MAD
 /*
 * S_start_force
 * Sets up for an eventual force_next().  start_force(0) basically does
 * an unshift, while start_force(-1) does a push.  yylex removes items
 * on the "pop" end.
 */

STATIC void
S_start_force(pTHX_ int where)
{
    int i;

    if (where < 0)	/* so people can duplicate start_force(PL_curforce) */
	where = PL_lasttoke;
    assert(PL_curforce < 0 || PL_curforce == where);
    if (PL_curforce != where) {
	for (i = PL_lasttoke; i > where; --i) {
	    PL_nexttoke[i] = PL_nexttoke[i-1];
	}
	PL_lasttoke++;
    }
    if (PL_curforce < 0)	/* in case of duplicate start_force() */
	Zero(&PL_nexttoke[where], 1, NEXTTOKE);
    PL_curforce = where;
    if (PL_nextwhite) {
	if (PL_madskills)
	    curmad('^', newSVpvs(""));
	CURMAD('_', PL_nextwhite);
    }
}

STATIC void
S_curmad(pTHX_ char slot, SV *sv)
{
    MADPROP **where;

    if (!sv)
	return;
    if (PL_curforce < 0)
	where = &PL_thismad;
    else
	where = &PL_nexttoke[PL_curforce].next_mad;

    if (PL_faketokens)
	sv_setpvn(sv, "", 0);
    else {
	if (!IN_BYTES) {
	    if (UTF && is_utf8_string((U8*)SvPVX(sv), SvCUR(sv)))
		SvUTF8_on(sv);
	    else if (PL_encoding) {
		sv_recode_to_utf8(sv, PL_encoding);
	    }
	}
    }

    /* keep a slot open for the head of the list? */
    if (slot != '_' && *where && (*where)->mad_key == '^') {
	(*where)->mad_key = slot;
	sv_free((*where)->mad_val);
	(*where)->mad_val = (void*)sv;
    }
    else
	addmad(newMADsv(slot, sv), where, 0);
}
#else
#  define start_force(where)    NOOP
#  define curmad(slot, sv)      NOOP
#endif

/*
 * S_force_next
 * When the lexer realizes it knows the next token (for instance,
 * it is reordering tokens for the parser) then it can call S_force_next
 * to know what token to return the next time the lexer is called.  Caller
 * will need to set PL_nextval[] (or PL_nexttoke[].next_val with PERL_MAD),
 * and possibly PL_expect to ensure the lexer handles the token correctly.
 */

STATIC void
S_force_next(pTHX_ I32 type)
{
    dVAR;
#ifdef PERL_MAD
    if (PL_curforce < 0)
	start_force(PL_lasttoke);
    PL_nexttoke[PL_curforce].next_type = type;
    if (PL_lex_state != LEX_KNOWNEXT)
 	PL_lex_defer = PL_lex_state;
    PL_lex_state = LEX_KNOWNEXT;
    PL_lex_expect = PL_expect;
    PL_curforce = -1;
#else
    PL_nexttype[PL_nexttoke] = type;
    PL_nexttoke++;
    if (PL_lex_state != LEX_KNOWNEXT) {
	PL_lex_defer = PL_lex_state;
	PL_lex_expect = PL_expect;
	PL_lex_state = LEX_KNOWNEXT;
    }
#endif
}

STATIC SV *
S_newSV_maybe_utf8(pTHX_ const char *start, STRLEN len)
{
    dVAR;
    SV * const sv = newSVpvn(start,len);
    if (UTF && !IN_BYTES && is_utf8_string((const U8*)start, len))
	SvUTF8_on(sv);
    return sv;
}

/*
 * S_force_word
 * When the lexer knows the next thing is a word (for instance, it has
 * just seen -> and it knows that the next char is a word char, then
 * it calls S_force_word to stick the next word into the PL_nexttoke/val
 * lookahead.
 *
 * Arguments:
 *   char *start : buffer position (must be within PL_linestr)
 *   int token   : PL_next* will be this type of bare word (e.g., METHOD,WORD)
 *   int check_keyword : if true, Perl checks to make sure the word isn't
 *       a keyword (do this if the word is a label, e.g. goto FOO)
 *   int allow_pack : if true, : characters will also be allowed (require,
 *       use, etc. do this)
 *   int allow_initial_tick : used by the "sub" lexer only.
 */

STATIC char *
S_force_word(pTHX_ register char *start, int token, int check_keyword, int allow_pack, int allow_initial_tick)
{
    dVAR;
    register char *s;
    STRLEN len;

    start = SKIPSPACE1(start);
    s = start;
    if (isIDFIRST_lazy_if(s,UTF) ||
	(allow_pack && *s == ':') ||
	(allow_initial_tick && *s == '\'') )
    {
	s = scan_word(s, PL_tokenbuf, sizeof PL_tokenbuf, allow_pack, &len);
	if (check_keyword && keyword(PL_tokenbuf, len, 0))
	    return start;
	start_force(PL_curforce);
	if (PL_madskills)
	    curmad('X', newSVpvn(start,s-start));
	if (token == METHOD) {
	    s = SKIPSPACE1(s);
	    if (*s == '(')
		PL_expect = XTERM;
	    else {
		PL_expect = XOPERATOR;
	    }
	}
	NEXTVAL_NEXTTOKE.opval
	    = (OP*)newSVOP(OP_CONST,0,
			   S_newSV_maybe_utf8(aTHX_ PL_tokenbuf, len));
	NEXTVAL_NEXTTOKE.opval->op_private |= OPpCONST_BARE;
	force_next(token);
    }
    return s;
}

/*
 * S_force_ident
 * Called when the lexer wants $foo *foo &foo etc, but the program
 * text only contains the "foo" portion.  The first argument is a pointer
 * to the "foo", and the second argument is the type symbol to prefix.
 * Forces the next token to be a "WORD".
 * Creates the symbol if it didn't already exist (via gv_fetchpv()).
 */

STATIC void
S_force_ident(pTHX_ register const char *s, int kind)
{
    dVAR;
    if (*s) {
	const STRLEN len = strlen(s);
	OP* const o = (OP*)newSVOP(OP_CONST, 0, newSVpvn(s, len));
	start_force(PL_curforce);
	NEXTVAL_NEXTTOKE.opval = o;
	force_next(WORD);
	if (kind) {
	    o->op_private = OPpCONST_ENTERED;
	    /* XXX see note in pp_entereval() for why we forgo typo
	       warnings if the symbol must be introduced in an eval.
	       GSAR 96-10-12 */
	    gv_fetchpvn_flags(s, len,
			      PL_in_eval ? (GV_ADDMULTI | GV_ADDINEVAL)
			      : GV_ADD,
			      kind == '$' ? SVt_PV :
			      kind == '@' ? SVt_PVAV :
			      kind == '%' ? SVt_PVHV :
			      SVt_PVGV
			      );
	}
    }
}

NV
Perl_str_to_version(pTHX_ SV *sv)
{
    NV retval = 0.0;
    NV nshift = 1.0;
    STRLEN len;
    const char *start = SvPV_const(sv,len);
    const char * const end = start + len;
    const bool utf = SvUTF8(sv) ? TRUE : FALSE;
    while (start < end) {
	STRLEN skip;
	UV n;
	if (utf)
	    n = utf8n_to_uvchr((U8*)start, len, &skip, 0);
	else {
	    n = *(U8*)start;
	    skip = 1;
	}
	retval += ((NV)n)/nshift;
	start += skip;
	nshift *= 1000;
    }
    return retval;
}

/*
 * S_force_version
 * Forces the next token to be a version number.
 * If the next token appears to be an invalid version number, (e.g. "v2b"),
 * and if "guessing" is TRUE, then no new token is created (and the caller
 * must use an alternative parsing method).
 */

STATIC char *
S_force_version(pTHX_ char *s, int guessing)
{
    dVAR;
    OP *version = NULL;
    char *d;
#ifdef PERL_MAD
    I32 startoff = s - SvPVX(PL_linestr);
#endif

    s = SKIPSPACE1(s);

    d = s;
    if (*d == 'v')
	d++;
    if (isDIGIT(*d)) {
	while (isDIGIT(*d) || *d == '_' || *d == '.')
	    d++;
#ifdef PERL_MAD
	if (PL_madskills) {
	    start_force(PL_curforce);
	    curmad('X', newSVpvn(s,d-s));
	}
#endif
        if (*d == ';' || isSPACE(*d) || *d == '}' || !*d) {
	    SV *ver;
            s = scan_num(s, &yylval);
            version = yylval.opval;
	    ver = cSVOPx(version)->op_sv;
	    if (SvPOK(ver) && !SvNIOK(ver)) {
		SvUPGRADE(ver, SVt_PVNV);
		SvNV_set(ver, str_to_version(ver));
		SvNOK_on(ver);		/* hint that it is a version */
	    }
        }
	else if (guessing) {
#ifdef PERL_MAD
	    if (PL_madskills) {
		sv_free(PL_nextwhite);	/* let next token collect whitespace */
		PL_nextwhite = 0;
		s = SvPVX(PL_linestr) + startoff;
	    }
#endif
	    return s;
	}
    }

#ifdef PERL_MAD
    if (PL_madskills && !version) {
	sv_free(PL_nextwhite);	/* let next token collect whitespace */
	PL_nextwhite = 0;
	s = SvPVX(PL_linestr) + startoff;
    }
#endif
    /* NOTE: The parser sees the package name and the VERSION swapped */
    start_force(PL_curforce);
    NEXTVAL_NEXTTOKE.opval = version;
    force_next(WORD);

    return s;
}

/*
 * S_tokeq
 * Tokenize a quoted string passed in as an SV.  It finds the next
 * chunk, up to end of string or a backslash.  It may make a new
 * SV containing that chunk (if HINT_NEW_STRING is on).  It also
 * turns \\ into \.
 */

STATIC SV *
S_tokeq(pTHX_ SV *sv)
{
    dVAR;
    register char *s;
    register char *send;
    register char *d;
    STRLEN len = 0;
    SV *pv = sv;

    if (!SvLEN(sv))
	goto finish;

    s = SvPV_force(sv, len);
    if (SvTYPE(sv) >= SVt_PVIV && SvIVX(sv) == -1)
	goto finish;
    send = s + len;
    while (s < send && *s != '\\')
	s++;
    if (s == send)
	goto finish;
    d = s;
    if ( PL_hints & HINT_NEW_STRING ) {
	pv = sv_2mortal(newSVpvn(SvPVX_const(pv), len));
	if (SvUTF8(sv))
	    SvUTF8_on(pv);
    }
    while (s < send) {
	if (*s == '\\') {
	    if (s + 1 < send && (s[1] == '\\'))
		s++;		/* all that, just for this */
	}
	*d++ = *s++;
    }
    *d = '\0';
    SvCUR_set(sv, d - SvPVX_const(sv));
  finish:
    if ( PL_hints & HINT_NEW_STRING )
       return new_constant(NULL, 0, "q", sv, pv, "q");
    return sv;
}

/*
 * Now come three functions related to double-quote context,
 * S_sublex_start, S_sublex_push, and S_sublex_done.  They're used when
 * converting things like "\u\Lgnat" into ucfirst(lc("gnat")).  They
 * interact with PL_lex_state, and create fake ( ... ) argument lists
 * to handle functions and concatenation.
 * They assume that whoever calls them will be setting up a fake
 * join call, because each subthing puts a ',' after it.  This lets
 *   "lower \luPpEr"
 * become
 *  join($, , 'lower ', lcfirst( 'uPpEr', ) ,)
 *
 * (I'm not sure whether the spurious commas at the end of lcfirst's
 * arguments and join's arguments are created or not).
 */

/*
 * S_sublex_start
 * Assumes that yylval.ival is the op we're creating (e.g. OP_LCFIRST).
 *
 * Pattern matching will set PL_lex_op to the pattern-matching op to
 * make (we return THING if yylval.ival is OP_NULL, PMFUNC otherwise).
 *
 * OP_CONST and OP_READLINE are easy--just make the new op and return.
 *
 * Everything else becomes a FUNC.
 *
 * Sets PL_lex_state to LEX_INTERPPUSH unless (ival was OP_NULL or we
 * had an OP_CONST or OP_READLINE).  This just sets us up for a
 * call to S_sublex_push().
 */

STATIC I32
S_sublex_start(pTHX)
{
    dVAR;
    register const I32 op_type = yylval.ival;

    if (op_type == OP_NULL) {
	yylval.opval = PL_lex_op;
	PL_lex_op = NULL;
	return THING;
    }
    if (op_type == OP_CONST || op_type == OP_READLINE) {
	SV *sv = tokeq(PL_lex_stuff);

	if (SvTYPE(sv) == SVt_PVIV) {
	    /* Overloaded constants, nothing fancy: Convert to SVt_PV: */
	    STRLEN len;
	    const char * const p = SvPV_const(sv, len);
	    SV * const nsv = newSVpvn(p, len);
	    if (SvUTF8(sv))
		SvUTF8_on(nsv);
	    SvREFCNT_dec(sv);
	    sv = nsv;
	}
	yylval.opval = (OP*)newSVOP(op_type, 0, sv);
	PL_lex_stuff = NULL;
	/* Allow <FH> // "foo" */
	if (op_type == OP_READLINE)
	    PL_expect = XTERMORDORDOR;
	return THING;
    }
    else if (op_type == OP_BACKTICK && PL_lex_op) {
	/* readpipe() vas overriden */
	cSVOPx(cLISTOPx(cUNOPx(PL_lex_op)->op_first)->op_first->op_sibling)->op_sv = tokeq(PL_lex_stuff);
	yylval.opval = PL_lex_op;
	PL_lex_op = NULL;
	PL_lex_stuff = NULL;
	return THING;
    }

    PL_sublex_info.super_state = PL_lex_state;
    PL_sublex_info.sub_inwhat = op_type;
    PL_sublex_info.sub_op = PL_lex_op;
    PL_lex_state = LEX_INTERPPUSH;

    PL_expect = XTERM;
    if (PL_lex_op) {
	yylval.opval = PL_lex_op;
	PL_lex_op = NULL;
	return PMFUNC;
    }
    else
	return FUNC;
}

/*
 * S_sublex_push
 * Create a new scope to save the lexing state.  The scope will be
 * ended in S_sublex_done.  Returns a '(', starting the function arguments
 * to the uc, lc, etc. found before.
 * Sets PL_lex_state to LEX_INTERPCONCAT.
 */

STATIC I32
S_sublex_push(pTHX)
{
    dVAR;
    ENTER;

    PL_lex_state = PL_sublex_info.super_state;
    SAVEI32(PL_lex_dojoin);
    SAVEI32(PL_lex_brackets);
    SAVEI32(PL_lex_casemods);
    SAVEI32(PL_lex_starts);
    SAVEI32(PL_lex_state);
    SAVEVPTR(PL_lex_inpat);
    SAVEI32(PL_lex_inwhat);
    SAVECOPLINE(PL_curcop);
    SAVEPPTR(PL_bufptr);
    SAVEPPTR(PL_bufend);
    SAVEPPTR(PL_oldbufptr);
    SAVEPPTR(PL_oldoldbufptr);
    SAVEPPTR(PL_last_lop);
    SAVEPPTR(PL_last_uni);
    SAVEPPTR(PL_linestart);
    SAVESPTR(PL_linestr);
    SAVEGENERICPV(PL_lex_brackstack);
    SAVEGENERICPV(PL_lex_casestack);

    PL_linestr = PL_lex_stuff;
    PL_lex_stuff = NULL;

    PL_bufend = PL_bufptr = PL_oldbufptr = PL_oldoldbufptr = PL_linestart
	= SvPVX(PL_linestr);
    PL_bufend += SvCUR(PL_linestr);
    PL_last_lop = PL_last_uni = NULL;
    SAVEFREESV(PL_linestr);

    PL_lex_dojoin = FALSE;
    PL_lex_brackets = 0;
    Newx(PL_lex_brackstack, 120, char);
    Newx(PL_lex_casestack, 12, char);
    PL_lex_casemods = 0;
    *PL_lex_casestack = '\0';
    PL_lex_starts = 0;
    PL_lex_state = LEX_INTERPCONCAT;
    CopLINE_set(PL_curcop, (line_t)PL_multi_start);

    PL_lex_inwhat = PL_sublex_info.sub_inwhat;
    if (PL_lex_inwhat == OP_MATCH || PL_lex_inwhat == OP_QR || PL_lex_inwhat == OP_SUBST)
	PL_lex_inpat = PL_sublex_info.sub_op;
    else
	PL_lex_inpat = NULL;

    return '(';
}

/*
 * S_sublex_done
 * Restores lexer state after a S_sublex_push.
 */

STATIC I32
S_sublex_done(pTHX)
{
    dVAR;
    if (!PL_lex_starts++) {
	SV * const sv = newSVpvs("");
	if (SvUTF8(PL_linestr))
	    SvUTF8_on(sv);
	PL_expect = XOPERATOR;
	yylval.opval = (OP*)newSVOP(OP_CONST, 0, sv);
	return THING;
    }

    if (PL_lex_casemods) {		/* oops, we've got some unbalanced parens */
	PL_lex_state = LEX_INTERPCASEMOD;
	return yylex();
    }

    /* Is there a right-hand side to take care of? (s//RHS/ or tr//RHS/) */
    if (PL_lex_repl && (PL_lex_inwhat == OP_SUBST || PL_lex_inwhat == OP_TRANS)) {
	PL_linestr = PL_lex_repl;
	PL_lex_inpat = 0;
	PL_bufend = PL_bufptr = PL_oldbufptr = PL_oldoldbufptr = PL_linestart = SvPVX(PL_linestr);
	PL_bufend += SvCUR(PL_linestr);
	PL_last_lop = PL_last_uni = NULL;
	SAVEFREESV(PL_linestr);
	PL_lex_dojoin = FALSE;
	PL_lex_brackets = 0;
	PL_lex_casemods = 0;
	*PL_lex_casestack = '\0';
	PL_lex_starts = 0;
	if (SvEVALED(PL_lex_repl)) {
	    PL_lex_state = LEX_INTERPNORMAL;
	    PL_lex_starts++;
	    /*	we don't clear PL_lex_repl here, so that we can check later
		whether this is an evalled subst; that means we rely on the
		logic to ensure sublex_done() is called again only via the
		branch (in yylex()) that clears PL_lex_repl, else we'll loop */
	}
	else {
	    PL_lex_state = LEX_INTERPCONCAT;
	    PL_lex_repl = NULL;
	}
	return ',';
    }
    else {
#ifdef PERL_MAD
	if (PL_madskills) {
	    if (PL_thiswhite) {
		if (!PL_endwhite)
		    PL_endwhite = newSVpvs("");
		sv_catsv(PL_endwhite, PL_thiswhite);
		PL_thiswhite = 0;
	    }
	    if (PL_thistoken)
		sv_setpvn(PL_thistoken,"",0);
	    else
		PL_realtokenstart = -1;
	}
#endif
	LEAVE;
	PL_bufend = SvPVX(PL_linestr);
	PL_bufend += SvCUR(PL_linestr);
	PL_expect = XOPERATOR;
	PL_sublex_info.sub_inwhat = 0;
	return ')';
    }
}

/*
  scan_const

  Extracts a pattern, double-quoted string, or transliteration.  This
  is terrifying code.

  It looks at PL_lex_inwhat and PL_lex_inpat to find out whether it's
  processing a pattern (PL_lex_inpat is true), a transliteration
  (PL_lex_inwhat == OP_TRANS is true), or a double-quoted string.

  Returns a pointer to the character scanned up to. If this is
  advanced from the start pointer supplied (i.e. if anything was
  successfully parsed), will leave an OP for the substring scanned
  in yylval. Caller must intuit reason for not parsing further
  by looking at the next characters herself.

  In patterns:
    backslashes:
      double-quoted style: \r and \n
      regexp special ones: \D \s
      constants: \x31
      backrefs: \1
      case and quoting: \U \Q \E
    stops on @ and $, but not for $ as tail anchor

  In transliterations:
    characters are VERY literal, except for - not at the start or end
    of the string, which indicates a range. If the range is in bytes,
    scan_const expands the range to the full set of intermediate
    characters. If the range is in utf8, the hyphen is replaced with
    a certain range mark which will be handled by pmtrans() in op.c.

  In double-quoted strings:
    backslashes:
      double-quoted style: \r and \n
      constants: \x31
      deprecated backrefs: \1 (in substitution replacements)
      case and quoting: \U \Q \E
    stops on @ and $

  scan_const does *not* construct ops to handle interpolated strings.
  It stops processing as soon as it finds an embedded $ or @ variable
  and leaves it to the caller to work out what's going on.

  embedded arrays (whether in pattern or not) could be:
      @foo, @::foo, @'foo, @{foo}, @$foo, @+, @-.

  $ in double-quoted strings must be the symbol of an embedded scalar.

  $ in pattern could be $foo or could be tail anchor.  Assumption:
  it's a tail anchor if $ is the last thing in the string, or if it's
  followed by one of "()| \r\n\t"

  \1 (backreferences) are turned into $1

  The structure of the code is
      while (there's a character to process) {
	  handle transliteration ranges
	  skip regexp comments /(?#comment)/ and codes /(?{code})/
	  skip #-initiated comments in //x patterns
	  check for embedded arrays
	  check for embedded scalars
	  if (backslash) {
	      leave intact backslashes from leaveit (below)
	      deprecate \1 in substitution replacements
	      handle string-changing backslashes \l \U \Q \E, etc.
	      switch (what was escaped) {
		  handle \- in a transliteration (becomes a literal -)
		  handle \132 (octal characters)
		  handle \x15 and \x{1234} (hex characters)
		  handle \N{name} (named characters)
		  handle \cV (control characters)
		  handle printf-style backslashes (\f, \r, \n, etc)
	      } (end switch)
	  } (end if backslash)
    } (end while character to read)
		
*/

STATIC char *
S_scan_const(pTHX_ char *start)
{
    dVAR;
    register char *send = PL_bufend;		/* end of the constant */
    SV *sv = newSV(send - start);		/* sv for the constant */
    register char *s = start;			/* start of the constant */
    register char *d = SvPVX(sv);		/* destination for copies */
    bool dorange = FALSE;			/* are we in a translit range? */
    bool didrange = FALSE;		        /* did we just finish a range? */
    I32  has_utf8 = FALSE;			/* Output constant is UTF8 */
    I32  this_utf8 = UTF;			/* The source string is assumed to be UTF8 */
    UV uv;
#ifdef EBCDIC
    UV literal_endpoint = 0;
    bool native_range = TRUE; /* turned to FALSE if the first endpoint is Unicode. */
#endif

    if (PL_lex_inwhat == OP_TRANS && PL_sublex_info.sub_op) {
	/* If we are doing a trans and we know we want UTF8 set expectation */
	has_utf8   = PL_sublex_info.sub_op->op_private & (OPpTRANS_FROM_UTF|OPpTRANS_TO_UTF);
	this_utf8  = PL_sublex_info.sub_op->op_private & (PL_lex_repl ? OPpTRANS_FROM_UTF : OPpTRANS_TO_UTF);
    }


    while (s < send || dorange) {
        /* get transliterations out of the way (they're most literal) */
	if (PL_lex_inwhat == OP_TRANS) {
	    /* expand a range A-Z to the full set of characters.  AIE! */
	    if (dorange) {
		I32 i;				/* current expanded character */
		I32 min;			/* first character in range */
		I32 max;			/* last character in range */

#ifdef EBCDIC
		UV uvmax = 0;
#endif

		if (has_utf8
#ifdef EBCDIC
		    && !native_range
#endif
		    ) {
		    char * const c = (char*)utf8_hop((U8*)d, -1);
		    char *e = d++;
		    while (e-- > c)
			*(e + 1) = *e;
		    *c = (char)UTF_TO_NATIVE(0xff);
		    /* mark the range as done, and continue */
		    dorange = FALSE;
		    didrange = TRUE;
		    continue;
		}

		i = d - SvPVX_const(sv);		/* remember current offset */
#ifdef EBCDIC
                SvGROW(sv,
		       SvLEN(sv) + (has_utf8 ?
				    (512 - UTF_CONTINUATION_MARK +
				     UNISKIP(0x100))
				    : 256));
                /* How many two-byte within 0..255: 128 in UTF-8,
		 * 96 in UTF-8-mod. */
#else
		SvGROW(sv, SvLEN(sv) + 256);	/* never more than 256 chars in a range */
#endif
		d = SvPVX(sv) + i;		/* refresh d after realloc */
#ifdef EBCDIC
                if (has_utf8) {
                    int j;
                    for (j = 0; j <= 1; j++) {
                        char * const c = (char*)utf8_hop((U8*)d, -1);
                        const UV uv    = utf8n_to_uvchr((U8*)c, d - c, NULL, 0);
                        if (j)
                            min = (U8)uv;
                        else if (uv < 256)
                            max = (U8)uv;
                        else {
                            max = (U8)0xff; /* only to \xff */
                            uvmax = uv; /* \x{100} to uvmax */
                        }
                        d = c; /* eat endpoint chars */
                     }
                }
               else {
#endif
		   d -= 2;		/* eat the first char and the - */
		   min = (U8)*d;	/* first char in range */
		   max = (U8)d[1];	/* last char in range  */
#ifdef EBCDIC
	       }
#endif

                if (min > max) {
		    Perl_croak(aTHX_
			       "Invalid range \"%c-%c\" in transliteration operator",
			       (char)min, (char)max);
                }

#ifdef EBCDIC
		if (literal_endpoint == 2 &&
		    ((isLOWER(min) && isLOWER(max)) ||
		     (isUPPER(min) && isUPPER(max)))) {
		    if (isLOWER(min)) {
			for (i = min; i <= max; i++)
			    if (isLOWER(i))
				*d++ = NATIVE_TO_NEED(has_utf8,i);
		    } else {
			for (i = min; i <= max; i++)
			    if (isUPPER(i))
				*d++ = NATIVE_TO_NEED(has_utf8,i);
		    }
		}
		else
#endif
		    for (i = min; i <= max; i++)
#ifdef EBCDIC
                        if (has_utf8) {
                            const U8 ch = (U8)NATIVE_TO_UTF(i);
                            if (UNI_IS_INVARIANT(ch))
                                *d++ = (U8)i;
                            else {
                                *d++ = (U8)UTF8_EIGHT_BIT_HI(ch);
                                *d++ = (U8)UTF8_EIGHT_BIT_LO(ch);
                            }
                        }
                        else
#endif
                            *d++ = (char)i;
 
#ifdef EBCDIC
                if (uvmax) {
                    d = (char*)uvchr_to_utf8((U8*)d, 0x100);
                    if (uvmax > 0x101)
                        *d++ = (char)UTF_TO_NATIVE(0xff);
                    if (uvmax > 0x100)
                        d = (char*)uvchr_to_utf8((U8*)d, uvmax);
                }
#endif

		/* mark the range as done, and continue */
		dorange = FALSE;
		didrange = TRUE;
#ifdef EBCDIC
		literal_endpoint = 0;
#endif
		continue;
	    }

	    /* range begins (ignore - as first or last char) */
	    else if (*s == '-' && s+1 < send  && s != start) {
		if (didrange) {
		    Perl_croak(aTHX_ "Ambiguous range in transliteration operator");
		}
		if (has_utf8
#ifdef EBCDIC
		    && !native_range
#endif
		    ) {
		    *d++ = (char)UTF_TO_NATIVE(0xff);	/* use illegal utf8 byte--see pmtrans */
		    s++;
		    continue;
		}
		dorange = TRUE;
		s++;
	    }
	    else {
		didrange = FALSE;
#ifdef EBCDIC
		literal_endpoint = 0;
		native_range = TRUE;
#endif
	    }
	}

	/* if we get here, we're not doing a transliteration */

	/* skip for regexp comments /(?#comment)/ and code /(?{code})/,
	   except for the last char, which will be done separately. */
	else if (*s == '(' && PL_lex_inpat && s[1] == '?') {
	    if (s[2] == '#') {
		while (s+1 < send && *s != ')')
		    *d++ = NATIVE_TO_NEED(has_utf8,*s++);
	    }
	    else if (s[2] == '{' /* This should match regcomp.c */
		     || ((s[2] == 'p' || s[2] == '?') && s[3] == '{'))
	    {
		I32 count = 1;
		char *regparse = s + (s[2] == '{' ? 3 : 4);
		char c;

		while (count && (c = *regparse)) {
		    if (c == '\\' && regparse[1])
			regparse++;
		    else if (c == '{')
			count++;
		    else if (c == '}')
			count--;
		    regparse++;
		}
		if (*regparse != ')')
		    regparse--;		/* Leave one char for continuation. */
		while (s < regparse)
		    *d++ = NATIVE_TO_NEED(has_utf8,*s++);
	    }
	}

	/* likewise skip #-initiated comments in //x patterns */
	else if (*s == '#' && PL_lex_inpat &&
	  ((PMOP*)PL_lex_inpat)->op_pmflags & PMf_EXTENDED) {
	    while (s+1 < send && *s != '\n')
		*d++ = NATIVE_TO_NEED(has_utf8,*s++);
	}

	/* check for embedded arrays
	   (@foo, @::foo, @'foo, @{foo}, @$foo, @+, @-)
	   */
	else if (*s == '@' && s[1]) {
	    if (isALNUM_lazy_if(s+1,UTF))
		break;
	    if (strchr(":'{$", s[1]))
		break;
	    if (!PL_lex_inpat && (s[1] == '+' || s[1] == '-'))
		break; /* in regexp, neither @+ nor @- are interpolated */
	}

	/* check for embedded scalars.  only stop if we're sure it's a
	   variable.
        */
	else if (*s == '$') {
	    if (!PL_lex_inpat)	/* not a regexp, so $ must be var */
		break;
	    if (s + 1 < send && !strchr("()| \r\n\t", s[1]))
		break;		/* in regexp, $ might be tail anchor */
	}

	/* End of else if chain - OP_TRANS rejoin rest */

	/* backslashes */
	if (*s == '\\' && s+1 < send) {
	    s++;

	    /* deprecate \1 in strings and substitution replacements */
	    if (PL_lex_inwhat == OP_SUBST && !PL_lex_inpat &&
		isDIGIT(*s) && *s != '0' && !isDIGIT(s[1]))
	    {
		if (ckWARN(WARN_SYNTAX))
		    Perl_warner(aTHX_ packWARN(WARN_SYNTAX), "\\%c better written as $%c", *s, *s);
		*--s = '$';
		break;
	    }

	    /* string-change backslash escapes */
	    if (PL_lex_inwhat != OP_TRANS && *s && strchr("lLuUEQ", *s)) {
		--s;
		break;
	    }
	    /* skip any other backslash escapes in a pattern */
	    else if (PL_lex_inpat) {
		*d++ = NATIVE_TO_NEED(has_utf8,'\\');
		goto default_action;
	    }

	    /* if we get here, it's either a quoted -, or a digit */
	    switch (*s) {

	    /* quoted - in transliterations */
	    case '-':
		if (PL_lex_inwhat == OP_TRANS) {
		    *d++ = *s++;
		    continue;
		}
		/* FALL THROUGH */
	    default:
	        {
		    if ((isALPHA(*s) || isDIGIT(*s)) &&
			ckWARN(WARN_MISC))
			Perl_warner(aTHX_ packWARN(WARN_MISC),
				    "Unrecognized escape \\%c passed through",
				    *s);
		    /* default action is to copy the quoted character */
		    goto default_action;
		}

	    /* \132 indicates an octal constant */
	    case '0': case '1': case '2': case '3':
	    case '4': case '5': case '6': case '7':
		{
                    I32 flags = 0;
                    STRLEN len = 3;
		    uv = grok_oct(s, &len, &flags, NULL);
		    s += len;
		}
		goto NUM_ESCAPE_INSERT;

	    /* \x24 indicates a hex constant */
	    case 'x':
		++s;
		if (*s == '{') {
		    char* const e = strchr(s, '}');
                    I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
                      PERL_SCAN_DISALLOW_PREFIX;
		    STRLEN len;

                    ++s;
		    if (!e) {
			yyerror("Missing right brace on \\x{}");
			continue;
		    }
                    len = e - s;
		    uv = grok_hex(s, &len, &flags, NULL);
		    s = e + 1;
		}
		else {
		    {
			STRLEN len = 2;
                        I32 flags = PERL_SCAN_DISALLOW_PREFIX;
			uv = grok_hex(s, &len, &flags, NULL);
			s += len;
		    }
		}

	      NUM_ESCAPE_INSERT:
		/* Insert oct or hex escaped character.
		 * There will always enough room in sv since such
		 * escapes will be longer than any UTF-8 sequence
		 * they can end up as. */
		
		/* We need to map to chars to ASCII before doing the tests
		   to cover EBCDIC
		*/
		if (!UNI_IS_INVARIANT(NATIVE_TO_UNI(uv))) {
		    if (!has_utf8 && uv > 255) {
		        /* Might need to recode whatever we have
			 * accumulated so far if it contains any
			 * hibit chars.
			 *
			 * (Can't we keep track of that and avoid
			 *  this rescan? --jhi)
			 */
			int hicount = 0;
			U8 *c;
			for (c = (U8 *) SvPVX(sv); c < (U8 *)d; c++) {
			    if (!NATIVE_IS_INVARIANT(*c)) {
			        hicount++;
			    }
			}
			if (hicount) {
			    const STRLEN offset = d - SvPVX_const(sv);
			    U8 *src, *dst;
			    d = SvGROW(sv, SvLEN(sv) + hicount + 1) + offset;
			    src = (U8 *)d - 1;
			    dst = src+hicount;
			    d  += hicount;
			    while (src >= (const U8 *)SvPVX_const(sv)) {
			        if (!NATIVE_IS_INVARIANT(*src)) {
				    const U8 ch = NATIVE_TO_ASCII(*src);
				    *dst-- = (U8)UTF8_EIGHT_BIT_LO(ch);
				    *dst-- = (U8)UTF8_EIGHT_BIT_HI(ch);
			        }
			        else {
				    *dst-- = *src;
			        }
				src--;
			    }
                        }
                    }

                    if (has_utf8 || uv > 255) {
		        d = (char*)uvchr_to_utf8((U8*)d, uv);
			has_utf8 = TRUE;
			if (PL_lex_inwhat == OP_TRANS &&
			    PL_sublex_info.sub_op) {
			    PL_sublex_info.sub_op->op_private |=
				(PL_lex_repl ? OPpTRANS_FROM_UTF
					     : OPpTRANS_TO_UTF);
			}
#ifdef EBCDIC
			if (uv > 255 && !dorange)
			    native_range = FALSE;
#endif
                    }
		    else {
		        *d++ = (char)uv;
		    }
		}
		else {
		    *d++ = (char) uv;
		}
		continue;

 	    /* \N{LATIN SMALL LETTER A} is a named character */
 	    case 'N':
 		++s;
 		if (*s == '{') {
 		    char* e = strchr(s, '}');
 		    SV *res;
 		    STRLEN len;
 		    const char *str;
 		    SV *type;

 		    if (!e) {
			yyerror("Missing right brace on \\N{}");
			e = s - 1;
			goto cont_scan;
		    }
		    if (e > s + 2 && s[1] == 'U' && s[2] == '+') {
		        /* \N{U+...} */
		        I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
			  PERL_SCAN_DISALLOW_PREFIX;
		        s += 3;
			len = e - s;
			uv = grok_hex(s, &len, &flags, NULL);
			if ( e > s && len != (STRLEN)(e - s) ) {
			    uv = 0xFFFD;
			}
			s = e + 1;
			goto NUM_ESCAPE_INSERT;
		    }
		    res = newSVpvn(s + 1, e - s - 1);
		    type = newSVpvn(s - 2,e - s + 3);
		    res = new_constant( NULL, 0, "charnames",
					res, NULL, SvPVX(type) );
		    SvREFCNT_dec(type);		
		    if (has_utf8)
			sv_utf8_upgrade(res);
		    str = SvPV_const(res,len);
#ifdef EBCDIC_NEVER_MIND
		    /* charnames uses pack U and that has been
		     * recently changed to do the below uni->native
		     * mapping, so this would be redundant (and wrong,
		     * the code point would be doubly converted).
		     * But leave this in just in case the pack U change
		     * gets revoked, but the semantics is still
		     * desireable for charnames. --jhi */
		    {
			 UV uv = utf8_to_uvchr((const U8*)str, 0);

			 if (uv < 0x100) {
			      U8 tmpbuf[UTF8_MAXBYTES+1], *d;

			      d = uvchr_to_utf8(tmpbuf, UNI_TO_NATIVE(uv));
			      sv_setpvn(res, (char *)tmpbuf, d - tmpbuf);
			      str = SvPV_const(res, len);
			 }
		    }
#endif
		    if (!has_utf8 && SvUTF8(res)) {
			const char * const ostart = SvPVX_const(sv);
			SvCUR_set(sv, d - ostart);
			SvPOK_on(sv);
			*d = '\0';
			sv_utf8_upgrade(sv);
			/* this just broke our allocation above... */
			SvGROW(sv, (STRLEN)(send - start));
			d = SvPVX(sv) + SvCUR(sv);
			has_utf8 = TRUE;
		    }
		    if (len > (STRLEN)(e - s + 4)) { /* I _guess_ 4 is \N{} --jhi */
			const char * const odest = SvPVX_const(sv);

			SvGROW(sv, (SvLEN(sv) + len - (e - s + 4)));
			d = SvPVX(sv) + (d - odest);
		    }
#ifdef EBCDIC
		    if (!dorange)
			native_range = FALSE; /* \N{} is guessed to be Unicode */
#endif
		    Copy(str, d, len, char);
		    d += len;
		    SvREFCNT_dec(res);
		  cont_scan:
		    s = e + 1;
		}
		else
		    yyerror("Missing braces on \\N{}");
		continue;

	    /* \c is a control character */
	    case 'c':
		s++;
		if (s < send) {
		    U8 c = *s++;
#ifdef EBCDIC
		    if (isLOWER(c))
			c = toUPPER(c);
#endif
		    *d++ = NATIVE_TO_NEED(has_utf8,toCTRL(c));
		}
		else {
		    yyerror("Missing control char name in \\c");
		}
		continue;

	    /* printf-style backslashes, formfeeds, newlines, etc */
	    case 'b':
		*d++ = NATIVE_TO_NEED(has_utf8,'\b');
		break;
	    case 'n':
		*d++ = NATIVE_TO_NEED(has_utf8,'\n');
		break;
	    case 'r':
		*d++ = NATIVE_TO_NEED(has_utf8,'\r');
		break;
	    case 'f':
		*d++ = NATIVE_TO_NEED(has_utf8,'\f');
		break;
	    case 't':
		*d++ = NATIVE_TO_NEED(has_utf8,'\t');
		break;
	    case 'e':
		*d++ = ASCII_TO_NEED(has_utf8,'\033');
		break;
	    case 'a':
		*d++ = ASCII_TO_NEED(has_utf8,'\007');
		break;
	    } /* end switch */

	    s++;
	    continue;
	} /* end if (backslash) */
#ifdef EBCDIC
	else
	    literal_endpoint++;
#endif

    default_action:
	/* If we started with encoded form, or already know we want it
	   and then encode the next character */
	if ((has_utf8 || this_utf8) && !NATIVE_IS_INVARIANT((U8)(*s))) {
	    STRLEN len  = 1;
	    const UV nextuv   = (this_utf8) ? utf8n_to_uvchr((U8*)s, send - s, &len, 0) : (UV) ((U8) *s);
	    const STRLEN need = UNISKIP(NATIVE_TO_UNI(nextuv));
	    s += len;
	    if (need > len) {
		/* encoded value larger than old, need extra space (NOTE: SvCUR() not set here) */
		const STRLEN off = d - SvPVX_const(sv);
		d = SvGROW(sv, SvLEN(sv) + (need-len)) + off;
	    }
	    d = (char*)uvchr_to_utf8((U8*)d, nextuv);
	    has_utf8 = TRUE;
#ifdef EBCDIC
	    if (uv > 255 && !dorange)
		native_range = FALSE;
#endif
	}
	else {
	    *d++ = NATIVE_TO_NEED(has_utf8,*s++);
	}
    } /* while loop to process each character */

    /* terminate the string and set up the sv */
    *d = '\0';
    SvCUR_set(sv, d - SvPVX_const(sv));
    if (SvCUR(sv) >= SvLEN(sv))
	Perl_croak(aTHX_ "panic: constant overflowed allocated space");

    SvPOK_on(sv);
    if (PL_encoding && !has_utf8) {
	sv_recode_to_utf8(sv, PL_encoding);
	if (SvUTF8(sv))
	    has_utf8 = TRUE;
    }
    if (has_utf8) {
	SvUTF8_on(sv);
	if (PL_lex_inwhat == OP_TRANS && PL_sublex_info.sub_op) {
	    PL_sublex_info.sub_op->op_private |=
		    (PL_lex_repl ? OPpTRANS_FROM_UTF : OPpTRANS_TO_UTF);
	}
    }

    /* shrink the sv if we allocated more than we used */
    if (SvCUR(sv) + 5 < SvLEN(sv)) {
	SvPV_shrink_to_cur(sv);
    }

    /* return the substring (via yylval) only if we parsed anything */
    if (s > PL_bufptr) {
	if ( PL_hints & ( PL_lex_inpat ? HINT_NEW_RE : HINT_NEW_STRING ) )
	    sv = new_constant(start, s - start,
			      (const char *)(PL_lex_inpat ? "qr" : "q"),
			      sv, NULL,
			      (const char *)
			      (( PL_lex_inwhat == OP_TRANS
				 ? "tr"
				 : ( (PL_lex_inwhat == OP_SUBST && !PL_lex_inpat)
				     ? "s"
				     : "qq"))));
	yylval.opval = (OP*)newSVOP(OP_CONST, 0, sv);
    } else
	SvREFCNT_dec(sv);
    return s;
}

/* S_intuit_more
 * Returns TRUE if there's more to the expression (e.g., a subscript),
 * FALSE otherwise.
 *
 * It deals with "$foo[3]" and /$foo[3]/ and /$foo[0123456789$]+/
 *
 * ->[ and ->{ return TRUE
 * { and [ outside a pattern are always subscripts, so return TRUE
 * if we're outside a pattern and it's not { or [, then return FALSE
 * if we're in a pattern and the first char is a {
 *   {4,5} (any digits around the comma) returns FALSE
 * if we're in a pattern and the first char is a [
 *   [] returns FALSE
 *   [SOMETHING] has a funky algorithm to decide whether it's a
 *      character class or not.  It has to deal with things like
 *      /$foo[-3]/ and /$foo[$bar]/ as well as /$foo[$\d]+/
 * anything else returns TRUE
 */

/* This is the one truly awful dwimmer necessary to conflate C and sed. */

STATIC int
S_intuit_more(pTHX_ register char *s)
{
    dVAR;
    if (PL_lex_brackets)
	return TRUE;
    if (*s == '-' && s[1] == '>' && (s[2] == '[' || s[2] == '{'))
	return TRUE;
    if (*s != '{' && *s != '[')
	return FALSE;
    if (!PL_lex_inpat)
	return TRUE;

    /* In a pattern, so maybe we have {n,m}. */
    if (*s == '{') {
	s++;
	if (!isDIGIT(*s))
	    return TRUE;
	while (isDIGIT(*s))
	    s++;
	if (*s == ',')
	    s++;
	while (isDIGIT(*s))
	    s++;
	if (*s == '}')
	    return FALSE;
	return TRUE;
	
    }

    /* On the other hand, maybe we have a character class */

    s++;
    if (*s == ']' || *s == '^')
	return FALSE;
    else {
        /* this is terrifying, and it works */
	int weight = 2;		/* let's weigh the evidence */
	char seen[256];
	unsigned char un_char = 255, last_un_char;
	const char * const send = strchr(s,']');
	char tmpbuf[sizeof PL_tokenbuf * 4];

	if (!send)		/* has to be an expression */
	    return TRUE;

	Zero(seen,256,char);
	if (*s == '$')
	    weight -= 3;
	else if (isDIGIT(*s)) {
	    if (s[1] != ']') {
		if (isDIGIT(s[1]) && s[2] == ']')
		    weight -= 10;
	    }
	    else
		weight -= 100;
	}
	for (; s < send; s++) {
	    last_un_char = un_char;
	    un_char = (unsigned char)*s;
	    switch (*s) {
	    case '@':
	    case '&':
	    case '$':
		weight -= seen[un_char] * 10;
		if (isALNUM_lazy_if(s+1,UTF)) {
		    int len;
		    scan_ident(s, send, tmpbuf, sizeof tmpbuf, FALSE);
		    len = (int)strlen(tmpbuf);
		    if (len > 1 && gv_fetchpvn_flags(tmpbuf, len, 0, SVt_PV))
			weight -= 100;
		    else
			weight -= 10;
		}
		else if (*s == '$' && s[1] &&
		  strchr("[#!%*<>()-=",s[1])) {
		    if (/*{*/ strchr("])} =",s[2]))
			weight -= 10;
		    else
			weight -= 1;
		}
		break;
	    case '\\':
		un_char = 254;
		if (s[1]) {
		    if (strchr("wds]",s[1]))
			weight += 100;
		    else if (seen[(U8)'\''] || seen[(U8)'"'])
			weight += 1;
		    else if (strchr("rnftbxcav",s[1]))
			weight += 40;
		    else if (isDIGIT(s[1])) {
			weight += 40;
			while (s[1] && isDIGIT(s[1]))
			    s++;
		    }
		}
		else
		    weight += 100;
		break;
	    case '-':
		if (s[1] == '\\')
		    weight += 50;
		if (strchr("aA01! ",last_un_char))
		    weight += 30;
		if (strchr("zZ79~",s[1]))
		    weight += 30;
		if (last_un_char == 255 && (isDIGIT(s[1]) || s[1] == '$'))
		    weight -= 5;	/* cope with negative subscript */
		break;
	    default:
		if (!isALNUM(last_un_char)
		    && !(last_un_char == '$' || last_un_char == '@'
			 || last_un_char == '&')
		    && isALPHA(*s) && s[1] && isALPHA(s[1])) {
		    char *d = tmpbuf;
		    while (isALPHA(*s))
			*d++ = *s++;
		    *d = '\0';
		    if (keyword(tmpbuf, d - tmpbuf, 0))
			weight -= 150;
		}
		if (un_char == last_un_char + 1)
		    weight += 5;
		weight -= seen[un_char];
		break;
	    }
	    seen[un_char]++;
	}
	if (weight >= 0)	/* probably a character class */
	    return FALSE;
    }

    return TRUE;
}

/*
 * S_intuit_method
 *
 * Does all the checking to disambiguate
 *   foo bar
 * between foo(bar) and bar->foo.  Returns 0 if not a method, otherwise
 * FUNCMETH (bar->foo(args)) or METHOD (bar->foo args).
 *
 * First argument is the stuff after the first token, e.g. "bar".
 *
 * Not a method if bar is a filehandle.
 * Not a method if foo is a subroutine prototyped to take a filehandle.
 * Not a method if it's really "Foo $bar"
 * Method if it's "foo $bar"
 * Not a method if it's really "print foo $bar"
 * Method if it's really "foo package::" (interpreted as package->foo)
 * Not a method if bar is known to be a subroutine ("sub bar; foo bar")
 * Not a method if bar is a filehandle or package, but is quoted with
 *   =>
 */

STATIC int
S_intuit_method(pTHX_ char *start, GV *gv, CV *cv)
{
    dVAR;
    char *s = start + (*start == '$');
    char tmpbuf[sizeof PL_tokenbuf];
    STRLEN len;
    GV* indirgv;
#ifdef PERL_MAD
    int soff;
#endif

    if (gv) {
	if (SvTYPE(gv) == SVt_PVGV && GvIO(gv))
	    return 0;
	if (cv) {
	    if (SvPOK(cv)) {
		const char *proto = SvPVX_const(cv);
		if (proto) {
		    if (*proto == ';')
			proto++;
		    if (*proto == '*')
			return 0;
		}
	    }
	} else
	    gv = NULL;
    }
    s = scan_word(s, tmpbuf, sizeof tmpbuf, TRUE, &len);
    /* start is the beginning of the possible filehandle/object,
     * and s is the end of it
     * tmpbuf is a copy of it
     */

    if (*start == '$') {
	if (gv || PL_last_lop_op == OP_PRINT || isUPPER(*PL_tokenbuf))
	    return 0;
#ifdef PERL_MAD
	len = start - SvPVX(PL_linestr);
#endif
	s = PEEKSPACE(s);
#ifdef PERL_MAD
	start = SvPVX(PL_linestr) + len;
#endif
	PL_bufptr = start;
	PL_expect = XREF;
	return *s == '(' ? FUNCMETH : METHOD;
    }
    if (!keyword(tmpbuf, len, 0)) {
	if (len > 2 && tmpbuf[len - 2] == ':' && tmpbuf[len - 1] == ':') {
	    len -= 2;
	    tmpbuf[len] = '\0';
#ifdef PERL_MAD
	    soff = s - SvPVX(PL_linestr);
#endif
	    goto bare_package;
	}
	indirgv = gv_fetchpvn_flags(tmpbuf, len, 0, SVt_PVCV);
	if (indirgv && GvCVu(indirgv))
	    return 0;
	/* filehandle or package name makes it a method */
	if (!gv || GvIO(indirgv) || gv_stashpvn(tmpbuf, len, FALSE)) {
#ifdef PERL_MAD
	    soff = s - SvPVX(PL_linestr);
#endif
	    s = PEEKSPACE(s);
	    if ((PL_bufend - s) >= 2 && *s == '=' && *(s+1) == '>')
		return 0;	/* no assumptions -- "=>" quotes bearword */
      bare_package:
	    start_force(PL_curforce);
	    NEXTVAL_NEXTTOKE.opval = (OP*)newSVOP(OP_CONST, 0,
						   newSVpvn(tmpbuf,len));
	    NEXTVAL_NEXTTOKE.opval->op_private = OPpCONST_BARE;
	    if (PL_madskills)
		curmad('X', newSVpvn(start,SvPVX(PL_linestr) + soff - start));
	    PL_expect = XTERM;
	    force_next(WORD);
	    PL_bufptr = s;
#ifdef PERL_MAD
	    PL_bufptr = SvPVX(PL_linestr) + soff; /* restart before space */
#endif
	    return *s == '(' ? FUNCMETH : METHOD;
	}
    }
    return 0;
}

/*
 * S_incl_perldb
 * Return a string of Perl code to load the debugger.  If PERL5DB
 * is set, it will return the contents of that, otherwise a
 * compile-time require of perl5db.pl.
 */

STATIC const char*
S_incl_perldb(pTHX)
{
    dVAR;
    if (PL_perldb) {
	const char * const pdb = PerlEnv_getenv("PERL5DB");

	if (pdb)
	    return pdb;
	SETERRNO(0,SS_NORMAL);
	return "BEGIN { require 'perl5db.pl' }";
    }
    return "";
}


/* Encoded script support. filter_add() effectively inserts a
 * 'pre-processing' function into the current source input stream.
 * Note that the filter function only applies to the current source file
 * (e.g., it will not affect files 'require'd or 'use'd by this one).
 *
 * The datasv parameter (which may be NULL) can be used to pass
 * private data to this instance of the filter. The filter function
 * can recover the SV using the FILTER_DATA macro and use it to
 * store private buffers and state information.
 *
 * The supplied datasv parameter is upgraded to a PVIO type
 * and the IoDIRP/IoANY field is used to store the function pointer,
 * and IOf_FAKE_DIRP is enabled on datasv to mark this as such.
 * Note that IoTOP_NAME, IoFMT_NAME, IoBOTTOM_NAME, if set for
 * private use must be set using malloc'd pointers.
 */

SV *
Perl_filter_add(pTHX_ filter_t funcp, SV *datasv)
{
    dVAR;
    if (!funcp)
	return NULL;

    if (!PL_rsfp_filters)
	PL_rsfp_filters = newAV();
    if (!datasv)
	datasv = newSV(0);
    SvUPGRADE(datasv, SVt_PVIO);
    IoANY(datasv) = FPTR2DPTR(void *, funcp); /* stash funcp into spare field */
    IoFLAGS(datasv) |= IOf_FAKE_DIRP;
    DEBUG_P(PerlIO_printf(Perl_debug_log, "filter_add func %p (%s)\n",
			  FPTR2DPTR(void *, IoANY(datasv)),
			  SvPV_nolen(datasv)));
    av_unshift(PL_rsfp_filters, 1);
    av_store(PL_rsfp_filters, 0, datasv) ;
    return(datasv);
}


/* Delete most recently added instance of this filter function.	*/
void
Perl_filter_del(pTHX_ filter_t funcp)
{
    dVAR;
    SV *datasv;

#ifdef DEBUGGING
    DEBUG_P(PerlIO_printf(Perl_debug_log, "filter_del func %p",
			  FPTR2DPTR(void*, funcp)));
#endif
    if (!PL_rsfp_filters || AvFILLp(PL_rsfp_filters)<0)
	return;
    /* if filter is on top of stack (usual case) just pop it off */
    datasv = FILTER_DATA(AvFILLp(PL_rsfp_filters));
    if (IoANY(datasv) == FPTR2DPTR(void *, funcp)) {
	IoFLAGS(datasv) &= ~IOf_FAKE_DIRP;
	IoANY(datasv) = (void *)NULL;
	sv_free(av_pop(PL_rsfp_filters));

        return;
    }
    /* we need to search for the correct entry and clear it	*/
    Perl_die(aTHX_ "filter_del can only delete in reverse order (currently)");
}


/* Invoke the idxth filter function for the current rsfp.	 */
/* maxlen 0 = read one text line */
I32
Perl_filter_read(pTHX_ int idx, SV *buf_sv, int maxlen)
{
    dVAR;
    filter_t funcp;
    SV *datasv = NULL;
    /* This API is bad. It should have been using unsigned int for maxlen.
       Not sure if we want to change the API, but if not we should sanity
       check the value here.  */
    const unsigned int correct_length
	= maxlen < 0 ?
#ifdef PERL_MICRO
	0x7FFFFFFF
#else
	INT_MAX
#endif
	: maxlen;

    if (!PL_rsfp_filters)
	return -1;
    if (idx > AvFILLp(PL_rsfp_filters)) {       /* Any more filters?	*/
	/* Provide a default input filter to make life easy.	*/
	/* Note that we append to the line. This is handy.	*/
	DEBUG_P(PerlIO_printf(Perl_debug_log,
			      "filter_read %d: from rsfp\n", idx));
	if (correct_length) {
 	    /* Want a block */
	    int len ;
	    const int old_len = SvCUR(buf_sv);

	    /* ensure buf_sv is large enough */
	    SvGROW(buf_sv, (STRLEN)(old_len + correct_length)) ;
	    if ((len = PerlIO_read(PL_rsfp, SvPVX(buf_sv) + old_len,
				   correct_length)) <= 0) {
		if (PerlIO_error(PL_rsfp))
	            return -1;		/* error */
	        else
		    return 0 ;		/* end of file */
	    }
	    SvCUR_set(buf_sv, old_len + len) ;
	} else {
	    /* Want a line */
            if (sv_gets(buf_sv, PL_rsfp, SvCUR(buf_sv)) == NULL) {
		if (PerlIO_error(PL_rsfp))
	            return -1;		/* error */
	        else
		    return 0 ;		/* end of file */
	    }
	}
	return SvCUR(buf_sv);
    }
    /* Skip this filter slot if filter has been deleted	*/
    if ( (datasv = FILTER_DATA(idx)) == &PL_sv_undef) {
	DEBUG_P(PerlIO_printf(Perl_debug_log,
			      "filter_read %d: skipped (filter deleted)\n",
			      idx));
	return FILTER_READ(idx+1, buf_sv, correct_length); /* recurse */
    }
    /* Get function pointer hidden within datasv	*/
    funcp = DPTR2FPTR(filter_t, IoANY(datasv));
    DEBUG_P(PerlIO_printf(Perl_debug_log,
			  "filter_read %d: via function %p (%s)\n",
			  idx, (void*)datasv, SvPV_nolen_const(datasv)));
    /* Call function. The function is expected to 	*/
    /* call "FILTER_READ(idx+1, buf_sv)" first.		*/
    /* Return: <0:error, =0:eof, >0:not eof 		*/
    return (*funcp)(aTHX_ idx, buf_sv, correct_length);
}

STATIC char *
S_filter_gets(pTHX_ register SV *sv, register PerlIO *fp, STRLEN append)
{
    dVAR;
#ifdef PERL_CR_FILTER
    if (!PL_rsfp_filters) {
	filter_add(S_cr_textfilter,NULL);
    }
#endif
    if (PL_rsfp_filters) {
	if (!append)
            SvCUR_set(sv, 0);	/* start with empty line	*/
        if (FILTER_READ(0, sv, 0) > 0)
            return ( SvPVX(sv) ) ;
        else
	    return NULL ;
    }
    else
        return (sv_gets(sv, fp, append));
}

STATIC HV *
S_find_in_my_stash(pTHX_ const char *pkgname, I32 len)
{
    dVAR;
    GV *gv;

    if (len == 11 && *pkgname == '_' && strEQ(pkgname, "__PACKAGE__"))
        return PL_curstash;

    if (len > 2 &&
        (pkgname[len - 2] == ':' && pkgname[len - 1] == ':') &&
        (gv = gv_fetchpvn_flags(pkgname, len, 0, SVt_PVHV)))
    {
        return GvHV(gv);			/* Foo:: */
    }

    /* use constant CLASS => 'MyClass' */
    gv = gv_fetchpvn_flags(pkgname, len, 0, SVt_PVCV);
    if (gv && GvCV(gv)) {
	SV * const sv = cv_const_sv(GvCV(gv));
	if (sv)
            pkgname = SvPV_nolen_const(sv);
    }

    return gv_stashpv(pkgname, FALSE);
}

/*
 * S_readpipe_override
 * Check whether readpipe() is overriden, and generates the appropriate
 * optree, provided sublex_start() is called afterwards.
 */
STATIC void
S_readpipe_override(pTHX)
{
    GV **gvp;
    GV *gv_readpipe = gv_fetchpvs("readpipe", GV_NOTQUAL, SVt_PVCV);
    yylval.ival = OP_BACKTICK;
    if ((gv_readpipe
		&& GvCVu(gv_readpipe) && GvIMPORTED_CV(gv_readpipe))
	    ||
	    ((gvp = (GV**)hv_fetchs(PL_globalstash, "readpipe", FALSE))
	     && (gv_readpipe = *gvp) != (GV*)&PL_sv_undef
	     && GvCVu(gv_readpipe) && GvIMPORTED_CV(gv_readpipe)))
    {
	PL_lex_op = (OP*)newUNOP(OP_ENTERSUB, OPf_STACKED,
	    append_elem(OP_LIST,
		newSVOP(OP_CONST, 0, &PL_sv_undef), /* value will be read later */
		newCVREF(0, newGVOP(OP_GV, 0, gv_readpipe))));
    }
    else {
	set_csh();
    }
}

#ifdef PERL_MAD 
 /*
 * Perl_madlex
 * The intent of this yylex wrapper is to minimize the changes to the
 * tokener when we aren't interested in collecting madprops.  It remains
 * to be seen how successful this strategy will be...
 */

int
Perl_madlex(pTHX)
{
    int optype;
    char *s = PL_bufptr;

    /* make sure PL_thiswhite is initialized */
    PL_thiswhite = 0;
    PL_thismad = 0;

    /* just do what yylex would do on pending identifier; leave PL_thiswhite alone */
    if (PL_pending_ident)
        return S_pending_ident(aTHX);

    /* previous token ate up our whitespace? */
    if (!PL_lasttoke && PL_nextwhite) {
	PL_thiswhite = PL_nextwhite;
	PL_nextwhite = 0;
    }

    /* isolate the token, and figure out where it is without whitespace */
    PL_realtokenstart = -1;
    PL_thistoken = 0;
    optype = yylex();
    s = PL_bufptr;
    assert(PL_curforce < 0);

    if (!PL_thismad || PL_thismad->mad_key == '^') {	/* not forced already? */
	if (!PL_thistoken) {
	    if (PL_realtokenstart < 0 || !CopLINE(PL_curcop))
		PL_thistoken = newSVpvs("");
	    else {
		char * const tstart = SvPVX(PL_linestr) + PL_realtokenstart;
		PL_thistoken = newSVpvn(tstart, s - tstart);
	    }
	}
	if (PL_thismad)	/* install head */
	    CURMAD('X', PL_thistoken);
    }

    /* last whitespace of a sublex? */
    if (optype == ')' && PL_endwhite) {
	CURMAD('X', PL_endwhite);
    }

    if (!PL_thismad) {

	/* if no whitespace and we're at EOF, bail.  Otherwise fake EOF below. */
	if (!PL_thiswhite && !PL_endwhite && !optype) {
	    sv_free(PL_thistoken);
	    PL_thistoken = 0;
	    return 0;
	}

	/* put off final whitespace till peg */
	if (optype == ';' && !PL_rsfp) {
	    PL_nextwhite = PL_thiswhite;
	    PL_thiswhite = 0;
	}
	else if (PL_thisopen) {
	    CURMAD('q', PL_thisopen);
	    if (PL_thistoken)
		sv_free(PL_thistoken);
	    PL_thistoken = 0;
	}
	else {
	    /* Store actual token text as madprop X */
	    CURMAD('X', PL_thistoken);
	}

	if (PL_thiswhite) {
	    /* add preceding whitespace as madprop _ */
	    CURMAD('_', PL_thiswhite);
	}

	if (PL_thisstuff) {
	    /* add quoted material as madprop = */
	    CURMAD('=', PL_thisstuff);
	}

	if (PL_thisclose) {
	    /* add terminating quote as madprop Q */
	    CURMAD('Q', PL_thisclose);
	}
    }

    /* special processing based on optype */

    switch (optype) {

    /* opval doesn't need a TOKEN since it can already store mp */
    case WORD:
    case METHOD:
    case FUNCMETH:
    case THING:
    case PMFUNC:
    case PRIVATEREF:
    case FUNC0SUB:
    case UNIOPSUB:
    case LSTOPSUB:
	if (yylval.opval)
	    append_madprops(PL_thismad, yylval.opval, 0);
	PL_thismad = 0;
	return optype;

    /* fake EOF */
    case 0:
	optype = PEG;
	if (PL_endwhite) {
	    addmad(newMADsv('p', PL_endwhite), &PL_thismad, 0);
	    PL_endwhite = 0;
	}
	break;

    case ']':
    case '}':
	if (PL_faketokens)
	    break;
	/* remember any fake bracket that lexer is about to discard */ 
	if (PL_lex_brackets == 1 &&
	    ((expectation)PL_lex_brackstack[0] & XFAKEBRACK))
	{
	    s = PL_bufptr;
	    while (s < PL_bufend && (*s == ' ' || *s == '\t'))
		s++;
	    if (*s == '}') {
		PL_thiswhite = newSVpvn(PL_bufptr, ++s - PL_bufptr);
		addmad(newMADsv('#', PL_thiswhite), &PL_thismad, 0);
		PL_thiswhite = 0;
		PL_bufptr = s - 1;
		break;	/* don't bother looking for trailing comment */
	    }
	    else
		s = PL_bufptr;
	}
	if (optype == ']')
	    break;
	/* FALLTHROUGH */

    /* attach a trailing comment to its statement instead of next token */
    case ';':
	if (PL_faketokens)
	    break;
	if (PL_bufptr > PL_oldbufptr && PL_bufptr[-1] == optype) {
	    s = PL_bufptr;
	    while (s < PL_bufend && (*s == ' ' || *s == '\t'))
		s++;
	    if (*s == '\n' || *s == '#') {
		while (s < PL_bufend && *s != '\n')
		    s++;
		if (s < PL_bufend)
		    s++;
		PL_thiswhite = newSVpvn(PL_bufptr, s - PL_bufptr);
		addmad(newMADsv('#', PL_thiswhite), &PL_thismad, 0);
		PL_thiswhite = 0;
		PL_bufptr = s;
	    }
	}
	break;

    /* pval */
    case LABEL:
	break;

    /* ival */
    default:
	break;

    }

    /* Create new token struct.  Note: opvals return early above. */
    yylval.tkval = newTOKEN(optype, yylval, PL_thismad);
    PL_thismad = 0;
    return optype;
}
#endif

STATIC char *
S_tokenize_use(pTHX_ int is_use, char *s) {
    dVAR;
    if (PL_expect != XSTATE)
	yyerror(Perl_form(aTHX_ "\"%s\" not allowed in expression",
		    is_use ? "use" : "no"));
    s = SKIPSPACE1(s);
    if (isDIGIT(*s) || (*s == 'v' && isDIGIT(s[1]))) {
	s = force_version(s, TRUE);
	if (*s == ';' || (s = SKIPSPACE1(s), *s == ';')) {
	    start_force(PL_curforce);
	    NEXTVAL_NEXTTOKE.opval = NULL;
	    force_next(WORD);
	}
	else if (*s == 'v') {
	    s = force_word(s,WORD,FALSE,TRUE,FALSE);
	    s = force_version(s, FALSE);
	}
    }
    else {
	s = force_word(s,WORD,FALSE,TRUE,FALSE);
	s = force_version(s, FALSE);
    }
    yylval.ival = is_use;
    return s;
}
#ifdef DEBUGGING
    static const char* const exp_name[] =
	{ "OPERATOR", "TERM", "REF", "STATE", "BLOCK", "ATTRBLOCK",
	  "ATTRTERM", "TERMBLOCK", "TERMORDORDOR"
	};
#endif

/*
  yylex

  Works out what to call the token just pulled out of the input
  stream.  The yacc parser takes care of taking the ops we return and
  stitching them into a tree.

  Returns:
    PRIVATEREF

  Structure:
      if read an identifier
          if we're in a my declaration
	      croak if they tried to say my($foo::bar)
	      build the ops for a my() declaration
	  if it's an access to a my() variable
	      are we in a sort block?
	          croak if my($a); $a <=> $b
	      build ops for access to a my() variable
	  if in a dq string, and they've said @foo and we can't find @foo
	      croak
	  build ops for a bareword
      if we already built the token before, use it.
*/


#ifdef __SC__
#pragma segment Perl_yylex
#endif
int
Perl_yylex(pTHX)
{
    dVAR;
    register char *s = PL_bufptr;
    register char *d;
    STRLEN len;
    bool bof = FALSE;

    /* orig_keyword, gvp, and gv are initialized here because
     * jump to the label just_a_word_zero can bypass their
     * initialization later. */
    I32 orig_keyword = 0;
    GV *gv = NULL;
    GV **gvp = NULL;

    DEBUG_T( {
	SV* tmp = newSVpvs("");
	PerlIO_printf(Perl_debug_log, "### %"IVdf":LEX_%s/X%s %s\n",
	    (IV)CopLINE(PL_curcop),
	    lex_state_names[PL_lex_state],
	    exp_name[PL_expect],
	    pv_display(tmp, s, strlen(s), 0, 60));
	SvREFCNT_dec(tmp);
    } );
    /* check if there's an identifier for us to look at */
    if (PL_pending_ident)
        return REPORT(S_pending_ident(aTHX));

    /* no identifier pending identification */

    switch (PL_lex_state) {
#ifdef COMMENTARY
    case LEX_NORMAL:		/* Some compilers will produce faster */
    case LEX_INTERPNORMAL:	/* code if we comment these out. */
	break;
#endif

    /* when we've already built the next token, just pull it out of the queue */
    case LEX_KNOWNEXT:
#ifdef PERL_MAD
	PL_lasttoke--;
	yylval = PL_nexttoke[PL_lasttoke].next_val;
	if (PL_madskills) {
	    PL_thismad = PL_nexttoke[PL_lasttoke].next_mad;
	    PL_nexttoke[PL_lasttoke].next_mad = 0;
	    if (PL_thismad && PL_thismad->mad_key == '_') {
		PL_thiswhite = (SV*)PL_thismad->mad_val;
		PL_thismad->mad_val = 0;
		mad_free(PL_thismad);
		PL_thismad = 0;
	    }
	}
	if (!PL_lasttoke) {
	    PL_lex_state = PL_lex_defer;
  	    PL_expect = PL_lex_expect;
  	    PL_lex_defer = LEX_NORMAL;
	    if (!PL_nexttoke[PL_lasttoke].next_type)
		return yylex();
  	}
#else
	PL_nexttoke--;
	yylval = PL_nextval[PL_nexttoke];
	if (!PL_nexttoke) {
	    PL_lex_state = PL_lex_defer;
	    PL_expect = PL_lex_expect;
	    PL_lex_defer = LEX_NORMAL;
	}
#endif
#ifdef PERL_MAD
	/* FIXME - can these be merged?  */
	return(PL_nexttoke[PL_lasttoke].next_type);
#else
	return REPORT(PL_nexttype[PL_nexttoke]);
#endif

    /* interpolated case modifiers like \L \U, including \Q and \E.
       when we get here, PL_bufptr is at the \
    */
    case LEX_INTERPCASEMOD:
#ifdef DEBUGGING
	if (PL_bufptr != PL_bufend && *PL_bufptr != '\\')
	    Perl_croak(aTHX_ "panic: INTERPCASEMOD");
#endif
	/* handle \E or end of string */
       	if (PL_bufptr == PL_bufend || PL_bufptr[1] == 'E') {
	    /* if at a \E */
	    if (PL_lex_casemods) {
		const char oldmod = PL_lex_casestack[--PL_lex_casemods];
		PL_lex_casestack[PL_lex_casemods] = '\0';

		if (PL_bufptr != PL_bufend
		    && (oldmod == 'L' || oldmod == 'U' || oldmod == 'Q')) {
		    PL_bufptr += 2;
		    PL_lex_state = LEX_INTERPCONCAT;
#ifdef PERL_MAD
		    if (PL_madskills)
			PL_thistoken = newSVpvs("\\E");
#endif
		}
		return REPORT(')');
	    }
#ifdef PERL_MAD
	    while (PL_bufptr != PL_bufend &&
	      PL_bufptr[0] == '\\' && PL_bufptr[1] == 'E') {
		if (!PL_thiswhite)
		    PL_thiswhite = newSVpvs("");
		sv_catpvn(PL_thiswhite, PL_bufptr, 2);
		PL_bufptr += 2;
	    }
#else
	    if (PL_bufptr != PL_bufend)
		PL_bufptr += 2;
#endif
	    PL_lex_state = LEX_INTERPCONCAT;
	    return yylex();
	}
	else {
	    DEBUG_T({ PerlIO_printf(Perl_debug_log,
              "### Saw case modifier\n"); });
	    s = PL_bufptr + 1;
	    if (s[1] == '\\' && s[2] == 'E') {
#ifdef PERL_MAD
		if (!PL_thiswhite)
		    PL_thiswhite = newSVpvs("");
		sv_catpvn(PL_thiswhite, PL_bufptr, 4);
#endif
	        PL_bufptr = s + 3;
		PL_lex_state = LEX_INTERPCONCAT;
		return yylex();
	    }
	    else {
		I32 tmp;
		if (!PL_madskills) /* when just compiling don't need correct */
		    if (strnEQ(s, "L\\u", 3) || strnEQ(s, "U\\l", 3))
			tmp = *s, *s = s[2], s[2] = (char)tmp;	/* misordered... */
		if ((*s == 'L' || *s == 'U') &&
		    (strchr(PL_lex_casestack, 'L') || strchr(PL_lex_casestack, 'U'))) {
		    PL_lex_casestack[--PL_lex_casemods] = '\0';
		    return REPORT(')');
		}
		if (PL_lex_casemods > 10)
		    Renew(PL_lex_casestack, PL_lex_casemods + 2, char);
		PL_lex_casestack[PL_lex_casemods++] = *s;
		PL_lex_casestack[PL_lex_casemods] = '\0';
		PL_lex_state = LEX_INTERPCONCAT;
		start_force(PL_curforce);
		NEXTVAL_NEXTTOKE.ival = 0;
		force_next('(');
		start_force(PL_curforce);
		if (*s == 'l')
		    NEXTVAL_NEXTTOKE.ival = OP_LCFIRST;
		else if (*s == 'u')
		    NEXTVAL_NEXTTOKE.ival = OP_UCFIRST;
		else if (*s == 'L')
		    NEXTVAL_NEXTTOKE.ival = OP_LC;
		else if (*s == 'U')
		    NEXTVAL_NEXTTOKE.ival = OP_UC;
		else if (*s == 'Q')
		    NEXTVAL_NEXTTOKE.ival = OP_QUOTEMETA;
		else
		    Perl_croak(aTHX_ "panic: yylex");
		if (PL_madskills) {
		    SV* const tmpsv = newSVpvs("");
		    Perl_sv_catpvf(aTHX_ tmpsv, "\\%c", *s);
		    curmad('_', tmpsv);
		}
		PL_bufptr = s + 1;
	    }
	    force_next(FUNC);
	    if (PL_lex_starts) {
		s = PL_bufptr;
		PL_lex_starts = 0;
#ifdef PERL_MAD
		if (PL_madskills) {
		    if (PL_thistoken)
			sv_free(PL_thistoken);
		    PL_thistoken = newSVpvs("");
		}
#endif
		/* commas only at base level: /$a\Ub$c/ => ($a,uc(b.$c)) */
		if (PL_lex_casemods == 1 && PL_lex_inpat)
		    OPERATOR(',');
		else
		    Aop(OP_CONCAT);
	    }
	    else
		return yylex();
	}

    case LEX_INTERPPUSH:
        return REPORT(sublex_push());

    case LEX_INTERPSTART:
	if (PL_bufptr == PL_bufend)
	    return REPORT(sublex_done());
	DEBUG_T({ PerlIO_printf(Perl_debug_log,
              "### Interpolated variable\n"); });
	PL_expect = XTERM;
	PL_lex_dojoin = (*PL_bufptr == '@');
	PL_lex_state = LEX_INTERPNORMAL;
	if (PL_lex_dojoin) {
	    start_force(PL_curforce);
	    NEXTVAL_NEXTTOKE.ival = 0;
	    force_next(',');
	    start_force(PL_curforce);
	    force_ident("\"", '$');
	    start_force(PL_curforce);
	    NEXTVAL_NEXTTOKE.ival = 0;
	    force_next('$');
	    start_force(PL_curforce);
	    NEXTVAL_NEXTTOKE.ival = 0;
	    force_next('(');
	    start_force(PL_curforce);
	    NEXTVAL_NEXTTOKE.ival = OP_JOIN;	/* emulate join($", ...) */
	    force_next(FUNC);
	}
	if (PL_lex_starts++) {
	    s = PL_bufptr;
#ifdef PERL_MAD
	    if (PL_madskills) {
		if (PL_thistoken)
		    sv_free(PL_thistoken);
		PL_thistoken = newSVpvs("");
	    }
#endif
	    /* commas only at base level: /$a\Ub$c/ => ($a,uc(b.$c)) */
	    if (!PL_lex_casemods && PL_lex_inpat)
		OPERATOR(',');
	    else
		Aop(OP_CONCAT);
	}
	return yylex();

    case LEX_INTERPENDMAYBE:
	if (intuit_more(PL_bufptr)) {
	    PL_lex_state = LEX_INTERPNORMAL;	/* false alarm, more expr */
	    break;
	}
	/* FALL THROUGH */

    case LEX_INTERPEND:
	if (PL_lex_dojoin) {
	    PL_lex_dojoin = FALSE;
	    PL_lex_state = LEX_INTERPCONCAT;
#ifdef PERL_MAD
	    if (PL_madskills) {
		if (PL_thistoken)
		    sv_free(PL_thistoken);
		PL_thistoken = newSVpvs("");
	    }
#endif
	    return REPORT(')');
	}
	if (PL_lex_inwhat == OP_SUBST && PL_linestr == PL_lex_repl
	    && SvEVALED(PL_lex_repl))
	{
	    if (PL_bufptr != PL_bufend)
		Perl_croak(aTHX_ "Bad evalled substitution pattern");
	    PL_lex_repl = NULL;
	}
	/* FALLTHROUGH */
    case LEX_INTERPCONCAT:
#ifdef DEBUGGING
	if (PL_lex_brackets)
	    Perl_croak(aTHX_ "panic: INTERPCONCAT");
#endif
	if (PL_bufptr == PL_bufend)
	    return REPORT(sublex_done());

	if (SvIVX(PL_linestr) == '\'') {
	    SV *sv = newSVsv(PL_linestr);
	    if (!PL_lex_inpat)
		sv = tokeq(sv);
	    else if ( PL_hints & HINT_NEW_RE )
		sv = new_constant(NULL, 0, "qr", sv, sv, "q");
	    yylval.opval = (OP*)newSVOP(OP_CONST, 0, sv);
	    s = PL_bufend;
	}
	else {
	    s = scan_const(PL_bufptr);
	    if (*s == '\\')
		PL_lex_state = LEX_INTERPCASEMOD;
	    else
		PL_lex_state = LEX_INTERPSTART;
	}

	if (s != PL_bufptr) {
	    start_force(PL_curforce);
	    if (PL_madskills) {
		curmad('X', newSVpvn(PL_bufptr,s-PL_bufptr));
	    }
	    NEXTVAL_NEXTTOKE = yylval;
	    PL_expect = XTERM;
	    force_next(THING);
	    if (PL_lex_starts++) {
#ifdef PERL_MAD
		if (PL_madskills) {
		    if (PL_thistoken)
			sv_free(PL_thistoken);
		    PL_thistoken = newSVpvs("");
		}
#endif
		/* commas only at base level: /$a\Ub$c/ => ($a,uc(b.$c)) */
		if (!PL_lex_casemods && PL_lex_inpat)
		    OPERATOR(',');
		else
		    Aop(OP_CONCAT);
	    }
	    else {
		PL_bufptr = s;
		return yylex();
	    }
	}

	return yylex();
    case LEX_FORMLINE:
	PL_lex_state = LEX_NORMAL;
	s = scan_formline(PL_bufptr);
	if (!PL_lex_formbrack)
	    goto rightbracket;
	OPERATOR(';');
    }

    s = PL_bufptr;
    PL_oldoldbufptr = PL_oldbufptr;
    PL_oldbufptr = s;

  retry:
#ifdef PERL_MAD
    if (PL_thistoken) {
	sv_free(PL_thistoken);
	PL_thistoken = 0;
    }
    PL_realtokenstart = s - SvPVX(PL_linestr);	/* assume but undo on ws */
#endif
    switch (*s) {
    default:
	if (isIDFIRST_lazy_if(s,UTF))
	    goto keylookup;
	Perl_croak(aTHX_ "Unrecognized character \\x%02X", *s & 255);
    case 4:
    case 26:
	goto fake_eof;			/* emulate EOF on ^D or ^Z */
    case 0:
#ifdef PERL_MAD
	if (PL_madskills)
	    PL_faketokens = 0;
#endif
	if (!PL_rsfp) {
	    PL_last_uni = 0;
	    PL_last_lop = 0;
	    if (PL_lex_brackets) {
		yyerror((const char *)
			(PL_lex_formbrack
			 ? "Format not terminated"
			 : "Missing right curly or square bracket"));
	    }