Fixed a couple of signed-unsigned comparisons.

[mmh] / sbr / fmt_compile.c

/*
 * fmt_compile.c -- "compile" format strings for fmt_scan
 *
 * This code is Copyright (c) 2002, by the authors of nmh.  See the
 * COPYRIGHT file in the root directory of the nmh distribution for
 * complete copyright information.
 *
 * This code compiles the format strings (documented in mh-format(5)) into
 * an internal form to be later processed by fmt_scan.c.
 *
 * What happens here is that the format strings are parsed and an array
 * of struct format structures are returned.  Each format structure is
 * a single operation interpreted by the the routines in fmt_scan.c.
 *
 * There is a NOT a one-to-one correspondence between format strings and
 * format instructions; some functions have side effects that can result
 * in multiple instructions being generated.  The exact list of instructions
 * generated by a format string can be seem with the nmh fmtdump utility.
 *
 * A list of format instructions can be found in fmt_compile.h.
 *
 * If you wish to add a new function, you will need to do the following
 * things:
 *
 * - Add a new instruction to the list of instructions in fmt_compile.h.
 *   Note that test instructions (starting with FT_IF_S_NULL) have special
 *   handling, so if you are NOT writing a test function then you need
 *   to insert it into the list before that _and_ bump all of the
 *   following instruction numbers.
 *
 * - Add the function name to the functable[] array below, and write any
 *   special code that your function may require in terms of parsing
 *   (it very well may not need anything).
 *
 * - Add the code in fmt_scan.c to handle your new function.
 *
 * - Add code to fmtdump.c to display your new function.
 *
 * - Document the new function in the mh-format(5) man page.
 *
 */

#include <h/mh.h>
#include <h/addrsbr.h>
#include <h/tws.h>
#include <h/fmt_scan.h>
#include <h/fmt_compile.h>
#include <h/mts.h>
#include <h/utils.h>

#ifdef HAVE_SYS_TIME_H
# include <sys/time.h>
#endif
#include <time.h>

/*
 * hash table for deciding if a component is "interesting"
 */
static struct comp *wantcomp[128];

static struct format *formatvec;        /* array to hold formats */
static struct format *next_fp;          /* next free format slot */
static struct format *fp;               /* current format slot   */
static struct comp *cm;                 /* most recent comp ref  */
static struct ftable *ftbl;             /* most recent func ref  */
static int ncomp;
static int infunction;                  /* function nesting cnt  */

extern struct mailname fmt_mnull;

/* ftable->type (argument type) */
#define TF_COMP    0        /* component expected                 */
#define TF_NUM     1        /* number expected                    */
#define TF_STR     2        /* string expected                    */
#define TF_EXPR    3        /* component or func. expected        */
#define TF_NONE    4        /* no argument                        */
#define TF_MYBOX   5        /* special - get current user's mbox  */
#define TF_NOW     6        /* special - get current unix time    */
#define TF_EXPR_SV 7        /* like expr but save current str reg */
#define TF_NOP     8        /* like expr but no result            */
#define TF_MYNAME  9        /* special - get current name of user */
#define TF_MYHOST  10       /* special - get "local" hostname     */
#define TF_LMBOX   11       /* special - get full local mailbox   */

/* ftable->flags */
/* NB that TFL_PUTS is also used to decide whether the test
 * in a "%<(function)..." should be a string or numeric one.
 */
#define TFL_PUTS   1        /* implicit putstr if top level */
#define TFL_PUTN   2        /* implicit putnum if top level */

/*
 * The functable array maps between the text names of format functions and
 * the format instructions interpreted by the engine in fmt_scan.c.
 *
 * The elements of this structure are as follows:
 *
 * name -   The name of the function as seen in the format string.  This is
 *          what maps a particular function name into a format instruction.
 * type -   The type of argument this function expects.  Those types are
 *          listed above (with the TF_ prefix).  This affects what gets
 *          placed in the format instruction (the f_un union).
 * f_type - The instruction corresponding to this function (from the list
 *          in fmt_compile.h).
 * extra  - Used by some functions to provide extra data to the compiler.
 *          Uses include:
 *              - Providing an alternate instruction to combine a load
 *                and test operation (see do_if()).
 *              - Passed in f_value in the format instruction to provide
 *                extra information for the engine (see FT_LV_DAT handling
 *                in fmt_scan.c).
 *              - Provide a hint as to preprocessing that is required for
 *                this instruction (see do_name()).
 * flags  - See the definitions for TFL_PUTS & TFL_PUTN above.
 */

struct ftable {
    char *name;         /* function name                  */
    char type;          /* argument type                  */
    char f_type;        /* fmt type                       */
    char extra;         /* arg. type dependent extra info */
    char flags;
};

static struct ftable functable[] = {
     { "nonzero",    TF_EXPR,   FT_V_NE,        FT_IF_V_NE,     0 },
     { "zero",       TF_EXPR,   FT_V_EQ,        FT_IF_V_EQ,     0 },
     { "eq",         TF_NUM,    FT_V_EQ,        FT_IF_V_EQ,     0 },
     { "ne",         TF_NUM,    FT_V_NE,        FT_IF_V_NE,     0 },
     { "gt",         TF_NUM,    FT_V_GT,        FT_IF_V_GT,     0 },
     { "null",       TF_EXPR,   FT_S_NULL,      FT_IF_S_NULL,   0 },
     { "nonnull",    TF_EXPR,   FT_S_NONNULL,   FT_IF_S,        0 },
     { "match",      TF_STR,    FT_V_MATCH,     FT_IF_MATCH,    0 },
     { "amatch",     TF_STR,    FT_V_AMATCH,    FT_IF_AMATCH,   0 },

     { "putstr",     TF_EXPR,   FT_STR,         0,              0 },
     { "putstrf",    TF_EXPR,   FT_STRF,        0,              0 },
     { "putnum",     TF_EXPR,   FT_NUM,         0,              0 },
     { "putnumf",    TF_EXPR,   FT_NUMF,        0,              0 },
     { "putaddr",    TF_STR,    FT_PUTADDR,     0,              0 },
     { "putlit",     TF_EXPR,   FT_STRLIT,      0,              0 },
     { "zputlit",    TF_EXPR,   FT_STRLITZ,     0,              0 },
     { "void",       TF_NOP,    0,              0,              0 },

     { "comp",       TF_COMP,   FT_LS_COMP,     0,              TFL_PUTS },
     { "lit",        TF_STR,    FT_LS_LIT,      0,              TFL_PUTS },
     { "getenv",     TF_STR,    FT_LS_GETENV,   0,              TFL_PUTS },
     { "profile",    TF_STR,    FT_LS_CFIND,    0,              TFL_PUTS },
     { "decodecomp", TF_COMP,   FT_LS_DECODECOMP, 0,            TFL_PUTS },
     { "decode",     TF_EXPR,   FT_LS_DECODE,   0,              TFL_PUTS },
     { "trim",       TF_EXPR,   FT_LS_TRIM,     0,              0 },
     { "compval",    TF_COMP,   FT_LV_COMP,     0,              TFL_PUTN },
     { "compflag",   TF_COMP,   FT_LV_COMPFLAG, 0,              TFL_PUTN },
     { "num",        TF_NUM,    FT_LV_LIT,      0,              TFL_PUTN },
     { "msg",        TF_NONE,   FT_LV_DAT,      0,              TFL_PUTN },
     { "cur",        TF_NONE,   FT_LV_DAT,      1,              TFL_PUTN },
     { "size",       TF_NONE,   FT_LV_DAT,      2,              TFL_PUTN },
     { "width",      TF_NONE,   FT_LV_DAT,      3,              TFL_PUTN },
     { "unseen",     TF_NONE,   FT_LV_DAT,      4,              TFL_PUTN },
     { "dat",        TF_NUM,    FT_LV_DAT,      0,              TFL_PUTN },
     { "strlen",     TF_NONE,   FT_LV_STRLEN,   0,              TFL_PUTN },
     { "me",         TF_MYBOX,  FT_LS_LIT,      0,              TFL_PUTS },
     { "myname",     TF_MYNAME, FT_LS_LIT,      0,              TFL_PUTS },
     { "myhost",     TF_MYHOST, FT_LS_LIT,      0,              TFL_PUTS },
     { "localmbox",  TF_LMBOX,  FT_LS_LIT,      0,              TFL_PUTS },
     { "plus",       TF_NUM,    FT_LV_PLUS_L,   0,              TFL_PUTN },
     { "minus",      TF_NUM,    FT_LV_MINUS_L,  0,              TFL_PUTN },
     { "divide",     TF_NUM,    FT_LV_DIVIDE_L, 0,              TFL_PUTN },
     { "modulo",     TF_NUM,    FT_LV_MODULO_L, 0,              TFL_PUTN },
     { "charleft",   TF_NONE,   FT_LV_CHAR_LEFT, 0,             TFL_PUTN },
     { "timenow",    TF_NOW,    FT_LV_LIT,      0,              TFL_PUTN },

     { "month",      TF_COMP,   FT_LS_MONTH,    FT_PARSEDATE,   TFL_PUTS },
     { "lmonth",     TF_COMP,   FT_LS_LMONTH,   FT_PARSEDATE,   TFL_PUTS },
     { "tzone",      TF_COMP,   FT_LS_ZONE,     FT_PARSEDATE,   TFL_PUTS },
     { "day",        TF_COMP,   FT_LS_DAY,      FT_PARSEDATE,   TFL_PUTS },
     { "weekday",    TF_COMP,   FT_LS_WEEKDAY,  FT_PARSEDATE,   TFL_PUTS },
     { "tws",        TF_COMP,   FT_LS_822DATE,  FT_PARSEDATE,   TFL_PUTS },
     { "sec",        TF_COMP,   FT_LV_SEC,      FT_PARSEDATE,   TFL_PUTN },
     { "min",        TF_COMP,   FT_LV_MIN,      FT_PARSEDATE,   TFL_PUTN },
     { "hour",       TF_COMP,   FT_LV_HOUR,     FT_PARSEDATE,   TFL_PUTN },
     { "mday",       TF_COMP,   FT_LV_MDAY,     FT_PARSEDATE,   TFL_PUTN },
     { "mon",        TF_COMP,   FT_LV_MON,      FT_PARSEDATE,   TFL_PUTN },
     { "year",       TF_COMP,   FT_LV_YEAR,     FT_PARSEDATE,   TFL_PUTN },
     { "yday",       TF_COMP,   FT_LV_YDAY,     FT_PARSEDATE,   TFL_PUTN },
     { "wday",       TF_COMP,   FT_LV_WDAY,     FT_PARSEDATE,   TFL_PUTN },
     { "zone",       TF_COMP,   FT_LV_ZONE,     FT_PARSEDATE,   TFL_PUTN },
     { "clock",      TF_COMP,   FT_LV_CLOCK,    FT_PARSEDATE,   TFL_PUTN },
     { "rclock",     TF_COMP,   FT_LV_RCLOCK,   FT_PARSEDATE,   TFL_PUTN },
     { "sday",       TF_COMP,   FT_LV_DAYF,     FT_PARSEDATE,   TFL_PUTN },
     { "szone",      TF_COMP,   FT_LV_ZONEF,    FT_PARSEDATE,   TFL_PUTN },
     { "dst",        TF_COMP,   FT_LV_DST,      FT_PARSEDATE,   TFL_PUTN },
     { "pretty",     TF_COMP,   FT_LS_PRETTY,   FT_PARSEDATE,   TFL_PUTS },
     { "nodate",     TF_COMP,   FT_LV_COMPFLAG, FT_PARSEDATE,   TFL_PUTN },
     { "date2local", TF_COMP,   FT_LOCALDATE,   FT_PARSEDATE,   0 },
     { "date2gmt",   TF_COMP,   FT_GMTDATE,     FT_PARSEDATE,   0 },

     { "pers",       TF_COMP,   FT_LS_PERS,     FT_PARSEADDR,   TFL_PUTS },
     { "mbox",       TF_COMP,   FT_LS_MBOX,     FT_PARSEADDR,   TFL_PUTS },
     { "host",       TF_COMP,   FT_LS_HOST,     FT_PARSEADDR,   TFL_PUTS },
     { "path",       TF_COMP,   FT_LS_PATH,     FT_PARSEADDR,   TFL_PUTS },
     { "gname",      TF_COMP,   FT_LS_GNAME,    FT_PARSEADDR,   TFL_PUTS },
     { "note",       TF_COMP,   FT_LS_NOTE,     FT_PARSEADDR,   TFL_PUTS },
     { "addr",       TF_COMP,   FT_LS_ADDR,     FT_PARSEADDR,   TFL_PUTS },
     { "proper",     TF_COMP,   FT_LS_822ADDR,  FT_PARSEADDR,   TFL_PUTS },
     { "type",       TF_COMP,   FT_LV_HOSTTYPE, FT_PARSEADDR,   TFL_PUTN },
     { "ingrp",      TF_COMP,   FT_LV_INGRPF,   FT_PARSEADDR,   TFL_PUTN },
     { "nohost",     TF_COMP,   FT_LV_NOHOSTF,  FT_PARSEADDR,   TFL_PUTN },
     { "formataddr", TF_EXPR_SV,FT_FORMATADDR,  FT_FORMATADDR,  0 },
     { "concataddr", TF_EXPR_SV,FT_CONCATADDR,  FT_FORMATADDR,  0 },
     { "friendly",   TF_COMP,   FT_LS_FRIENDLY, FT_PARSEADDR,   TFL_PUTS },

     { "mymbox",     TF_COMP,   FT_LV_COMPFLAG, FT_MYMBOX,      TFL_PUTN },
     { "addtoseq",   TF_STR,    FT_ADDTOSEQ,    0,              0 },

     { "unquote",   TF_EXPR,    FT_LS_UNQUOTE,  0,              TFL_PUTS},

     { NULL,         0,         0,              0,              0 }
};

/* 
 * Hash function for component name.  The function should be
 * case independent and probably shouldn't involve a routine
 * call.  This function is pretty good but will not work on
 * single character component names.  
 */
#define CHASH(nm) (((((nm)[0]) - ((nm)[1])) & 0x1f) + (((nm)[2]) & 0x5f))

/*
 * Find a component in the hash table.
 */
#define FINDCOMP(comp,name) \
                for (comp = wantcomp[CHASH(name)]; \
                     comp && strcmp(comp->c_name,name); \
                     comp = comp->c_next) \
                ;

/* Add new component to the hash table */
#define NEWCOMP(cm,name) do { \
        cm = ((struct comp *) calloc(1, sizeof (struct comp)));\
        cm->c_name = getcpy(name);\
        cm->c_refcount++;\
        ncomp++;\
        i = CHASH(name);\
        cm->c_next = wantcomp[i];\
        wantcomp[i] = cm; \
        } while (0)

#define NEWFMT (next_fp++)
#define NEW(type,fill,wid) do {\
        fp=NEWFMT; fp->f_type=(type); fp->f_fill=(fill); fp->f_width=(wid); \
        } while (0)

/* Add (possibly new) component to the hash table */
#define ADDC(name) do { \
        FINDCOMP(cm, name);\
        if (!cm) {\
            NEWCOMP(cm,name);\
        }\
        fp->f_comp = cm; \
        fp->f_flags |= FF_COMPREF; \
        cm->c_refcount++; \
        } while (0)

#define LV(type, value)         do { NEW(type,0,0); fp->f_value = (value); } while (0)
#define LS(type, str)           do { NEW(type,0,0); fp->f_text = getcpy(str); fp->f_flags |= FF_STRALLOC; } while (0)

#define PUTCOMP(comp)           do { NEW(FT_COMP,0,0); ADDC(comp); } while (0)
#define PUTLIT(str)             do { NEW(FT_LIT,0,0); fp->f_text = getcpy(str); fp->f_flags |= FF_STRALLOC; } while (0)
#define PUTC(c)                 do { NEW(FT_CHAR,0,0); fp->f_char = (c); } while (0)

static char *format_string;
static unsigned char *usr_fstring;      /* for CERROR */

#define CERROR(str) compile_error (str, cp)

/*
 * static prototypes
 */
static struct ftable *lookup(char *);
static void compile_error(char *, char *);
static char *compile (char *);
static char *do_spec(char *);
static char *do_name(char *, int);
static char *do_func(char *);
static char *do_expr (char *, int);
static char *do_loop(char *);
static char *do_if(char *);
static void free_component(struct comp *);
static void free_comptable(void);


/*
 * Lookup a function name in the functable
 */
static struct ftable *
lookup(char *name)
{
    register struct ftable *t = functable;
    register char *nm;
    register char c = *name;

    while ((nm = t->name)) {
        if (*nm == c && strcmp (nm, name) == 0)
            return (ftbl = t);

        t++;
    }
    return (struct ftable *) 0;
}


static void
compile_error(char *str, char *cp)
{
    int i, errpos, errctx;

    errpos = cp - format_string;
    errctx = errpos > 20 ? 20 : errpos;
    usr_fstring[errpos] = '\0';

    for (i = errpos-errctx; i < errpos; i++) {
#ifdef LOCALE
        if (iscntrl(usr_fstring[i]))
#else
        if (usr_fstring[i] < 32)
#endif
            usr_fstring[i] = '_';
    }

    advise(NULL, "\"%s\": format compile error - %s",
           &usr_fstring[errpos-errctx], str);
    adios (NULL, "%*s", errctx+1, "^");
}

/*
 * Compile format string "fstring" into format list "fmt".
 * Return the number of header components found in the format
 * string.
 */

int
fmt_compile(char *fstring, struct format **fmt, int reset_comptable)
{
    register char *cp;
    size_t i;
    static int comptable_initialized = 0;

    format_string = getcpy (fstring);
    usr_fstring = fstring;

    if (reset_comptable || !comptable_initialized) {
        free_comptable();
        comptable_initialized = 1;
    }

    memset((char *) &fmt_mnull, 0, sizeof(fmt_mnull));

    /* it takes at least 4 char to generate one format so we
     * allocate a worst-case format array using 1/4 the length
     * of the format string.  We actually need twice this much
     * to handle both pre-processing (e.g., address parsing) and
     * normal processing.
     */
    i = strlen(fstring)/2 + 1;
                if (i==1) i++;
    next_fp = formatvec = (struct format *)calloc ((size_t) i,
                                                   sizeof(struct format));
    if (next_fp == NULL)
        adios (NULL, "unable to allocate format storage");

    infunction = 0;

    cp = compile(format_string);
    if (*cp) {
        CERROR("extra '%>', '%|' or '%?'");
    }
    LV(FT_DONE, 0);             /* really done */
    *fmt = formatvec;

    free(format_string);
    return (ncomp);
}

static char *
compile (char *sp)
{
    register char *cp = sp;
    register int  c;

    for (;;) {
        sp = cp;
        while ((c = *cp) && c != '%')
            cp++;
        *cp = 0;
        switch (cp-sp) {
        case 0:
            break;
        case 1:
            PUTC(*sp);
            break;
        default:
            PUTLIT(sp);
            break;
        }
        if (c == 0)
            return (cp);

        switch (c = *++cp) {
        case '%':
            PUTC (*cp);
            cp++;
            break;

        case '|':
        case '>':
        case '?':
        case ']':
            return (cp);

        case '<':
            cp = do_if(++cp);
            break;

        case '[':       /* ] */
            cp = do_loop(++cp);
            break;

        case ';':       /* comment line */
            cp++;
            while ((c = *cp++) && c != '\n')
                continue;
            break;

        default:
            cp = do_spec(cp);
            break;
        }
    }
}


/*
 * Process functions & components (handle field width here as well
 */
static char *
do_spec(char *sp)
{
    register char *cp = sp;
    register int c;
#ifndef lint
    register int ljust = 0;
#endif  /* not lint */
    register int wid = 0;
    register char fill = ' ';

    c = *cp++;
    if (c == '-') {
        ljust++;
        c = *cp++;
    }
    if (c == '0') {
        fill = c;
        c = *cp++;
    }
    while (isdigit(c)) {
        wid = wid*10 + (c - '0');
        c = *cp++;
    }
    if (c == '{') {
        cp = do_name(cp, 0);
        if (! infunction)
            fp->f_type = wid? FT_COMPF : FT_COMP;
    }
    else if (c == '(') {
        cp = do_func(cp);
        if (! infunction) {
            if (ftbl->flags & TFL_PUTS) {
                LV( wid? FT_STRF : FT_STR, ftbl->extra);
            }
            else if (ftbl->flags & TFL_PUTN) {
                LV( wid? FT_NUMF : FT_NUM, ftbl->extra);
            }
        }
    }
    else {
        CERROR("component or function name expected");
    }
    if (ljust)
        wid = -wid;
    fp->f_width = wid;
    fp->f_fill = fill;

    return (cp);
}

/*
 * Process a component name.  Normally this involves generating an FT_COMP
 * instruction for the specified component.  If preprocess is set, then we
 * do some extra processing.
 */
static char *
do_name(char *sp, int preprocess)
{
    register char *cp = sp;
    register int c;
    register int i;
    static int primed = 0;

    while (isalnum(c = *cp++) || c == '-' || c == '_')
        ;
    if (c != '}') {
        CERROR("'}' expected");
    }
    cp[-1] = '\0';
    PUTCOMP(sp);
    switch (preprocess) {

    case FT_PARSEDATE:
        if (cm->c_type & CT_ADDR) {
            CERROR("component used as both date and address");
        }
        cm->c_tws = (struct tws *)
            calloc((size_t) 1, sizeof(*cm->c_tws));
        fp->f_type = preprocess;
        PUTCOMP(sp);
        cm->c_type |= CT_DATE;
        break;

    case FT_MYMBOX:
        if (!primed) {
            ismymbox ((struct mailname *) 0);
            primed++;
        }
        /* fall through */
    case FT_PARSEADDR:
        if (cm->c_type & CT_DATE) {
            CERROR("component used as both date and address");
        }
        cm->c_mn = &fmt_mnull;
        fp->f_type = preprocess;
        PUTCOMP(sp);
        cm->c_type |= CT_ADDR;
        break;

    case FT_FORMATADDR:
        if (cm->c_type & CT_DATE) {
            CERROR("component used as both date and address");
        }
        cm->c_type |= CT_ADDR;
        break;
    }
    return (cp);
}

/*
 * Generate one or more instructions corresponding to the named function.
 * The different type of function arguments are handled here.
 */
static char *
do_func(char *sp)
{
    register char *cp = sp;
    register int c;
    register struct ftable *t;
    register int n;
    int mflag;          /* minus sign in NUM */

    infunction++;

    while (isalnum(c = *cp++)) 
        ;
    if (c != '(' && c != '{' && c != ' ' && c != ')') {
        CERROR("'(', '{', ' ' or ')' expected");
    }
    cp[-1] = '\0';
    if ((t = lookup (sp)) == 0) {
        CERROR("unknown function");
    }
    if (isspace(c))
        c = *cp++;

    switch (t->type) {

    case TF_COMP:
        if (c != '{') {
            CERROR("component name expected");
        }
        cp = do_name(cp, t->extra);
        fp->f_type = t->f_type;
        c = *cp++;
        break;

    case TF_NUM:
        if ((mflag = (c == '-')))
            c = *cp++;
        n = 0;
        while (isdigit(c)) {
            n = n*10 + (c - '0');
            c = *cp++;
        }
        if (mflag)
            n = (-n);
        LV(t->f_type,n);
        break;

    case TF_STR:
        sp = cp - 1;
        while (c && c != ')')
            c = *cp++;
        cp[-1] = '\0';
        LS(t->f_type,sp);
        break;

    case TF_NONE:
        LV(t->f_type,t->extra);
        break;

    case TF_MYBOX:
        LS(t->f_type, getusername());
        break;

    case TF_MYNAME:
        LS(t->f_type, getfullname());
        break;

    case TF_MYHOST:
        LS(t->f_type, LocalName(0));
        break;

    case TF_LMBOX:
        LS(t->f_type, getlocalmbox());
        break;

    case TF_NOW:
        LV(t->f_type, time((time_t *) 0));
        break;

    case TF_EXPR_SV:
        LV(FT_SAVESTR, 0);
        /* fall through */
    case TF_EXPR:
        *--cp = c;
        cp = do_expr(cp, t->extra);
        LV(t->f_type, 0);
        c = *cp++;
        ftbl = t;
        break;

    case TF_NOP:
        *--cp = c;
        cp = do_expr(cp, t->extra);
        c = *cp++;
        ftbl = t;
        break;
    }
    if (c != ')') {
        CERROR("')' expected");
    }
    --infunction;
    return (cp);
}

/*
 * Handle an expression as an argument.  Basically we call one of do_name(),
 * do_func(), or do_if()
 */
static char *
do_expr (char *sp, int preprocess)
{
    register char *cp = sp;
    register int  c;

    if ((c = *cp++) == '{') {
        cp = do_name (cp, preprocess);
        fp->f_type = FT_LS_COMP;
    } else if (c == '(') {
        cp = do_func (cp);
    } else if (c == ')') {
        return (--cp);
    } else if (c == '%' && *cp == '<') {
        cp = do_if (cp+1);
    } else {
        CERROR ("'(', '{', '%<' or ')' expected");
    }
    return (cp);
}

/*
 * I am guessing this was for some kind of loop statement, which would have
 * looked like %[ .... %].  It looks like the way this would have worked
 * is that the format engine would have seen that FT_DONE had a 1 in the
 * f_un.f_un_value and then decided whether or not to continue the loop.
 * There is no support for this in the format engine, so right now if
 * you try using it you will reach the FT_DONE and simply stop.  I'm leaving
 * this here in case someone wants to continue the work.
 *
 * Okay, got some more information on this from John L. Romine!  From an
 * email he sent to the nmh-workers mailing list on December 2, 2010, he
 * explains it thusly:
 *
 *    In this case (scan, formatsbr) it has to do with an extension to
 *    the mh-format syntax to allow for looping.
 *
 *    The scan format is processed once for each message.  Those #ifdef
 *    JLR changes allowed for the top part of the format file to be
 *    processed once, then a second, looping part to be processed
 *    once per message.  As I recall, there were new mh-format escape
 *    sequences to delimit the loop.  This would have allowed for things
 *    like per-format column headings in the scan output.
 *
 *    Since existing format files didn't include the scan listing
 *    header (it was hard-coded in scan.c) it would not have been
 *    backward-compatible.  All existing format files (including any
 *    local ones) would have needed to be changed to include the format
 *    codes for a header.  The practice at the time was not to introduce
 *    incompatible changes in a minor release, and I never managed to
 *    put out a newer major release.
 *
 * I can see how this would work, and I suspect part of the motivation was
 * because the format compiler routines (at the time) couldn't really be
 * called multiple times on the same message because the memory management
 * was so lousy.  That's been reworked and things are now a lot cleaner,
 * so I suspect if we're going to allow a format string to be used for the
 * scan header it might be simpler to have a separate format string just
 * for the header.  But I'll leave this code in for now just in case we
 * decide that we want some kind of looping support.
 */
static char *
do_loop(char *sp)
{
    register char *cp = sp;
    struct format *floop;

    floop = next_fp;
    cp = compile (cp);
    if (*cp++ != ']')
        CERROR ("']' expected");

    LV(FT_DONE, 1);             /* not yet done */
    LV(FT_GOTO, 0);
    fp->f_skip = floop - fp;    /* skip backwards */

    return cp;
}

/*
 * Handle an if-elsif-endif statement.  Note here that the branching
 * is handled by the f_skip member of the struct format (which is really
 * just f_width overloaded).  This number controls how far to move forward
 * (or back) in the format instruction array.
 */
static char *
do_if(char *sp)
{
    register char *cp = sp;
    register struct format *fexpr,
                           *fif = (struct format *)NULL;
    register int c = '<';

    for (;;) {
        if (c == '<') {                 /* doing an IF */
            if ((c = *cp++) == '{') /*}*/{
                cp = do_name(cp, 0);
                fp->f_type = FT_LS_COMP;
                LV (FT_IF_S, 0);
            }
            else if (c == '(') {
                cp = do_func(cp);
                /* see if we can merge the load and the "if" */
                if (ftbl->f_type >= IF_FUNCS)
                    fp->f_type = ftbl->extra;
                else {
                    /* Put out a string test or a value test depending
                     * on what this function's return type is.
                     */
                    if (ftbl->flags & TFL_PUTS) {
                        LV (FT_IF_S, 0);
                    } else {
                        LV (FT_IF_V_NE, 0);
                    }
                }
            }
            else {
                CERROR("'(' or '{' expected");  /*}*/
            }
        }

        fexpr = fp;                     /* loc of [ELS]IF */
        cp = compile (cp);              /* compile IF TRUE stmts */
        if (fif)
            fif->f_skip = next_fp - fif;

        if ((c = *cp++) == '|') {       /* the last ELSE */
            LV(FT_GOTO, 0);
            fif = fp;                   /* loc of GOTO */
            fexpr->f_skip = next_fp - fexpr;

            fexpr = (struct format *)NULL;/* no extra ENDIF */

            cp = compile (cp);          /* compile ELSE stmts */
            fif->f_skip = next_fp - fif;
            c = *cp++;
        }
        else if (c == '?') {            /* another ELSIF */
            LV(FT_GOTO, 0);
            fif = fp;                   /* loc of GOTO */
            fexpr->f_skip = next_fp - fexpr;

            c = '<';                    /* impersonate an IF */
            continue;
        }
        break;
    }

    if (c != '>') {
        CERROR("'>' expected.");
    }

    if (fexpr)                          /* IF ... [ELSIF ...] ENDIF */
        fexpr->f_skip = next_fp - fexpr;

    return (cp);
}

/*
 * Free a set of format instructions.
 *
 * What we do here is:
 *
 * - Iterate through the list of format instructions, freeing any references
 *   to allocated memory in each instruction.
 * - Free component references.
 * - If requested, reset the component hash table; that will also free any
 *   references to components stored there.
 *
 */

void
fmt_free(struct format *fmt, int reset_comptable)
{
    struct format *fp = fmt;

    if (fp) {
        while (! (fp->f_type == FT_DONE && fp->f_value == 0)) {
            if (fp->f_flags & FF_STRALLOC)
                free(fp->f_text);
            if (fp->f_flags & FF_COMPREF)
                free_component(fp->f_comp);
            fp++;
        }
        free(fmt);
    }

    if (reset_comptable)
        free_comptable();
}

/*
 * Find a component in our hash table.  This is just a public interface to
 * the FINDCOMP macro, so we don't have to expose our hash table.
 */

struct comp *
fmt_findcomp(char *component)
{
    struct comp *cm;

    FINDCOMP(cm, component);

    return cm;
}

/*
 * Like fmt_findcomp, but case-insensitive.
 */

struct comp *
fmt_findcasecomp(char *component)
{
    struct comp *cm;

    for (cm = wantcomp[CHASH(component)]; cm; cm = cm->c_next)
        if (mh_strcasecmp(component, cm->c_name) == 0)
            break;

    return cm;
}

/*
 * Add an entry to the component hash table
 *
 * Returns true if the component was added, 0 if it already existed.
 *
 */

int
fmt_addcompentry(char *component)
{
    struct comp *cm;
    int i;

    FINDCOMP(cm, component);

    if (cm)
        return 0;

    NEWCOMP(cm, component);

    /*
     * ncomp is really meant for fmt_compile() and this function is
     * meant to be used outside of it.  So decrement it just to be safe
     * (internal callers should be using NEWCOMP()).
     */

    ncomp--;

    return 1;
}

/*
 * Add a string to a component hash table entry.
 *
 * Note the special handling for components marked with CT_ADDR.  The comments
 * in fmt_scan.h explain this in more detail.
 */

int
fmt_addcomptext(char *component, char *text)
{
    int i, found = 0, bucket = CHASH(component);
    struct comp *cptr = wantcomp[bucket];
    char *cp;

    while (cptr) {
        if (mh_strcasecmp(component, cptr->c_name) == 0) {
            found++;
            if (! cptr->c_text) {
                cptr->c_text = getcpy(text);
            } else {
                i = strlen(cp = cptr->c_text) - 1;
                if (cp[i] == '\n') {
                    if (cptr->c_type & CT_ADDR) {
                        cp[i] = '\0';
                        cp = add(",\n\t", cp);
                    } else {
                        cp = add("\t", cp);
                    }
                }
                cptr->c_text = add(text, cp);
            }
        }
        cptr = cptr->c_next;
    }

    return found ? bucket : -1;
}

/*
 * Append text to a component we've already found.  See notes in fmt_scan.h
 * for more information.
 */

void
fmt_appendcomp(int bucket, char *component, char *text)
{
    struct comp *cptr;

    if (bucket != -1) {
        for (cptr = wantcomp[bucket]; cptr; cptr = cptr->c_next)
            if (mh_strcasecmp(component, cptr->c_name) == 0)
                cptr->c_text = add(text, cptr->c_text);
    }
}

/*
 * Free and reset our component hash table
 */

static void
free_comptable(void)
{
    unsigned int i;
    struct comp *cm, *cm2;

    for (i = 0; i < sizeof(wantcomp)/sizeof(wantcomp[0]); i++) {
        cm = wantcomp[i];
        while (cm != NULL) {
            cm2 = cm->c_next;
            free_component(cm);
            cm = cm2;
        }
        wantcomp[i] = 0;
    }

    ncomp = 0;
}

/*
 * Decrement the reference count of a component structure.  If it reaches
 * zero, free it
 */

static void
free_component(struct comp *cm)
{
    if (--cm->c_refcount <= 0) {
        /* Shouldn't ever be NULL, but just in case ... */
        if (cm->c_name)
            free(cm->c_name);
        if (cm->c_text)
            free(cm->c_text);
        if (cm->c_type & CT_DATE)
            free(cm->c_tws);
        if (cm->c_type & CT_ADDR && cm->c_mn && cm->c_mn != &fmt_mnull)
            mnfree(cm->c_mn);
        free(cm);
    }
}