printf.w

\input mwebmac
\input ctmac

\RCSID{$Id: printf.w,v 1.5 2016/07/25 14:55:24 mike Exp $}

\TOC{Mawk Printf}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\section{Introduction}
This is the source and documentation for a new [[mawk]] implementation
of [[s?printf()]].  The purpose of the reimplementation is,

\list
\item{(1)} Remove the use of a fixed buffer for [[sprintf()]].
\item{(2)} Handle null character, [['\0']], in the format string
and the argument strings.
\item{(3)} For integer formats, e.g. [[%d]], handle bigger integers.
The original implementation was restricted to 32-bit integers.
This implementation uses 64-bit integers.

\item{(4)} When the format string is a constant, which is the usual case,
parse it at compile-time as opposed to run-time.  This removes
some redundant computation at run-time.
\endlist

\section{The Interface}
To execute [[printf(form,...)]] or [[sprintf(form,...)]], a call is
made to [[parse_form(form)]], which returns a [[const Form*]].  To the caller,
this is an abstract type.

<<[[printf]] definitions>>=
typedef struct form Form ;
const Form* parse_form(const STRING*) ;

@
To execute [[s?printf(form,arg1,...argn)]], a call is made to
[[do_xprint(fp,form,cp)]], where

\I[[fp]] is a [[FILE*]] for output or zero for [[sprintf]].
\I[[form]] is the [[const Form*]] returned from [[parse_form()]].
\I[[cp]] is a pointer to [[arg1..argn]] in an array of [[CELL]].
\endhitems

\noindent
If doing [[printf]], the return is zero.  If doing [[sprintf]], the
return is the resulting [[STRING*]].


@ 
\section{Form Data}
This section handles the parsing and storage of [[form]] in
the function calls {\hfil\break}[[printf(form,...)]] and [[sprintf(form,...)]].

<<[[printf]] definitions>>=
typedef struct spec Spec ;
struct form {
    Form* link ;
    STRING* form ;
    unsigned num_args ;  /* number of args to satisfy form */
    Spec* specs ;  /* each  %..C  is one Spec */
    STRING* ending ;   /* after last Spec */
} ;
@ %def Form

All the [[Form]]s are kept on a singly linked list, [[the_forms]].

<<[[form]] definitions and data>>=
static Form* the_forms ;
@ %def the_forms

This function finds a [[form]] on the list.  Successful search causes
move to the front.

<<[[printf]] functions>>=
static Form* find_form(const STRING* form)
{
    Form* p = the_forms ;
    Form* q = 0 ;
    while(p) {
        if (STRING_eq(form,p->form)) {
	    if (q) { /* move to the front */
	        q->link = p->link ;
		p->link = the_forms ;
		the_forms = p ;
	    }
	    return p ;
	}
	else {
	    q = p ;
	    p = p->link ;
	}
    }
    /* not found */
    return 0 ;
}
@ %def find_form

<<static prototypes>>=
static Form* find_form(const STRING*) ;

<<[[printf]] functions>>=
const Form* parse_form(const STRING* form)
{
    Form* ret = find_form(form) ;
    if (ret) return ret ;
    ret = parse_form0(form) ;
    /* zero if an error */
    if (ret) {
        ret->link = the_forms ;
        the_forms = ret ;
    }
    return ret ;
}
@ %def parse_form

The real work of parsing a format happens here.

<<[[printf]] functions>>=
static Form* parse_form0(const STRING* form)
{
    STRING* prefix ;
    Spec tmp_head ;
    Spec* tail = &tmp_head ;
    char xbuffer[1024] ;
    char* buffer = form->len <= 1024 ? xbuffer : (char*)emalloc(form->len) ;
    const char* str = form->str ; /* walks form */
    const char* end = str + form->len ;

    tmp_head.link = 0 ;
    Form* ret = (Form*)zmalloc(sizeof(Form)) ;

    while(1) {
        <<get the [[prefix]]>>
	if (str == end) 
	    <<done, so finish up and return>>
	else {
	    str = parse_spec(str+1,end,tail) ;
	    if (str == 0) {
	        /* error, msg already done */
		return 0 ;
	    }
	    tail = tail->link ;
	    tail->prefix = prefix ;
	}
    }
}
@ %def parse_form0

<<static prototypes>>=
static Form* parse_form0(const STRING*) ;

@
Everything up to a [[%]] goes into the [[prefix]], which is
collected in [[buffer]].  [[%%]] becomes [[%]] in the [[prefix]].

<<get the [[prefix]]>>=
{
    char* bp = buffer ;
    prefix = 0 ;
    while(str < end) {
        if (*str == '%') {
	    if (str[1] == '%') {
	        *bp++ = '%' ;
		str += 2 ;
		continue ;
	    }
	    else { /* prefix is done */
	        if (bp > buffer) {
	            prefix = new_STRING2(buffer, bp - buffer) ;
		    bp = buffer ;
		}
		break ;
	    }
	}
	else *bp++ = *str++ ;
    }
    if (bp > buffer) {
        /* prefix is really the ending */
	if (bp == buffer+1 && buffer[0] == '\n') {
	    prefix = STRING_dup(the_nl_str) ;
	}
	else {
	    prefix = new_STRING2(buffer, bp - buffer) ;
	}
    }
}
@
When we get here,
the collected [[prefix]] is actually the [[ending]], everything after
the last control character.  This code finishes the construction
of the [[Form]], returning a pointer to it.
	    
<<done, so finish up and return>>=
{
    ret->ending = prefix ;
    ret->form = STRING_dup(form) ;
    ret->link = 0 ;
    ret->specs = tmp_head.link ;
    ret->num_args = 0 ;
    {
        const Spec* sp = ret->specs ;
	while(sp) {
	    ret->num_args += 1 + sp->ast_cnt ;
	    sp = sp->link ;
	}
    }
    if (buffer != xbuffer) free(buffer) ;
    return ret ;
}


@
Often the [[ending]] is only a new-line, here is a [[STRING]]
to handle that case.

<<[[form]] definitions and data>>=
static STRING the_nl = {1,1,"\n"} ;
static STRING* const the_nl_str = &the_nl ;
@ %def the_nl_str 


@
Function [[parse_spec()]] analyzes [["%...C"]], where [[C]] is
the terminating control character.  On success, a pointer to past the
end of the [[Spec]] is returned and the
[[Spec]] is placed at the back of
a singly-linked list at [[tail]].

<<[[form]] definitions and data>>=
struct spec {
    Spec* link ;
    unsigned type ;
    const STRING* prefix ;
    int width ;
    int prec ;
    int minus_flag ;
    unsigned ast_cnt ;
    const char* form ;
} ;
@ %def Spec

<<[[printf]] functions>>=
const char* parse_spec(const char* str, const char* end, Spec* tail)
{
    const char* const start = str-1 ;
    Spec* spec = (Spec*)zmalloc(sizeof(Spec)) ;
    spec->link = 0 ;
    spec->width = WP_NONE ;
    spec->prec  = WP_NONE ;
    spec->minus_flag = 0 ;
    spec->ast_cnt = 0 ;
    spec->form = 0 ;
    unsigned l_cnt = 0 ;

    <<do flags>>
    <<do width and precision>>
    <<count [['l']]>>
    <<finish with the control character>>
    tail->link = spec ;
    return str+1 ;
}
@ %def parse_spec
@ %def l_cnt 
	    
<<static prototypes>>=
static const char* parse_spec(const char*, const char*, Spec*) ;

@
The [[width]] and [[prec]] can be an [[*]], [[WP_AST]], 
or nothing, [[WP_NONE]], or a non-negative integer.

<<[[form]] definitions and data>>=
enum { WP_AST = -1, WP_NONE = -2 } ;
    
<<do flags>>=
{
    while(str < end) {
        switch(*str) {
	    case '-' : 
	        spec->minus_flag = 1 ;
		/* fall thru */
	    case '+' : case ' ' : case '#' : case '0' :
	        str++ ;
		continue ;
	    default:
	        break ;
	}
	break ; /* while */
    }
    if (str == end) {
        c_or_rt_error_or_silence(
	    "incomplete format specification \"%s\"", start) ;
	<<[[return]] error>>
    }
}

@
A null return indicates a misformed spec.  We do not attempt to
recover allocated memory as the program will exit soon.

<<[[return]] error>>=
{
    return 0 ;
}

<<do width and precision>>=
{
    if (*str == '*') {
	spec->width = WP_AST ;
	spec->ast_cnt = 1 ;
	str++ ;
    }
    else {
        unsigned char c = *(unsigned char*) str ;

	if (scan_code[c] == SC_DIGIT) {
	    spec->width = c - '0' ;
	    str++ ;
	    while(1) {
                c = *(unsigned char*) str ;
		if (scan_code[c] != SC_DIGIT) break ;
		spec->width = spec->width * 10 + c - '0' ;
		str++ ;
	    }
	}
    }
    /* width is done */
    if (*str == '.') {
        /* have precision , no numbers will be prec 0 */
	str++ ;
	if (*str == '*') {
	    spec->prec = WP_AST ;
	    spec->ast_cnt++ ;
	    str++ ;
	}
	else {
	    unsigned char c ;
	    spec->prec = 0 ;
	    while(1) {
	        c = * (unsigned char*) str ;
		if (scan_code[c] != SC_DIGIT) break ;
		spec->prec = spec->prec * 10 + c - '0' ;
		str++ ;
	    }
	}
    }
}

@
An [['l']] modifier is not necessary in [[awk]], but some awk's accept it and
earlier [[mawk]] did, so we accept it as long as followed by an integer
control character, [[d]], [[i]], etc.

<<count [['l']]>>=
{
    while(*str == 'l') {
        str++ ;
	l_cnt++ ;
    }
}

@
The next expected character is the control character and it determines
the type of the [[Spec]].   The types are,

\I[[PF_C]]  print a character, [[%c]].
\I[[PF_I]]  print a signed integer, [[%d]] or [[%i]].
\I[[PF_U]]  print an unsigned integer, [[%u]], [[%x]], [[%X]] or [[%o]].
\I[[PF_F]]  print a floating point number, [[%f]], [[%g]], [[%G]],
[[%e]] or [[%E]].
\I[[PF_S]]  print a string, [[%s]].
\endhitems

<<[[form]] definitions and data>>=
enum { PF_C = 0, PF_I, PF_U, PF_F, PF_S } ;
@ %def PF_C PF_I PF_U PF_F PF_S
	    
	    
<<finish with the control character>>=
{
    switch(*str) {
        case 's':
	    spec->type = PF_S ;
	    break ;
        case 'c':
	case 'e': case 'g': case 'f': case 'E': case 'G':
	    <<make the [[spec]] type [[PF_F]] or type [[PF_C]]>>
	    break ;
	case 'd': case 'i':
	case 'o': case 'x': case 'X': case 'u':
	    <<make the [[spec]] type [[PF_I]] or type [[PF_U]]>>
	    break ;
	default:
	   c_or_rt_error_or_silence(
  "invalid control character '%c' in [s]printf format (\"%s\")",
           *(unsigned char*) str, start) ;
	   <<[[return]] error>>
	   break ;
    }
}

@
An [['l']] modifier does not make sense here, but rather than make it
an error, we silently remove it.  This is a minor change in
[[mawk]] behavior.

<<make the [[spec]] type [[PF_F]] or type [[PF_C]]>>=
{
    size_t len = str+1 - start - l_cnt ;
    char* form = (char*) zmalloc(len+1) ;
    spec->type = *str == 'c' ? PF_C : PF_F ;
    memcpy(form, start, len-1) ;
    form[len] = 0 ;
    form[len-1] = *str ;
    spec->form = form ;
}

@
A [[printf]] form such as [["%2x"]] in an [[awk]] program needs to
be converted to [["%2lx"]] for a call to the 64 bit long C-library
or [["%2llx"]] for the 32 bit long C-library.  The user is allowed to use
[['l']] in an [[awk]] program, but it has no real effect.

<<make the [[spec]] type [[PF_I]] or type [[PF_U]]>>=
{
    size_t len = str+1 - start ;
    int delta = (have_long64? 1 : 2) - l_cnt ;
        /* delta is number of 'l' to add or remove */
    char* form ;

    len += delta ;
    form = (char*) zmalloc(len+1) ;
    if (*str == 'd' || *str == 'i') {
        spec->type = PF_I ;
    }
    else {
        spec->type = PF_U ;
    }
    form[len] = 0 ;
    form[len-1] = *str ;
    form[len-2] = 'l' ;
    if (!have_long64) {
        form[len-3] = 'l' ;
    }
    memcpy(form, start, len - (have_long64?2:3)) ;
    spec->form = form ;
}

   
@
\section{Output}
Up to the final output, to a file or copy to a string, the execution
of [[printf()]] and [[sprintf()]] is the same.  Both start at
[[do_xprintf(fp,form,cp)]] with arguments,

\I[[fp]] is [[FILE*]] to the output file, if doing [[printf()]]. If doing
[[sprintf()]] is zero.

\I[[form]] is the [[Form*]] parsed from the format string.

\I[[cp]] points into an array of arguments. The number of arguments is
[[form->num_args]].  The [[CELL]] arguments can be modified, but
the caller owns the cells.
\endhitems
\noindent
If doing [[sprintf()]], the return is the [[STRING*]] output, else zero.


<<[[printf]] functions>>=
STRING*  do_xprintf(
    FILE* fp,
    const Form* form,
    CELL* cp)
{
    const Spec* spec = form->specs ;

    sprintf_ptr = string_buff ;
    while(spec) {
	if (spec->prefix) {
	    xprint_string(spec->prefix->str,spec->prefix->len,fp) ;
	}
        if (spec->type == PF_S) {
	    xprint_string_spec(fp,spec,cp) ;
	}
	else {
            xprint_num_spec(fp,spec,cp) ;
	}
	cp +=  1 + spec->ast_cnt ;
	spec = spec->link ;
    }
    if (form->ending) {
        xprint_string(form->ending->str, form->ending->len, fp) ;
    }
    if (fp) {
        if (ferror(fp)) {
            write_error(fp) ;
	    mawk_exit(2) ;
	}
	return 0 ;
    }
    return new_STRING2(string_buff, sprintf_ptr - string_buff) ;
}
@ %def do_xprintf

<<[[printf]] definitions>>=
STRING* do_xprintf(FILE*, const Form*, CELL*) ;

@
Here are two helper functions.  The first outputs an array of characters
and the second outputs an array of spaces.

<<[[printf]] functions>>=
static void xprint_string(const char* str, size_t len,  FILE* fp) 
{
    if (fp) {
        fwrite(str, 1, len, fp) ;
    }
    else {
	size_t room = string_buff_end - sprintf_ptr ;
	while(room < len) {
	    sprintf_ptr = enlarge_string_buff(sprintf_ptr) ;
	    room = string_buff_end - sprintf_ptr ;
	}
	memcpy(sprintf_ptr, str, len) ;
	sprintf_ptr += len ;
    }
}

static void xprint_spaces(size_t len, FILE* fp)
{
    char buffer[1024] ;
    memset(buffer, ' ', len > 1024 ? 1024 : len) ;
    while(len > 1024) {
        xprint_string(buffer, 1024, fp) ;
	len -= 1024 ;
    }
    xprint_string(buffer, len, fp) ;
}
@ %def xprint_string
@ %def xprint_spaces

<<static prototypes>>=
static void xprint_string(const char*,size_t,FILE*) ;
static void xprint_spaces(size_t,FILE*) ;

@
When outputting for [[sprintf()]] to [[string_buff]], the
place to write is [[sprintf_ptr]] ;

<<static data>>=
char* sprintf_ptr ;
@ %def sprintf_ptr

@
\subsection{Output for a [[%..s]] Specification}
This function outputs the [[STRING]] at [[cp]] using a
[[PF_S]] [[spec]].  Note that since the [[STRING]] might have
[[\0]], we do the implementation instead of passing it to
a C-library function.

<<[[printf]] functions>>=
static 
void xprint_string_spec(FILE* fp, const Spec* spec, CELL* cp)
{
    int width = spec->width ;
    int prec = spec->prec ;
    const char* str ;
    size_t len ;
    int minus_flag = spec->minus_flag ;

    <<get [[width]] and [[prec]] for [[*]]>>
    cast1_to_s(cp) ;
    str = string(cp)->str ;
    len = string(cp)->len ;
    
    if (width < 0) width = 0 ; /* effectively ignore */
    if (prec < 0)  prec = len ; /* ditto */
    if (prec < len) len = prec ;

    /* output str and spaces if needed */
    if (len >= width) {
       xprint_string(str,len,fp) ;
    }
    else if (minus_flag) {
        /* left justify */
	xprint_string(str,len,fp) ;
	xprint_spaces(width-len,fp) ;
    }
    else {
	xprint_spaces(width-len,fp) ;
	xprint_string(str,len,fp) ;
    }
}
@ %def xprint_string_spec

<<static prototypes>>=
static void xprint_string_spec(FILE*,const Spec*,CELL*) ;

@
Here is a tricky point: for [[%*s]], if the value passed for [[*]] is
negative, e.g., [[-5]], then it is the same as passing [[5]] to
[[%-*s]].

<<get [[width]] and [[prec]] for [[*]]>>=
{
    if (width == WP_AST) {
	cast1_to_d(cp) ;
	width = d_to_int(cp->dval) ;
	if (width < 0) {
	    minus_flag = 1 ;
	    width = -width ;
	}
	cp++ ;
    }
    if (prec == WP_AST) {
	cast1_to_d(cp) ;
	prec = d_to_int(cp->dval) ;
	cp++ ;
    }
}

@
\subsection{Output for Numeric Specifications}
There are four numeric specifications. 
The function,{\hfil\break}[[xprint_num_spec(fp,spec,cp)]] breaks the
output into the four cases, converting the data in [[*cp]] to
the appropriate type.

<<[[printf]] functions>>=
static
void xprint_num_spec(FILE* fp, const Spec* spec, CELL* cp)
{
    unsigned ast_cnt = spec->ast_cnt ;
    int ast[2] ;

    <<set [[ast[2]]], incrementing [[cp]] as needed>>

    switch(spec->type) {
        case PF_F:
	    if (cp->type != C_DOUBLE) cast1_to_d(cp) ;
	    xprint_pf_f(fp, spec->form, ast_cnt, ast[0],ast[1],cp->dval) ;
	    break ;

        case PF_I:
	    if (cp->type != C_DOUBLE) cast1_to_d(cp) ;
	    xprint_pf_i(fp, spec->form, ast_cnt, ast[0],ast[1],
	                d_to_i64(cp->dval)) ;
	    break ;

        case PF_U:
	    if (cp->type != C_DOUBLE) cast1_to_d(cp) ;
	    xprint_pf_u(fp, spec->form, ast_cnt, ast[0],ast[1],
	                d_to_u64(cp->dval)) ;
	    break ;

        case PF_C:
	    {
	        unsigned c ;
		<<get the value of [[*cp]] into [[c]]>>
	        xprint_pf_c(fp, spec->form, ast_cnt, ast[0],ast[1],c) ;
	    }
	    break ;
    }
}
@ %def print_spec

<<static prototypes>>=
static void xprint_num_spec(FILE*, const Spec*, CELL*) ;

@
Note the value assigned to [[*]] is supposed to be non-negative,
but we do not check as the C-library function will handle it.

<<set [[ast[2]]], incrementing [[cp]] as needed>>=
{
    if (ast_cnt >= 1)  {
        cast1_to_d(cp) ;
        ast[0] = d_to_int(cp->dval) ;
	cp++ ;
    }
    if (ast_cnt == 2) {
        cast1_to_d(cp) ;
        ast[1] = d_to_int(cp->dval) ;
	cp++ ;
    }
}

@
[[%c]] is a little weird in [[awk]], for numeric values it uses the value,
but for string values, it uses the first character in the string or
zero for the empty string.

<<get the value of [[*cp]] into [[c]]>>=
{
    int64_t val ;

    if (cp->type == C_MBSTRN) {
        check_strnum(cp) ;
    }
    if (cp->type == C_STRING) {
        c = string(cp)->str[0] ;
    }
    else {
        cast1_to_d(cp) ;
	val = d_to_i64(cp->dval) ;
	c = (unsigned) (val & 0xff) ;
    }
}

@
If working in [[C++]], to get the four functions,
[[xprint_fp_{f,i,u,c}()]], we would write one template function and
instantiate it four times.  Since we are working
in ansi [[C]], the following sub-sections are instantiation by
cut and paste.
    

\subsubsection{Floating Point Format}

<<[[printf]] functions>>=
static void xprint_pf_f(FILE* fp, const char* form, 
    unsigned cnt, int x, int y, double d)
{
    if (fp) {
        switch(cnt) {
	    case 0:
	        fprintf(fp, form, d) ;
		break ;
	    case 1:
	        fprintf(fp, form, x, d) ;
		break ;
	    case 2:
	        fprintf(fp, form, x, y, d) ;
		break ;
	}
    }
    else {
        size_t room = string_buff_end - sprintf_ptr ;
	size_t used = (size_t)-1 ;

	while(1) {
	    switch(cnt) {
	        case 0:
	            used = snprintf(sprintf_ptr, room, form, d) ;
		    break ;
	        case 1:
	            used = snprintf(sprintf_ptr, room, form, x ,d) ;
		    break ;
	        case 2:
	            used = snprintf(sprintf_ptr, room, form, x,y,d) ;
		    break ;
	    }
	    if ((int) used < 0) snprintf_bozo() ;
	    
	    if (used <= room) break ; /* while */
	    do {
	        sprintf_ptr = enlarge_string_buff(sprintf_ptr) ;
		room = string_buff_end - sprintf_ptr ;
	    }
	    while(room < used) ;
	}
	sprintf_ptr += used ;
    }
}
@ %def xprint_pf_f

<<static prototypes>>=
static void xprint_pf_f(FILE*,const char*,unsigned,int,int,double) ;

@
\subsubsection{Integer Format}

<<[[printf]] functions>>=
static void xprint_pf_i(FILE* fp, const char* form, 
    unsigned cnt, int x, int y, int64_t val)
{
    if (fp) {
        switch(cnt) {
	    case 0:
	        fprintf(fp, form, val) ;
		break ;
	    case 1:
	        fprintf(fp, form, x, val) ;
		break ;
	    case 2:
	        fprintf(fp, form, x, y, val) ;
		break ;
	}
    }
    else {
        size_t room = string_buff_end - sprintf_ptr ;
	size_t used = (size_t)-1 ;

	while(1) {
	    switch(cnt) {
	        case 0:
	            used = snprintf(sprintf_ptr, room, form, val) ;
		    break ;
	        case 1:
	            used = snprintf(sprintf_ptr, room, form, x ,val) ;
		    break ;
	        case 2:
	            used = snprintf(sprintf_ptr, room, form, x,y,val) ;
		    break ;
	    }
	    if ((int) used < 0) snprintf_bozo() ;
	    
	    if (used <= room) break ; /* while */
	    do {
	        sprintf_ptr = enlarge_string_buff(sprintf_ptr) ;
		room = string_buff_end - sprintf_ptr ;
	    }
	    while(room < used) ;
	}
	sprintf_ptr += used ;
    }
}
@ %def xprint_pf_i

<<static prototypes>>=
static void xprint_pf_i(FILE*,const char*,unsigned,int,int, int64_t) ;

@
\subsubsection{Unsigned Integer Format}

<<[[printf]] functions>>=
static void xprint_pf_u(FILE* fp, const char* form, 
    unsigned cnt, int x, int y, uint64_t val)
{
    if (fp) {
        switch(cnt) {
	    case 0:
	        fprintf(fp, form, val) ;
		break ;
	    case 1:
	        fprintf(fp, form, x, val) ;
		break ;
	    case 2:
	        fprintf(fp, form, x, y, val) ;
		break ;
	}
    }
    else {
        size_t room = string_buff_end - sprintf_ptr ;
	size_t used = (size_t)-1 ;

	while(1) {
	    switch(cnt) {
	        case 0:
	            used = snprintf(sprintf_ptr, room, form, val) ;
		    break ;
	        case 1:
	            used = snprintf(sprintf_ptr, room, form, x ,val) ;
		    break ;
	        case 2:
	            used = snprintf(sprintf_ptr, room, form, x,y,val) ;
		    break ;
	    }
	    if ((int) used < 0) snprintf_bozo() ;
	    
	    if (used <= room) break ; /* while */
	    do {
	        sprintf_ptr = enlarge_string_buff(sprintf_ptr) ;
		room = string_buff_end - sprintf_ptr ;
	    }
	    while(room < used) ;
	}
	sprintf_ptr += used ;
    }
}
@ %def xprint_pf_u

<<static prototypes>>=
static void xprint_pf_u(FILE*,const char*,unsigned,int,int, uint64_t) ;

@
\subsubsection{Single Character Format, [[%c]]}
Here the input, [[val]], is known to be in the range, [[[0,255)]].

<<[[printf]] functions>>=
static void xprint_pf_c(FILE* fp, const char* form, 
    unsigned cnt, int x, int y, unsigned val)
{
    if (fp) {
        switch(cnt) {
	    case 0:
	        fprintf(fp, form, val) ;
		break ;
	    case 1:
	        fprintf(fp, form, x, val) ;
		break ;
	    case 2:
	        fprintf(fp, form, x, y, val) ;
		break ;
	}
    }
    else {
        size_t room = string_buff_end - sprintf_ptr ;
	size_t used = (size_t)-1 ;

	while(1) {
	    switch(cnt) {
	        case 0:
	            used = snprintf(sprintf_ptr, room, form, val) ;
		    break ;
	        case 1:
	            used = snprintf(sprintf_ptr, room, form, x ,val) ;
		    break ;
	        case 2:
	            used = snprintf(sprintf_ptr, room, form, x,y,val) ;
		    break ;
	    }
	    if ((int) used < 0) snprintf_bozo() ;
	    
	    if (used <= room) break ; /* while */
	    do {
	        sprintf_ptr = enlarge_string_buff(sprintf_ptr) ;
		room = string_buff_end - sprintf_ptr ;
	    }
	    while(room < used) ;
	}
	sprintf_ptr += used ;
    }
}
@ %def xprint_pf_c

<<static prototypes>>=
static void xprint_pf_c(FILE*,const char*,unsigned,int,int, unsigned) ;

<<[[printf]] functions>>=
static void snprintf_bozo() {
    rt_error("unexpected error (%d) in internal call to snprintf", errno) ;
    /* no return */
}
@ %def snprintf_bozo

<<static prototypes>>=
static void snprintf_bozo(void) ;

@
\section{Loose Ends}
@
\subsection{[[Form]] Disassembly}
Here is support for [[-Wd]].

<<[[printf]] functions>>=
void visible_string(FILE*, const STRING*, int) ; /* in da.c */
static void da_Spec(FILE*, const Spec*) ;
void da_Form(FILE* fp, const Form* form)
{
    const Spec* spec = form->specs ;
    while(spec) {
        da_Spec(fp,spec) ;
	spec = spec->link ;
    }
    if (form->ending) {
        visible_string(fp, form->ending,'\"') ;
    }
}
@ %def da_Form

<<[[printf]] definitions>>=
void da_Form(FILE*,const Form*) ;

<<[[printf]] functions>>=
static void da_Spec(FILE* fp, const Spec* spec)
{
    if (spec->prefix) visible_string(fp, spec->prefix, '\"') ;
    if (spec->type != PF_S) {
        fprintf(fp, "%s", spec->form) ;
    }
    else {
        fputc('%', fp) ;
	if (spec->minus_flag) fputc('-',fp) ;
	if (spec->width == WP_AST) {
	    fputc('*',fp) ;
	}
	else if (spec->width >= 0) {
	    fprintf(fp, "%d", spec->width) ;
	}
	if (spec->prec >= 0) {
	    fprintf(fp, ".%d", spec->prec) ;
	}
	else if (spec->prec == WP_AST) {
	    fputs(".*", fp) ;
	}
	fputc('s', fp) ;
    }
}
@ %def da_Spec

\subsection{Checking [[CONVFMT]]}
A bug report, against [[mawk]], reported a core-dump on bad assignment
to [[CONVFMT]].  The user is either careless or malicious, but we
will protect him/her by no longer allowing weird 
assignment to [[CONVFMT]] or [[OFMT]].
This is a minor change in [[mawk]]'s behavior.

If [[conv]] is a valid string for assignment to [[CONVFMT]], then
[[conv]] is returned else zero.  Valid means it parses to a [[Form*]]
that takes one argument of type [[PF_F]].  Any prefix or ending
does not contain [['\0']].

<<[[printf]] functions>>=
STRING* check_convfmt(const STRING* conv)
{
    const Form* form ;
    checking_convfmt = 1 ;
    form = parse_form(conv) ;
    checking_convfmt = 0 ;
    if (form == 0 || form->num_args != 1 || form->specs->type != PF_F) {
        return 0 ;
    }
    /* no \0 in prefix or ending */
    if (form->ending && memchr(form->ending->str, 0, form->ending->len)) {
        return 0 ;
    }

    {
        const STRING* prefix = form->specs->prefix ;
	if (prefix && memchr(prefix->str, 0, prefix->len)) {
	    return 0 ;
	}
    }
    return  new_STRING(form->specs->form) ;
}
@ %def check_convfmt

Setting this to [[1]], turns off error messages in [[parse_form()]].

<<static data>>=
static int checking_convfmt = 0 ;
@ %def checking_convfmt


<<[[printf]] functions>>=
static void c_or_rt_error_or_silence(const char* format, ...)
{
    if (! checking_convfmt) {
	/* up to caller not to exceed this buffer */
	char buffer[1024] ;
	va_list args ;

	va_start(args,format) ;
	vsprintf(buffer, format, args) ;
	if (mawk_state == EXECUTION) {
	    rt_error(buffer) ;
	}
	else {
	    compile_error(buffer) ;
	}
    }
}
@ %def c_or_rt_error_or_silence

<<static prototypes>>=
static void c_or_rt_error_or_silence(const char*,...) ;


@
\section{Source Files}

<<"printf.h">>=
/* printf.h */
<<blurb>>
#ifndef PRINTF_H
#define PRINTF_H 1
#include "mawk.h"
#include "types.h"
#include "memory.h"
#include "files.h"

<<[[printf]] definitions>>
#endif /* PRINTF_H */

<<"printf.c">>=
/* printf.c */
<<blurb>>

#include <stdarg.h>
#include "mawk.h"
#include "scan.h"
#include "printf.h"
#include "int.h"

<<[[form]] definitions and data>>
<<static prototypes>>
<<static data>>
<<[[printf]] functions>>

<<blurb>>=
/*
copyright 2016 Michael D. Brennan

This is a source file for mawk, an implementation of
the AWK programming language.

Mawk is distributed without warranty under the terms of
the GNU General Public License, version 3, 2007.

printf.c and printf.h were generated with the commands

   notangle -R'"printf.c"' printf.w > printf.c 
   notangle -R'"printf.h"' printf.w > printf.h 

Notangle is part of Norman Ramsey's noweb literate programming package.
Noweb home page: http://www.cs.tufts.edu/~nr/noweb/

It's easiest to read or modify this file by working with printf.w.
*/

@
\idindex