Extended sprintf implementation (for the browser and nodejs). Emphasis on compliance, performance and IE6+ support.

PRINTJ.sprintf("Hello %s!", "World");

A self-contained specification of the printf format string is included below in this README, as well as a summary of the support against various printf implementations

With npm:

$ npm install printj

In the browser:

<script src="printj.js"></script>

The browser exposes a variable PRINTJ

When installed globally, npm installs a script printj that renders the format string with the given arguments. Running the script with -h displays help.

The script will manipulate module.exports if available. This is not always desirable. To prevent the behavior, define DO_NOT_EXPORT_PRINTJ

The bundle ships with a printj.mjs script that acts as an ES Module.

NodeJS

NodeJS 12+ support ES modules. The default import uses the CommonJS script:

import PRINTJ from "printj";

It is possible to use the ESM powered script referencing printj.mjs directly:

import * as PRINTJ from "printj/printj.mjs"; // pull all exports
import { sprintf } from "printj/printj.mjs"; // pull `sprintf`

Browser Module Support

Chrome 61+ and Safari 11+ support module imports in the web browser. The .mjs script can be imported from a script type=module block:

<script type="module">
import { sprintf } from './printj.mjs';
console.log(sprintf("%02hhx", 123));
</script>

In all cases, the relevant function takes a format and arguments to be rendered.

The return value is a JS string.

• PRINTJ.sprintf(format, ...args) assumes the arguments are passed directly

• PRINTJ.vsprintf(format, argv) assumes the arguments are passed in an array

For example:

> // var PRINTJ = require('printj');       // uncomment this line if in node
> var sprintf = PRINTJ.sprintf, vsprintf = PRINTJ.vsprintf;
> sprintf("Hello %s", "SheetJS")           // 'Hello SheetJS'
> sprintf("%d + %d = %d", 2,3,2+3)         // '2 + 3 = 5'
> vsprintf("%d + %d = %d", [2,3,5])        // '2 + 3 = 5'
> sprintf("%1$02hhx %1$u %1$i %1$o", -69)  // 'bb 4294967227 -69 37777777673'

The command line script takes a format and arguments:

usage: printj [options] <format> [args...]

Options:
    -h, --help      output usage information
    -d, --dump      print debug information about format string

Arguments are treated as strings unless prefaced by a type indicator:
    n:<integer>     call parseInt (ex. n:3 -> 3)
    f:<float>       call parseFloat (ex. f:3.1 -> 3.1)
    b:<boolean>     false when lowercase value is "FALSE" or "0", else true
    s:<string>      interpret as string (ex. s:n:3 -> "n:3")
    j:<JSON>        interpret as an object using JSON.parse
    e:<JS>          evaluate argument (ex. e:1+1 -> 2, e:"1"+1 -> "11")

samples:
    $ printj '|%02hhx%d|' n:50 e:0x7B                # |32123|
    $ printj '|%2$d + %3$d is %1$d|' e:1+2 n:1 n:2   # |1 + 2 is 3|
    $ printj '|%s is %s|' s:1+2 e:1+2                # |1+2 is 3|
    $ printj '|%c %c|' s:69 n:69                     # |6 E|

make test will run the nodejs-based test.

make stress will run a larger test encompassing every possible conversion. It requires access to a C compiler.

Please consult the attached LICENSE file for details. All rights not explicitly granted by the Apache 2.0 license are reserved by the Original Author.

The printf family of functions attempt to generate and output a string of characters from a series of arguments, following a user-supplied "format string" specification. The format string contains normal characters that are written to the output string as well as specifiers that describe which parameter to insert and how to render the parameter. This specification describes how a conformant implementation should process the format string and generate an output string. Any discrepancies between this document and the reference implementation are considered bugs in the implementation.

Every function in the printf family follows the same logic to generate strings but have different interfaces reflecting different input and output behaviors. Some functions have wide variants that use wide wchar_t * strings rather than normal C char *. The following variants are required by the POSIX spec:

Each function has a dual function, whose name begins with v, that accepts the parameters as a va_list rather than formal parameters. In all cases, they return the number of characters written or a negative value to indicate error:

int   sprintf(char *ostr, const char *fmt, ...);
int  vsprintf(char *ostr, const char *fmt, va_list arg_list);

int  swprintf(wchar_t *ostr, const wchar_t *fmt, ...);
int vswprintf(wchar_t *ostr, const wchar_t *fmt, va_list arg_list);

C "strings" are really just arrays of numbers. An external code page (such as ASCII) maps those numbers to characters. K&R defines two types of strings: basic character set strings (char *) and extended character set strings (wchar_t *). In contrast, JS has a true string value type.

Unlike in C, JS strings do not treat the null character as an end-of-string marker. As a result, characters beyond the first null character will be used.

The JS equivalent of a C extended string would be an array of the individual character codes. The C basic string equivalent would involve specifying a code page and mapping back. The codepage JS library supports common codepages.

While capturing the essence of C strings, using arrays of character codes is not idiomatic JS. Few developers leverage this and the downsides far exceed the benefits of a more direct translation. The effect can be feigned, as shown in the js2c code sample at the end of the document.

In the absence of a standard output or even a standard concept of a stream, the non-string outputs are irrelevant. Similarly there is no JS analogue of wide characters. While useful, lack of direct memory management obviates snprintf. This implementation exports the remaining functions, sprintf and vsprintf.

Instead of replicating the original C signature and errno, functions directly return the output string and throw Errors:

function  sprintf(fmt:string, ...args):string;
function vsprintf(fmt:string, args:Array<any>):string;

The C functions return the number of characters written to the string, which is directly accessible in JS via the length property. A direct replica of the various string functions are included at the end of the document.

Note: The regular expressions follow Perl /x style. Whitespace characters outside of character classes are ignored. # is a comment character and every character until the end of the line is ignored. To convert to a standard regex:

regex_string.replace(/#.*$/gm,"").replace(/^\s*/gm,"").replace(/\s*\n/gm,"");

Based on K&R, conversions originally followed the format:

• required: leading %
• optional: - (POSIX refers to this as the "flags")
• optional: positive number or * (POSIX "width")
• optional: period followed by positive number or * (POSIX "precision")
• optional: an h or l to indicate size of data (POSIX "length")
• required: character describing output behavior (POSIX "conversion specifier")

This is captured by the regular expression:

/%(?:
    ([-])?                             # flags (only minus sign)
    (\d+|\*)?                          # width
    (?:\.(\d+|\*))?                    # period + precision
    ([hl])?                            # length
    ([dioxXucsfeEgGp%])                # conversion specifier
)/x

Various implementations of printf have added different functionality.

ANSI standards up through C99:

• more flags "+" " " "0" "#"
• more lengths "L" "hh" "ll" "j" "z" "t"
• more conversions "F" "a" "A" "n"

The POSIX specification of printf added:

• positional parameters to identify argument indices
• more flags "'"
• more conversions "C" "S"
• clarifications regarding corner cases and "undefined behavior"

BSD implementations added:

• more lengths "q"
• more conversions "D" "U" "O"

glibc (GNU) added:

• more lengths "Z"
• more conversions "m"

Windows C Runtime (CRT) added:

• more lengths "I" "I32" "I64" "w"

glibc and CRT both added Z. glibc uses Z for the length size_t. CRT uses Z as a conversion for length-prefixed strings. This implementation takes the former approach, handling Z in the same way as z.

BSD and IBM C library both added D. BSD uses D as a conversion, namely as an alias of ld. IBM uses D for the length for _Decimal64, a decimal floating point type, in accordance with ISO/IEC TR 24732. This implementation takes the former approach.

This implementation also adds new conversions:

• "b" and "B" for binary (base-2) integer renderings
• "y" and "Y" for true/false and yes/no Boolean conversions
• "J" for JSON
• "T" and "V" for JS typeof and valueOf inspection

Combining the various extensions yields the following regular expression:

/%(?:
    %|                                  # literal %% (flags etc prohibited)
    ([1-9]\d*\$)?                       # positional parameter
    ([-+ 0\x23\x27]*)?                  # flags
    ([1-9]\d*|\*(?:[1-9]\d*\$)?)?       # width
    (?:\.(\d+|\*(?:[1-9]\d*\$)?))?      # precision
    (hh?|ll?|[LzjtqZIw])?               # length
    ([diouxXfFeEgGaAcCsSpnDUOmbByYJVT]) # conversion specifier
)/x

This implementation explicitly does not support certain non-standard extensions:

• AltiVec vector length extensions (v with h/l/ll):
• CRT fixed width lengths I32 and I64

The default behavior is to consume arguments in order:

printf("Count to 3: %d %d %d", 1, 2, 3); // Count to 3: 1 2 3

POSIX printf permits explicit argument selection, bypassing the standard behavior of consuming arguments in order. To specify the argument at position n, use n$ immediately after the % token:

printf("%d %d %d",       1, 2, 3);        // 1 2 3 (implicit order 1, 2, 3 )
printf("%1$s %2$s %3$s", "a", "b", "c");  // a b c (explicit order 1, 2, 3 )
printf("%1$s %3$s %2$s", "a", "b", "c");  // a c b (explicit order 1, 3, 2 )

The POSIX standard asserts that mixing positional and non-positional conversions is undefined behavior. This implementation handles mixing by tracking the index for non-positional conversions:

printf("%s %4$s %s %5$s %s", "a", "b", "c", "d", "e"); // a d b e c

The POSIX standard requires that if an argument is used in the format, every preceding argument must be used. This implementation relaxes that requirement:

printf("%3$s", "a", "b", "c"); // c (technically invalid since "a"/"b" unused)

The width and precision specifiers may include the dynamic specifier * which instructs the engine to read the next argument (assumed to be an integer). Just as with the positional parameter, idx$ immediately after the * token selects the numeric argument.

For example:

printf("|%5s|", "sheetjs");               // |sheetjs|    (width = 5)
printf("|%*s|", 5, "sheetjs");            // |sheetjs|    (width first argument)
printf("|%2$*1$s|", 5, "sheetjs", 10);    // |sheetjs|    (width is argument #1)

printf("|%10s|", "sheetjs");              // |   sheetjs| (width = 10)
printf("|%2$*3$s|", 5, "sheetjs", 10);    // |   sheetjs| (width is argument #3)

Arguments are generally consumed in order as presented in the format string:

printf("|%s|", val);
printf("|%*s|", width, val);
printf("|%.*s|", prec, val);
printf("|%*.*s|", width, prec, val);
printf("|%0*.*d|", 4, 2, 1);  // |  01| width=4 prec=2 value=1

Positional arguments can be applied to width and precision:

printf("|%*.*d|", width, prec, val);
printf("|%2$0*3$.*1$d|", prec, val, width);
printf("|%0*.*d|", 4, 2, 1);        // |  01| width=4 prec=2 value=1 flags='0'
printf("|%1$0*3$.*2$d|", 1, 2, 4);  // |  01| width=4 prec=2 value=1 flags='0'

A negative width is interpreted as the - flag with a positive width:

printf("|%*.*d|",   4, 2, 1);        // |  01| width=4 prec=2 value=1 flags=''
printf("|%-*.*d|",  4, 2, 1);        // |01  | width=4 prec=2 value=1 flags='-'
printf("|%*.*d|",  -4, 2, 1);        // |01  | width=4 prec=2 value=1 flags='-'
printf("|%-*.*d|", -4, 2, 1);        // |01  | width=4 prec=2 value=1 flags='-'

A negative precision is discarded:

printf("|%*s|", 4, "sheetjs");       // |sheetjs|  width=4
printf("|%*.*s|", 4,  3, "sheetjs"); // | she|     width=4 prec=3
printf("|%*.*s|", 4,  2, "sheetjs"); // |  sh|     width=4 prec=2
printf("|%*.*s|", 4,  1, "sheetjs"); // |   s|     width=4 prec=1
printf("|%*.*s|", 4,  0, "sheetjs"); // |    |     width=4 prec=0
printf("|%*.*s|", 4, -1, "sheetjs"); // |sheetjs|  width=4 (prec ignored)

JS has one numeric type Number which represents an IEEE754 double-precision (64-bit) floating point number. C has a multitude of numeric types, including floating point as well as integer types. The sizes of those data types are implementation-dependent. A "C data model" specifies the sizes of the core C data types.

POSIX printf specification references 8 integer types in integer conversions:

C99 does not officially define a signed size_t or unsigned ptrdiff_t type. POSIX does define ssize_t but no equivalent uptrdiff_t.

BSD additionally recognizes the types quad_t and u_quad_t, which this implementation treats as long long int and unsigned long long int.

Two integer types are used in character and string conversions:

Both wide types wchar_t and wint_t can be signed or unsigned according to C99. Both types are used only in character and string conversions. Based on K&R "printable characters are always positive", the types are assumed unsigned.

K&R recognizes 3 floating point types. C99 later tied it to IEC 60559:

Numerous "C data models", specifying the bit/byte sizes of the various types, have been and continue to be used. For example, OSX and other modern 64-bit UNIX flavors use the "LP64" C data model. 64-bit Windows currently uses the "LLP64" model. 32-bit systems generally use the "ILP32" model. The 8-bit byte sizes for the data types under the various models are defined in ctypes.json in the Models object as per the following table:

By default the source assumes the LP64 data model. Other data models are supported in the source tree, controlled by the JSFLAGS variable in the build process. Set the JS_MODEL variable to the desired index as specified in the ModelNames array in bits/ctype.json:

$ <bits/ctypes.json jq -r '.ModelNames|.[]'  # LP64 ILP32 LLP64
$ JSFLAGS=-DJS_MODEL=0 make                  # LP64
$ JSFLAGS=-DJS_MODEL=1 make                  # ILP32
$ JSFLAGS=-DJS_MODEL=2 make                  # LLP64

To create a custom model, add the spec to bits/ctypes.json by appending the model name to the end of the ModelNames array and adding an entry to the Models object. The current models are defined as follows:

{
  "ModelNames":["LP64", "ILP32", "LLP64"],
  "Models": {
    "LP64":  { "char":1, "short":2, "int":4, "long":8, "longlong":8, "wint_t":4, "wchar_t":4, "size_t":8, "intmax_t":8, "ptrdiff_t":8 },
    "ILP32": { "char":1, "short":2, "int":4, "long":4, "longlong":8, "wint_t":4, "wchar_t":4, "size_t":4, "intmax_t":8, "ptrdiff_t":4 },
    "LLP64": { "char":1, "short":2, "int":4, "long":4, "longlong":8, "wint_t":2, "wchar_t":2, "size_t":8, "intmax_t":8, "ptrdiff_t":8 }
  }
}

This section covers the conversions diouxXDUO. The base-2 conversions bB are an extension and are discussed at the end, but the same basic rules apply.

JS has one Number type (representing an IEEE754 8-byte floating point number) that is capable of representing a 32-bit integer. It cannot represent the full range of 64-bit integers exactly. Care is taken to avoid operations that may inadvertently result in a conversion to a smaller integral type.

JS Bitwise operations convert numbers to 32-bit integers before performing operations. With the exception of the unsigned right shift operator >>>, all operations act on signed integers. For example:

Math.pow(2,31) | 0;        // -2147483648 == -Math.pow(2,31)
(Math.pow(2,32)-2) ^ 0;    // -2
-1 >>> 0                   // 4294967295 == Math.pow(2,32) - 1

JS Number can exactly represent every integer in the range -2^53 .. 2^53. For lengths exceeding 32 bits, Math.round is appropriate.

When a length specifier implies a certain size (such as hh for a single-byte integer), the number will be converted before rendering strings. For example:

printf("%1$02hhx %1$02hx %1$02lx %1$02llx", 256);       // |00 100 100 100|
printf("%1$02hhx %1$02hx %1$02lx %1$02llx", 4096);      // |00 1000 1000 1000|
printf("%1$02hhx %1$02hx %1$02lx %1$02llx", 65536);     // |00 00 10000 10000|

Values are restricted by first limiting the result to a specified number of bytes (appropriate bit-and) and then adding or subtracting to ensure the value is signed or unsigned according to the conversion specifier. If a length is specified, it overrides the implied length of the conversion. The following table describes the behavior of this implementation:

num.toString(10) produces the correct result for exact integers.

"u" conversion restricts values to int; "U" restricts to long.

Even though num.toString(8) is implementation-dependent, all browser implementations use standard form for integers in the exact range.

The alternate form (#) prints a "0" prefix.

"o" conversion restricts values to int; "O" restricts to long.

Even though num.toString(16) is implementation-dependent, all browser implementations use standard form for integers in the exact range.

The alternate form (#) prints a "0x" or "0X" prefix.

Unlike "U" "O" "D", "X" conversion uses A-F instead of a-f in hex.

num.toString(10) produces the correct result for exact integers. The flags " +" control prefixes for positive integers.

"di" conversions restrict values to int; "D" restricts to long.

This section covers the conversions fFeEgGaA.

Due to C variadic argument promotion rules, float types are always promoted to double. None of the conversions or length specifiers signal that an argument is to be interpreted as a float. There is no JS canonical representation of an extended floating point number, so JS Number suffices.

JS recognizes a few special IEEE754 values, as described in the following table:

JS Number methods render different strings from the POSIX spec:

This implementation performs the required adjustments.

Aside from the special cases discussed above, JS num.toExponential(prec) differs from POSIX printf("%1$.*2$e", num, prec) in the exponent field: JS writes exponents with the fewest digits (POSIX requires 2+ digits). This is addressed by inspecting the output string and inserting a "0" when needed.

The optional # flag forces the decimal point to appear when precision is 0. This is also fixed by adding a decimal point just before the "e".

The POSIX spec only requires that the number of digits after the decimal point is equal to the precision. It does not specify how many digits appear before the decimal point, nor does it specify how to handle numbers that cannot be exactly represented.

For values less than 1e21 the JS num.toFixed(n) generally matches %f with the specified precision. However, for larger values toFixed defaults to the exponential form.

The final form (exponential or standard) is determined based on the value. The threshold is different from the JS toString / toPrecision thresholds and depends on the specified precision as well as the base-10 exponent:

According to JS spec, toPrecision uses standard form when precision > E and E >= -6. For printf standard form is used when precision > E and E >= -4.

A general exponential form involves 3 parameters: radix of the mantissa, base of the exponent expression, and radix of the exponent expression. The standard exponential form uses decimal for all three parts. For base 16, there are quite a few reasonable combinations. Consider the value 1.234567e-80:

POSIX "%a" uses a hex mantissa (16), decimal exponent radix (10), and binary exponent base (2). The general normalized form requires that the integral part of the mantissa to exceed 0 and not to exceed exponent base - 1 except in the special case of 0. The sigil is p and exponent sign is always used.

JS num.toString(radix) is implementation-dependent for radices other than 10 (2-9, 11-36). IE uses hex-mantissa decimal-hex-exponent form when the absolute value of the base-2 exponent exceeds 60. Otherwise, IE uses an exact standard hexadecimal form. Chrome, Safari and other browsers always use the exact standard hexadecimal form. Both forms are converted to "%a" by calculating and dividing by the appropriate power of 2.

For each non-zero normal floating point value, there are 4 acceptable strings that represent the value, derived by multiplying the normalized value by powers of 2 and adjusting the exponent accordingly:

JS engines follow the glibc model: multiply by a suitable power of 16 so that the mantissa is between 1 and 16, render left to right one digit at a time, then fix the result at the end. FreeBSD and OSX always show the normalized form. This implementation defaults to the normalized form. To switch to the glibc form, define DO_NOT_NORMALIZE in the JSFLAGS variable when building:

$ JSFLAGS=-DDO_NOT_NORMALIZE make

This section covers the conversions sScC.

JS has no concept of "wide strings" (wchar_t * in C), so the length modifiers are ignored. s and S are treated as equivalent.

Arguments are first interpreted as strings by calling the String function. Implementing toString on the argument to be converted may lead to unexpected results:

var O = {valueOf:function() {return 456;}, toString:function() {return "123"}};
printf("%1$s %1$d", O); // "123 456"

If a positive precision is specified, up to that many characters will be taken from the string. Otherwise the entire string will be used:

printf("|%s|", "sheetjs");    // '|sheetjs|' (no precision)
printf("|%.9s|", "sheetjs");  // '|sheetjs|' (string shorter than precision)
printf("|%.5s|", "sheetjs");  // '|sheet|'   (string truncated)

Lengths are measured using the JS string length accessor. Since there is no attempt to correct for multi-character sequences like combining marks, the results may be unexpected:

printf("%.1s","ñ");  // 'n' not "ñ"

If the width is specified and is greater than the width of the string to be rendered, padding will be applied. If the "-" flag is specified, then the string will be right-padded, otherwise it will be left-padded. If the "0" flag is specified, the final string is left-padded with zeroes. The "-" flag takes precedence over 0.

printf(   "|%s|", "sheetjs");   // '|sheetjs|'   (no width)
printf(  "|%5s|", "sheetjs");   // '|sheetjs|'   (string longer than width)
printf(  "|%9s|", "sheetjs");   // '|  sheetjs|' (no flag = left pad spaces)
printf( "|%09s|", "sheetjs");   // '|00sheetjs|' ("0" = left pad "0")
printf( "|%-9s|", "sheetjs");   // '|sheetjs  |' ("-" = right pad space)
printf("|%-09s|", "sheetjs");   // '|sheetjs  |' ("0" ignored)

JS has no concept of "wide characters" (wchar_t in C). The length modifier is used in determining whether the number should be interpreted as one or two 16-bit character codes (when the "C" format or the "l" or "ll" specifiers are used) or a single 8-bit char code. Precision and flags are ignored.

All other parameters are ignored.

JS has no true concept of pointers. In array and typed array contexts, it is common to associate a position object that stores the address relative to the start of the array. This implementation reads the l key and interprets as a 32-bit or 52-bit unsigned integer depending on size_t in the data model.

The normal output format is equivalent to "%#x" but the alternate form emits using the "%d" format. When the pointer is invalid, -1 is rendered. Only the "#" flag is interpreted.

var x = {}, y = {l:3};
printf("%1$p %1$#p", y); // 0x3 3
printf("%1$p %1$#p", x); // 0xFFFFFFFF -1

C printf permits a special n conversion which interprets the argument as an integral pointer (interpreted size controlled by the length specifier) and writes the number of characters printed to that pointer.

JS has no true concept of pointers in the C sense. The library works around the limitation by interpreting the argument as an object and assigning to the len key. The conversion does not write any characters to the output string:

var x = {};
printf("%1$s %2$J%2$n abc", "foo", x); // "foo {} abc", also sets x.len = 6
//     |........|                         |......|  (6 chars at that point)

This implementation mutates the object while processing:

var x = {};
printf("%1$s %2$J%2$n %3$s %2$J", "foo", x, "bar"); // 'foo {} bar {"len":6}'

glibc supports an m conversion that does not consume arguments. It renders the string strerror(errno) where strerror is the libc function and errno is the global error number.

JS has no equivalent of errno and no standard JS runtime exposes a similar global error variable, so %m will write the default message "Success". A positional parameter or # flag changes the behavior:

In all forms other than "%m", an argument will be processed as follows:

• If the argument is not an instance of an Error, emit "Success"
• If the message field is set, emit the error message.
• If the errno field is set, emit "Error number " followed by the errno
• Otherwise emit "Error " followed by the error interpreted as a String

var x = new Error("sheetjs");
x.errno = 69; x.toString = function() { return "SHEETJS"; };
printf("|%#m|", x);      // |sheetjs|
delete x.message;
printf("|%#m|", x);      // |Error number 69|
delete x.errno;
printf("|%#m|", x);      // |Error SHEETJS|

These additional conversions take advantage of unused format characters:

Values are converted to Boolean and tested for truthiness. The Y rendering is the uppercase version of the equivalent rendering with format y.

Width and precision are applied in the same manner as the s conversion.

printf("|%1$y|%2$Y|%1$#Y|%2$#y|%2$.1y|", 1, 0); // |true|FALSE|YES|no|f|
printf("|%05.2Y|%-5.2y|", 1, 0);  // |000TR|fa   |

The default rendering is the standard output from JSON.stringify. Alternate form ("#" flag) renders using util.inspect if available.

var x = {
  a: [1,[2,3,4],5,6,7],
  b: {
    c: {
      d: { e:"f" },
      g:"h",
      i:"j"
    },
    k:"l",
    m:"n",
    o:"p"},
  q: "r"
};
printf("%J", x) // '{"a":[1,[2,3,4],5,6,7],"b":{"c":{"d":{"e":"f"}, ..(ctnd)..
printf("%#J", x) // '{ a: [ 1, [ 2, 3, 4 ], 5, 6, 7 ],\n  b: { c: { ..(ctnd)..

Width, precision and other flags are ignored.

Under the "T" conversion, the result of typeof arg is rendered. If the # flag is specified, the type is derived from Object.prototype.toString:

printf("%1$T %1$#T", 1);          // 'number Number'
printf("%1$T %1$#T", 'foo');      // 'string String'
printf("%1$T %1$#T", [1,2,3]);    // 'object Array'
printf("%1$T %1$#T", null);       // 'object Null'
printf("%1$T %1$#T", undefined);  // 'undefined Undefined'

Under the "V" conversion, the result of arg.valueOf() is rendered:

var _f = function() { return "f"; };
var _3 = function() { return 3; };
printf("%1$d %1$s %1$V", {toString:_f});               // '0 f f'
printf("%1$d %1$s %1$V", {valueOf:_3});                // '3 [object Object] 3'
printf("%1$d %1$s %1$V", {valueOf:_3, toString:_f});   // '3 f 3'

The implementation is similar to the octal "o" and "O" conversions, except for the radix (2 for "b" and "B") and the alternate-form prefix ("0b")

For compatibility purposes, format characters must be printable ASCII characters (ASCII codes 0x20 - 0x7E). The 95 eligible characters are listed below:

C provides no guidance on the actual character set. According to K&R all valid characters in source code must be in a character set that is a subset of the 7-bit ASCII set. This implementation falls back on the UTF-16 base required by JS. When converting C literal strings, there are a few differences in escaping:

Opera does not always include the last significant digit in base 16 rendering. For example, (-6.9e-11).toString(16) is "0.000000004bddc5fd160168" in every other browser but is "0.000000004bddc5fd16017" in Opera. The test suite skips the %a/%A precision-less formats in Opera.

Object.prototype.toString.call gives unexpected results in older browsers, and no attempt is made to correct for them. The test suite ignores those cases:

• Full POSIX conversion support with extensions! Conversion Specifier Table
• Full support for POSIX flags and positional parameters
• Emulation of BSD quad_t and u_quad_t conversion
• Parser accepts but does not emulate CRT wide and unicode character conversions
• glibc Z length conversion and extended m error support
• Parser fails on CRT I32/I64 fixed lengths
• Default LP64 data model but can be configured to support ILP32 or LLP64