mirrors/cpython
1
0
Fork 0
mirror of https://github.com/python/cpython.git synced 2025-08-04 17:08:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 13

Implementation of PEP 3101, Advanced String Formatting.

Browse source 
Known issues:

The string.Formatter class, as discussed in the PEP, is incomplete.

Error handling needs to conform to the PEP.

Need to fix this warning that I introduced in Python/formatter_unicode.c:
Objects/stringlib/unicodedefs.h:26: warning: `STRINGLIB_CMP' defined but not used

Need to make sure sign formatting is correct, more tests needed.

Need to remove '()' sign formatting, left over from an earlier version of the PEP.
This commit is contained in:
Eric Smith 2007-08-25 02:26:07 +00:00
parent e4dc324884
commit 8c66326368
22 changed files with 2669 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

966
Objects/stringlib/formatter.h Normal file
View file

@ -0,0 +1,966 @@
/* implements the string, long, and float formatters. that is,
string.__format__, etc. */
/* Before including this, you must include either:
stringlib/unicodedefs.h
stringlib/stringdefs.h
Also, you should define the names:
FORMAT_STRING
FORMAT_LONG
FORMAT_FLOAT
to be whatever you want the public names of these functions to
be. These are the only non-static functions defined here.
*/
/*
get_integer consumes 0 or more decimal digit characters from an
input string, updates *result with the corresponding positive
integer, and returns the number of digits consumed.
returns -1 on error.
*/
static int
get_integer(STRINGLIB_CHAR **ptr, STRINGLIB_CHAR *end,
Py_ssize_t *result)
{
Py_ssize_t accumulator, digitval, oldaccumulator;
int numdigits;
accumulator = numdigits = 0;
for (;;(*ptr)++, numdigits++) {
if (*ptr >= end)
break;
digitval = STRINGLIB_TODECIMAL(**ptr);
if (digitval < 0)
break;
/*
This trick was copied from old Unicode format code. It's cute,
but would really suck on an old machine with a slow divide
implementation. Fortunately, in the normal case we do not
expect too many digits.
*/
oldaccumulator = accumulator;
accumulator *= 10;
if ((accumulator+10)/10 != oldaccumulator+1) {
PyErr_Format(PyExc_ValueError,
"Too many decimal digits in format string");
return -1;
}
accumulator += digitval;
}
*result = accumulator;
return numdigits;
}
/************************************************************************/
/*********** standard format specifier parsing **************************/
/************************************************************************/
/* returns true if this character is a specifier alignment token */
Py_LOCAL_INLINE(int)
is_alignment_token(STRINGLIB_CHAR c)
{
switch (c) {
case '<': case '>': case '=': case '^':
return 1;
default:
return 0;
}
}
/* returns true if this character is a sign element */
Py_LOCAL_INLINE(int)
is_sign_element(STRINGLIB_CHAR c)
{
switch (c) {
case ' ': case '+': case '-': case '(':
return 1;
default:
return 0;
}
}
typedef struct {
STRINGLIB_CHAR fill_char;
STRINGLIB_CHAR align;
STRINGLIB_CHAR sign;
Py_ssize_t width;
Py_ssize_t precision;
STRINGLIB_CHAR type;
} InternalFormatSpec;
/*
ptr points to the start of the format_spec, end points just past its end.
fills in format with the parsed information.
returns 1 on success, 0 on failure.
if failure, sets the exception
*/
static int
parse_internal_render_format_spec(PyObject *format_spec,
InternalFormatSpec *format,
char default_type)
{
STRINGLIB_CHAR *ptr = STRINGLIB_STR(format_spec);
STRINGLIB_CHAR *end = ptr + STRINGLIB_LEN(format_spec);
/* end-ptr is used throughout this code to specify the length of
the input string */
Py_ssize_t specified_width;
format->fill_char = '\0';
format->align = '\0';
format->sign = '\0';
format->width = -1;
format->precision = -1;
format->type = default_type;
/* If the second char is an alignment token,
then parse the fill char */
if (end-ptr >= 2 && is_alignment_token(ptr[1])) {
format->align = ptr[1];
format->fill_char = ptr[0];
ptr += 2;
} else if (end-ptr >= 1 && is_alignment_token(ptr[0])) {
format->align = ptr[0];
ptr++;
}
/* Parse the various sign options */
if (end-ptr >= 1 && is_sign_element(ptr[0])) {
format->sign = ptr[0];
ptr++;
if (end-ptr >= 1 && ptr[0] == ')') {
ptr++;
}
}
/* The special case for 0-padding (backwards compat) */
if (format->fill_char == '\0' &&
end-ptr >= 1 && ptr[0] == '0') {
format->fill_char = '0';
if (format->align == '\0') {
format->align = '=';
}
ptr++;
}
/* XXX add error checking */
specified_width = get_integer(&ptr, end, &format->width);
/* if specified_width is 0, we didn't consume any characters for
the width. in that case, reset the width to -1, because
get_integer() will have set it to zero */
if (specified_width == 0) {
format->width = -1;
}
/* Parse field precision */
if (end-ptr && ptr[0] == '.') {
ptr++;
/* XXX add error checking */
specified_width = get_integer(&ptr, end, &format->precision);
/* not having a precision after a dot is an error */
if (specified_width == 0) {
PyErr_Format(PyExc_ValueError,
"Format specifier missing precision");
return 0;
}
}
/* Finally, parse the type field */
if (end-ptr > 1) {
/* invalid conversion spec */
PyErr_Format(PyExc_ValueError, "Invalid conversion specification");
return 0;
}
if (end-ptr == 1) {
format->type = ptr[0];
ptr++;
}
return 1;
}
/************************************************************************/
/*********** common routines for numeric formatting *********************/
/************************************************************************/
/* describes the layout for an integer, see the comment in
_calc_integer_widths() for details */
typedef struct {
Py_ssize_t n_lpadding;
Py_ssize_t n_spadding;
Py_ssize_t n_rpadding;
char lsign;
Py_ssize_t n_lsign;
char rsign;
Py_ssize_t n_rsign;
Py_ssize_t n_total; /* just a convenience, it's derivable from the
other fields */
} NumberFieldWidths;
/* not all fields of format are used. for example, precision is
unused. should this take discrete params in order to be more clear
about what it does? or is passing a single format parameter easier
and more efficient enough to justify a little obfuscation? */
static void
calc_number_widths(NumberFieldWidths *r, STRINGLIB_CHAR actual_sign,
Py_ssize_t n_digits, const InternalFormatSpec *format)
{
r->n_lpadding = 0;
r->n_spadding = 0;
r->n_rpadding = 0;
r->lsign = '\0';
r->n_lsign = 0;
r->rsign = '\0';
r->n_rsign = 0;
/* the output will look like:
| |
| <lpadding> <lsign> <spadding> <digits> <rsign> <rpadding> |
| |
lsign and rsign are computed from format->sign and the actual
sign of the number
digits is already known
the total width is either given, or computed from the
actual digits
only one of lpadding, spadding, and rpadding can be non-zero,
and it's calculated from the width and other fields
*/
/* compute the various parts we're going to write */
if (format->sign == '+') {
/* always put a + or - */
r->n_lsign = 1;
r->lsign = (actual_sign == '-' ? '-' : '+');
} else if (format->sign == '(') {
if (actual_sign == '-') {
r->n_lsign = 1;
r->lsign = '(';
r->n_rsign = 1;
r->rsign = ')';
}
} else if (format->sign == ' ') {
r->n_lsign = 1;
r->lsign = (actual_sign == '-' ? '-' : ' ');
} else {
/* non specified, or the default (-) */
if (actual_sign == '-') {
r->n_lsign = 1;
r->lsign = '-';
}
}
/* now the number of padding characters */
if (format->width == -1) {
/* no padding at all, nothing to do */
} else {
/* see if any padding is needed */
if (r->n_lsign + n_digits + r->n_rsign >= format->width) {
/* no padding needed, we're already bigger than the
requested width */
} else {
/* determine which of left, space, or right padding is
needed */
Py_ssize_t padding = format->width - (r->n_lsign + n_digits + r->n_rsign);
if (format->align == '<')
r->n_rpadding = padding;
else if (format->align == '>')
r->n_lpadding = padding;
else if (format->align == '^') {
r->n_lpadding = padding / 2;
r->n_rpadding = padding - r->n_lpadding;
} else
/* must be '=' */
r->n_spadding = padding;
}
}
r->n_total = r->n_lpadding + r->n_lsign + r->n_spadding +
n_digits + r->n_rsign + r->n_rpadding;
}
/* fill in the non-digit parts of a numbers's string representation,
as determined in _calc_integer_widths(). returns the pointer to
where the digits go. */
static STRINGLIB_CHAR *
fill_number(STRINGLIB_CHAR *p_buf, const NumberFieldWidths *spec,
Py_ssize_t n_digits, STRINGLIB_CHAR fill_char)
{
STRINGLIB_CHAR* p_digits;
if (spec->n_lpadding) {
STRINGLIB_FILL(p_buf, fill_char, spec->n_lpadding);
p_buf += spec->n_lpadding;
}
if (spec->n_lsign == 1) {
*p_buf++ = spec->lsign;
}
if (spec->n_spadding) {
STRINGLIB_FILL(p_buf, fill_char, spec->n_spadding);
p_buf += spec->n_spadding;
}
p_digits = p_buf;
p_buf += n_digits;
if (spec->n_rsign == 1) {
*p_buf++ = spec->rsign;
}
if (spec->n_rpadding) {
STRINGLIB_FILL(p_buf, fill_char, spec->n_rpadding);
p_buf += spec->n_rpadding;
}
return p_digits;
}
/************************************************************************/
/*********** string formatting ******************************************/
/************************************************************************/
static PyObject *
format_string_internal(PyObject *value, const InternalFormatSpec *format)
{
Py_ssize_t width; /* total field width */
Py_ssize_t lpad;
STRINGLIB_CHAR *dst;
STRINGLIB_CHAR *src = STRINGLIB_STR(value);
Py_ssize_t len = STRINGLIB_LEN(value);
PyObject *result = NULL;
/* sign is not allowed on strings */
if (format->sign != '\0') {
PyErr_SetString(PyExc_ValueError,
"Sign not allowed in string format specifier");
goto done;
}
/* '=' alignment not allowed on strings */
if (format->align == '=') {
PyErr_SetString(PyExc_ValueError,
"'=' alignment not allowed "
"in string format specifier");
goto done;
}
/* if precision is specified, output no more that format.precision
characters */
if (format->precision >= 0 && len >= format->precision) {
len = format->precision;
}
if (format->width >= 0) {
width = format->width;
/* but use at least len characters */
if (len > width) {
width = len;
}
} else {
/* not specified, use all of the chars and no more */
width = len;
}
/* allocate the resulting string */
result = STRINGLIB_NEW(NULL, width);
if (result == NULL)
goto done;
/* now write into that space */
dst = STRINGLIB_STR(result);
/* figure out how much leading space we need, based on the
aligning */
if (format->align == '>')
lpad = width - len;
else if (format->align == '^')
lpad = (width - len) / 2;
else
lpad = 0;
/* if right aligning, increment the destination allow space on the
left */
memcpy(dst + lpad, src, len * sizeof(STRINGLIB_CHAR));
/* do any padding */
if (width > len) {
STRINGLIB_CHAR fill_char = format->fill_char;
if (fill_char == '\0') {
/* use the default, if not specified */
fill_char = ' ';
}
/* pad on left */
if (lpad)
STRINGLIB_FILL(dst, fill_char, lpad);
/* pad on right */
if (width - len - lpad)
STRINGLIB_FILL(dst + len + lpad, fill_char, width - len - lpad);
}
done:
return result;
}
/************************************************************************/
/*********** long formatting ********************************************/
/************************************************************************/
static PyObject *
format_long_internal(PyObject *value, const InternalFormatSpec *format)
{
PyObject *result = NULL;
int total_leading_chars_to_skip = 0; /* also includes sign, if
present */
STRINGLIB_CHAR sign = '\0';
STRINGLIB_CHAR *p;
Py_ssize_t n_digits; /* count of digits need from the computed
string */
Py_ssize_t len;
Py_ssize_t tmp;
NumberFieldWidths spec;
long x;
/* no precision allowed on integers */
if (format->precision != -1) {
PyErr_SetString(PyExc_ValueError,
"Precision not allowed in integer format specifier");
goto done;
}
/* special case for character formatting */
if (format->type == 'c') {
/* error to specify a sign */
if (format->sign != '\0') {
PyErr_SetString(PyExc_ValueError,
"Sign not allowed with integer"
" format specifier 'c'");
goto done;
}
/* taken from unicodeobject.c formatchar() */
/* Integer input truncated to a character */
x = PyInt_AsLong(value);
if (x == -1 && PyErr_Occurred())
goto done;
#ifdef Py_UNICODE_WIDE
if (x < 0 || x > 0x10ffff) {
PyErr_SetString(PyExc_OverflowError,
"%c arg not in range(0x110000) "
"(wide Python build)");
goto done;
}
#else
if (x < 0 || x > 0xffff) {
PyErr_SetString(PyExc_OverflowError,
"%c arg not in range(0x10000) "
"(narrow Python build)");
goto done;
}
#endif
result = STRINGLIB_NEW(NULL, 1);
if (result == NULL)
goto done;
p = STRINGLIB_STR(result);
p[0] = (Py_UNICODE) x;
n_digits = len = 1;
} else {
int base;
int format_leading_chars_to_skip; /* characters added by
PyNumber_ToBase that we
want to skip over.
instead of using them,
we'll compute our
own. */
/* compute the base and how many characters will be added by
PyNumber_ToBase */
switch (format->type) {
case 'b':
base = 2;
format_leading_chars_to_skip = 2; /* 0b */
break;
case 'o':
base = 8;
format_leading_chars_to_skip = 2; /* 0o */
break;
case 'x':
case 'X':
base = 16;
format_leading_chars_to_skip = 2; /* 0x */
break;
default: /* shouldn't be needed, but stops a compiler warning */
case 'd':
base = 10;
format_leading_chars_to_skip = 0;
break;
}
/* do the hard part, converting to a string in a given base */
result = PyNumber_ToBase(value, base);
if (result == NULL)
goto done;
n_digits = STRINGLIB_LEN(result);
len = n_digits;
p = STRINGLIB_STR(result);
/* if X, convert to uppercase */
if (format->type == 'X')
for (tmp = 0; tmp < len; tmp++)
p[tmp] = STRINGLIB_TOUPPER(p[tmp]);
/* is a sign character present in the output? if so, remember it
and skip it */
sign = p[0];
if (sign == '-') {
total_leading_chars_to_skip += 1;
n_digits--;
}
/* skip over the leading digits (0x, 0b, etc.) */
assert(n_digits >= format_leading_chars_to_skip + 1);
n_digits -= format_leading_chars_to_skip;
total_leading_chars_to_skip += format_leading_chars_to_skip;
}
calc_number_widths(&spec, sign, n_digits, format);
/* if the buffer is getting bigger, realloc it. if it's getting
smaller, don't realloc because we need to move the results
around first. realloc after we've done that */
if (spec.n_total > len) {
if (STRINGLIB_RESIZE(&result, spec.n_total) < 0)
goto done;
/* recalc, because string might have moved */
p = STRINGLIB_STR(result);
}
/* copy the characters into position first, since we're going to
overwrite some of that space */
/* we need to move if the number of left padding in the output is
different from the number of characters we need to skip */
if ((spec.n_lpadding + spec.n_lsign + spec.n_spadding) !=
total_leading_chars_to_skip) {
memmove(p + (spec.n_lpadding + spec.n_lsign + spec.n_spadding),
p + total_leading_chars_to_skip,
n_digits * sizeof(STRINGLIB_CHAR));
}
/* now fill in the non-digit parts */
fill_number(p, &spec, n_digits,
format->fill_char == '\0' ? ' ' : format->fill_char);
/* if we're getting smaller, realloc now */
if (spec.n_total < len) {
if (STRINGLIB_RESIZE(&result, spec.n_total) < 0)
goto done;
}
done:
return result;
}
/************************************************************************/
/*********** float formatting *******************************************/
/************************************************************************/
/* taken from unicodeobject.c */
static Py_ssize_t
strtounicode(Py_UNICODE *buffer, const char *charbuffer)
{
register Py_ssize_t i;
Py_ssize_t len = strlen(charbuffer);
for (i = len - 1; i >= 0; i--)
buffer[i] = (Py_UNICODE) charbuffer[i];
return len;
}
/* the callback function to call to do the actual float formatting.
it matches the definition of PyOS_ascii_formatd */
typedef char*
(*DoubleSnprintfFunction)(char *buffer, size_t buf_len,
const char *format, double d);
/* just a wrapper to make PyOS_snprintf look like DoubleSnprintfFunction */
static char*
snprintf_double(char *buffer, size_t buf_len, const char *format, double d)
{
PyOS_snprintf(buffer, buf_len, format, d);
return NULL;
}
/* see FORMATBUFLEN in unicodeobject.c */
#define FLOAT_FORMATBUFLEN 120
/* much of this is taken from unicodeobject.c */
/* use type instead of format->type, so that it can be overridden by
format_number() */
static PyObject *
_format_float(STRINGLIB_CHAR type, PyObject *value,
const InternalFormatSpec *format,
DoubleSnprintfFunction snprintf)
{
/* fmt = '%.' + `prec` + `type` + '%%'
worst case length = 2 + 10 (len of INT_MAX) + 1 + 2 = 15 (use 20)*/
char fmt[20];
/* taken from unicodeobject.c */
/* Worst case length calc to ensure no buffer overrun:
'g' formats:
fmt = %#.<prec>g
buf = '-' + [0-9]*prec + '.' + 'e+' + (longest exp
for any double rep.)
len = 1 + prec + 1 + 2 + 5 = 9 + prec
'f' formats:
buf = '-' + [0-9]*x + '.' + [0-9]*prec (with x < 50)
len = 1 + 50 + 1 + prec = 52 + prec
If prec=0 the effective precision is 1 (the leading digit is
always given), therefore increase the length by one.
*/
char charbuf[FLOAT_FORMATBUFLEN];
Py_ssize_t n_digits;
double x;
Py_ssize_t precision = format->precision;
PyObject *result = NULL;
STRINGLIB_CHAR sign;
char* trailing = "";
STRINGLIB_CHAR *p;
NumberFieldWidths spec;
#if STRINGLIB_IS_UNICODE
Py_UNICODE unicodebuf[FLOAT_FORMATBUFLEN];
#endif
/* first, do the conversion as 8-bit chars, using the platform's
snprintf. then, if needed, convert to unicode. */
/* 'F' is the same as 'f', per the PEP */
if (type == 'F')
type = 'f';
x = PyFloat_AsDouble(value);
if (x == -1.0 && PyErr_Occurred())
goto done;
if (type == '%') {
type = 'f';
x *= 100;
trailing = "%";
}
if (precision < 0)
precision = 6;
if (type == 'f' && (fabs(x) / 1e25) >= 1e25)
type = 'g';
/* cast "type", because if we're in unicode we need to pass a
8-bit char. this is safe, because we've restricted what "type"
can be */
PyOS_snprintf(fmt, sizeof(fmt), "%%.%zd%c", precision, (char)type);
/* call the passed in function to do the actual formatting */
snprintf(charbuf, sizeof(charbuf), fmt, x);
/* adding trailing to fmt with PyOS_snprintf doesn't work, not
sure why. we'll just concatentate it here, no harm done. we
know we can't have a buffer overflow from the fmt size
analysis */
strcat(charbuf, trailing);
/* rather than duplicate the code for snprintf for both unicode
and 8 bit strings, we just use the 8 bit version and then
convert to unicode in a separate code path. that's probably
the lesser of 2 evils. */
#if STRINGLIB_IS_UNICODE
n_digits = strtounicode(unicodebuf, charbuf);
p = unicodebuf;
#else
/* compute the length. I believe this is done because the return
value from snprintf above is unreliable */
n_digits = strlen(charbuf);
p = charbuf;
#endif
/* is a sign character present in the output? if so, remember it
and skip it */
sign = p[0];
if (sign == '-') {
p++;
n_digits--;
}
calc_number_widths(&spec, sign, n_digits, format);
/* allocate a string with enough space */
result = STRINGLIB_NEW(NULL, spec.n_total);
if (result == NULL)
goto done;
/* fill in the non-digit parts */
fill_number(STRINGLIB_STR(result), &spec, n_digits,
format->fill_char == '\0' ? ' ' : format->fill_char);
/* fill in the digit parts */
memmove(STRINGLIB_STR(result) + (spec.n_lpadding + spec.n_lsign + spec.n_spadding),
p,
n_digits * sizeof(STRINGLIB_CHAR));
done:
return result;
}
static PyObject *
format_float_internal(PyObject *value, const InternalFormatSpec *format)
{
if (format->type == 'n')
return _format_float('f', value, format, snprintf_double);
else
return _format_float(format->type, value, format, PyOS_ascii_formatd);
}
/************************************************************************/
/*********** built in formatters ****************************************/
/************************************************************************/
PyObject *
FORMAT_STRING(PyObject* value, PyObject* args)
{
PyObject *format_spec;
PyObject *tmp = NULL;
PyObject *result = NULL;
InternalFormatSpec format;
if (!PyArg_ParseTuple(args, "O:__format__", &format_spec))
goto done;
if (!STRINGLIB_CHECK(format_spec)) {
PyErr_SetString(PyExc_TypeError, STRINGLIB_TYPE_NAME " object required");
goto done;
}
/* check for the special case of zero length format spec, make
it equivalent to str(value) */
if (STRINGLIB_LEN(format_spec) == 0) {
result = STRINGLIB_TOSTR(value);
goto done;
}
/* parse the format_spec */
if (!parse_internal_render_format_spec(format_spec, &format, 's'))
goto done;
/* type conversion? */
switch (format.type) {
case 's':
/* no type conversion needed, already a string. do the formatting */
result = format_string_internal(value, &format);
break;
#if 0
case 'b':
case 'c':
case 'd':
case 'o':
case 'x':
case 'X':
/* convert to integer */
/* XXX: make a stringlib function to do this when backporting,
since FromUnicode differs from FromString */
tmp = PyLong_FromUnicode(STRINGLIB_STR(value), STRINGLIB_LEN(value), 0);
if (tmp == NULL)
goto done;
result = format_long_internal(tmp, &format);
break;
case 'e':
case 'E':
case 'f':
case 'F':
case 'g':
case 'G':
case 'n':
case '%':
/* convert to float */
tmp = PyFloat_FromString(value);
if (tmp == NULL)
goto done;
result = format_float_internal(tmp, &format);
break;
#endif
default:
/* unknown */
PyErr_Format(PyExc_ValueError, "Unknown conversion type %c",
format.type);
goto done;
}
done:
Py_XDECREF(tmp);
return result;
}
PyObject *
FORMAT_LONG(PyObject* value, PyObject* args)
{
PyObject *format_spec;
PyObject *result = NULL;
PyObject *tmp = NULL;
InternalFormatSpec format;
if (!PyArg_ParseTuple(args, "O:__format__", &format_spec))
goto done;
if (!STRINGLIB_CHECK(format_spec)) {
PyErr_SetString(PyExc_TypeError, STRINGLIB_TYPE_NAME " object required");
goto done;
}
/* check for the special case of zero length format spec, make
it equivalent to str(value) */
if (STRINGLIB_LEN(format_spec) == 0) {
result = STRINGLIB_TOSTR(value);
goto done;
}
/* parse the format_spec */
if (!parse_internal_render_format_spec(format_spec, &format, 'd'))
goto done;
/* type conversion? */
switch (format.type) {
#if 0
case 's':
/* convert to string/unicode */
tmp = STRINGLIB_TOSTR(value);
if (tmp == NULL)
goto done;
result = format_string_internal(tmp, &format);
break;
#endif
case 'b':
case 'c':
case 'd':
case 'o':
case 'x':
case 'X':
/* no type conversion needed, already an int. do the formatting */
result = format_long_internal(value, &format);
break;
case 'e':
case 'E':
case 'f':
case 'F':
case 'g':
case 'G':
case 'n':
case '%':
/* convert to float */
tmp = PyNumber_Float(value);
if (tmp == NULL)
goto done;
result = format_float_internal(value, &format);
break;
default:
/* unknown */
PyErr_Format(PyExc_ValueError, "Unknown conversion type %c",
format.type);
goto done;
}
done:
Py_XDECREF(tmp);
return result;
}
PyObject *
FORMAT_FLOAT(PyObject *value, PyObject *args)
{
PyObject *format_spec;
PyObject *result = NULL;
PyObject *tmp = NULL;
InternalFormatSpec format;
if (!PyArg_ParseTuple(args, "O:__format__", &format_spec))
goto done;
if (!STRINGLIB_CHECK(format_spec)) {
PyErr_SetString(PyExc_TypeError, STRINGLIB_TYPE_NAME " object required");
goto done;
}
/* check for the special case of zero length format spec, make
it equivalent to str(value) */
if (STRINGLIB_LEN(format_spec) == 0) {
result = STRINGLIB_TOSTR(value);
goto done;
}
/* parse the format_spec */
if (!parse_internal_render_format_spec(format_spec, &format, 'g'))
goto done;
/* type conversion? */
switch (format.type) {
#if 0
case 's':
/* convert to string/unicode */
tmp = STRINGLIB_TOSTR(value);
if (tmp == NULL)
goto done;
result = format_string_internal(tmp, &format);
break;
#endif
case 'b':
case 'c':
case 'd':
case 'o':
case 'x':
case 'X':
/* convert to integer */
tmp = PyNumber_Long(value);
if (tmp == NULL)
goto done;
result = format_long_internal(tmp, &format);
break;
case 'e':
case 'E':
case 'f':
case 'F':
case 'g':
case 'G':
case 'n':
case '%':
/* no conversion, already a float. do the formatting */
result = format_float_internal(value, &format);
break;
default:
/* unknown */
PyErr_Format(PyExc_ValueError, "Unknown conversion type %c",
format.type);
goto done;
}
done:
Py_XDECREF(tmp);
return result;
}

831
Objects/stringlib/string_format.h Normal file
View file

@ -0,0 +1,831 @@
/*
string_format.h -- implementation of string.format().
It uses the Objects/stringlib conventions, so that it can be
compiled for both unicode and string objects.
*/
/* Defines for more efficiently reallocating the string buffer */
#define INITIAL_SIZE_INCREMENT 100
#define SIZE_MULTIPLIER 2
#define MAX_SIZE_INCREMENT 3200
/************************************************************************/
/*********** Global data structures and forward declarations *********/
/************************************************************************/
/*
A SubString consists of the characters between two string or
unicode pointers.
*/
typedef struct {
STRINGLIB_CHAR *ptr;
STRINGLIB_CHAR *end;
} SubString;
/* forward declaration for recursion */
static PyObject *
build_string(SubString *input, PyObject *args, PyObject *kwargs,
int *recursion_level);
/************************************************************************/
/************************** Utility functions ************************/
/************************************************************************/
/* fill in a SubString from a pointer and length */
Py_LOCAL_INLINE(void)
SubString_init(SubString *str, STRINGLIB_CHAR *p, Py_ssize_t len)
{
str->ptr = p;
if (p == NULL)
str->end = NULL;
else
str->end = str->ptr + len;
}
Py_LOCAL_INLINE(PyObject *)
SubString_new_object(SubString *str)
{
return STRINGLIB_NEW(str->ptr, str->end - str->ptr);
}
/************************************************************************/
/*********** Error handling and exception generation **************/
/************************************************************************/
/*
Most of our errors are value errors, because to Python, the
format string is a "value". Also, it's convenient to return
a NULL when we are erroring out.
XXX: need better error handling, per PEP 3101.
*/
static void *
SetError(const char *s)
{
/* PyErr_Format always returns NULL */
return PyErr_Format(PyExc_ValueError, "%s in format string", s);
}
/*
check_input returns True if we still have characters
left in the input string.
XXX: make this function go away when better error handling is
implemented.
*/
Py_LOCAL_INLINE(int)
check_input(SubString *input)
{
if (input->ptr < input->end)
return 1;
PyErr_SetString(PyExc_ValueError,
"unterminated replacement field");
return 0;
}
/************************************************************************/
/*********** Output string management functions ****************/
/************************************************************************/
typedef struct {
STRINGLIB_CHAR *ptr;
STRINGLIB_CHAR *end;
PyObject *obj;
Py_ssize_t size_increment;
} OutputString;
/* initialize an OutputString object, reserving size characters */
static int
output_initialize(OutputString *output, Py_ssize_t size)
{
output->obj = STRINGLIB_NEW(NULL, size);
if (output->obj == NULL)
return 0;
output->ptr = STRINGLIB_STR(output->obj);
output->end = STRINGLIB_LEN(output->obj) + output->ptr;
output->size_increment = INITIAL_SIZE_INCREMENT;
return 1;
}
/*
output_extend reallocates the output string buffer.
It returns a status: 0 for a failed reallocation,
1 for success.
*/
static int
output_extend(OutputString *output, Py_ssize_t count)
{
STRINGLIB_CHAR *startptr = STRINGLIB_STR(output->obj);
Py_ssize_t curlen = output->ptr - startptr;
Py_ssize_t maxlen = curlen + count + output->size_increment;
if (STRINGLIB_RESIZE(&output->obj, maxlen) < 0)
return 0;
startptr = STRINGLIB_STR(output->obj);
output->ptr = startptr + curlen;
output->end = startptr + maxlen;
if (output->size_increment < MAX_SIZE_INCREMENT)
output->size_increment *= SIZE_MULTIPLIER;
return 1;
}
/*
output_data dumps characters into our output string
buffer.
In some cases, it has to reallocate the string.
It returns a status: 0 for a failed reallocation,
1 for success.
*/
static int
output_data(OutputString *output, const STRINGLIB_CHAR *s, Py_ssize_t count)
{
if ((count > output->end - output->ptr) && !output_extend(output, count))
return 0;
memcpy(output->ptr, s, count * sizeof(STRINGLIB_CHAR));
output->ptr += count;
return 1;
}
/************************************************************************/
/*********** Format string parsing -- integers and identifiers *********/
/************************************************************************/
/*
end_identifier returns true if a character marks
the end of an identifier string.
Although the PEP specifies that identifiers are
numbers or valid Python identifiers, we just let
getattr/getitem handle that, so the implementation
is more flexible than the PEP would indicate.
*/
Py_LOCAL_INLINE(int)
end_identifier(STRINGLIB_CHAR c)
{
switch (c) {
case '.': case '[': case ']':
return 1;
default:
return 0;
}
}
/*
get_integer consumes 0 or more decimal digit characters from an
input string, updates *result with the corresponding positive
integer, and returns the number of digits consumed.
returns -1 on error.
*/
static int
get_integer(STRINGLIB_CHAR **ptr, STRINGLIB_CHAR *end,
Py_ssize_t *result)
{
Py_ssize_t accumulator, digitval, oldaccumulator;
int numdigits;
accumulator = numdigits = 0;
for (;;(*ptr)++, numdigits++) {
if (*ptr >= end)
break;
digitval = STRINGLIB_TODECIMAL(**ptr);
if (digitval < 0)
break;
/*
This trick was copied from old Unicode format code. It's cute,
but would really suck on an old machine with a slow divide
implementation. Fortunately, in the normal case we do not
expect too many digits.
*/
oldaccumulator = accumulator;
accumulator *= 10;
if ((accumulator+10)/10 != oldaccumulator+1) {
PyErr_Format(PyExc_ValueError,
"Too many decimal digits in format string");
return -1;
}
accumulator += digitval;
}
*result = accumulator;
return numdigits;
}
/*
get_identifier is a bit of a misnomer. It returns a value for use
with getattr or getindex. This value will a string/unicode
object. The input cannot be zero length. Continues until end of
input, or end_identifier() returns true.
*/
static PyObject *
get_identifier(SubString *input)
{
STRINGLIB_CHAR *start;
for (start = input->ptr;
input->ptr < input->end && !end_identifier(*input->ptr);
input->ptr++)
;
return STRINGLIB_NEW(start, input->ptr - start);
/*
We might want to add code here to check for invalid Python
identifiers. All identifiers are eventually passed to getattr
or getitem, so there is a check when used. However, we might
want to remove (or not) the ability to have strings like
"a/b" or " ab" or "-1" (which is not parsed as a number).
For now, this is left as an exercise for the first disgruntled
user...
if (XXX -- need check function) {
Py_DECREF(result);
PyErr_SetString(PyExc_ValueError,
"Invalid embedded Python identifier");
return NULL;
}
*/
}
/************************************************************************/
/******** Functions to get field objects and specification strings ******/
/************************************************************************/
/* get_field_and_spec is the main function in this section. It parses
the format string well enough to return a field object to render along
with a field specification string.
*/
/*
look up key in our keyword arguments
*/
static PyObject *
key_lookup(PyObject *kwargs, PyObject *key)
{
PyObject *result;
if (kwargs && (result = PyDict_GetItem(kwargs, key)) != NULL) {
Py_INCREF(result);
return result;
}
return NULL;
}
/*
get_field_object returns the object inside {}, before the
format_spec. It handles getindex and getattr lookups and consumes
the entire input string.
*/
static PyObject *
get_field_object(SubString *input, PyObject *args, PyObject *kwargs)
{
PyObject *myobj, *subobj, *newobj;
STRINGLIB_CHAR c;
Py_ssize_t index;
int isindex, isnumeric, isargument;
index = isnumeric = 0; /* Just to shut up the compiler warnings */
myobj = args;
Py_INCREF(myobj);
for (isindex=1, isargument=1;;) {
if (!check_input(input))
break;
if (!isindex) {
if ((subobj = get_identifier(input)) == NULL)
break;
newobj = PyObject_GetAttr(myobj, subobj);
Py_DECREF(subobj);
} else {
isnumeric = (STRINGLIB_ISDECIMAL(*input->ptr));
if (isnumeric)
/* XXX: add error checking */
get_integer(&input->ptr, input->end, &index);
if (isnumeric && PySequence_Check(myobj))
newobj = PySequence_GetItem(myobj, index);
else {
/* XXX -- do we need PyLong_FromLongLong?
Using ssizet, not int... */
subobj = isnumeric ?
PyInt_FromLong(index) :
get_identifier(input);
if (subobj == NULL)
break;
if (isargument) {
newobj = key_lookup(kwargs, subobj);
} else {
newobj = PyObject_GetItem(myobj, subobj);
}
Py_DECREF(subobj);
}
}
Py_DECREF(myobj);
myobj = newobj;
if (myobj == NULL)
break;
if (!isargument && isindex)
if ((!check_input(input)) || (*(input->ptr++) != ']')) {
SetError("Expected ]");
break;
}
/* if at the end of input, return with myobj */
if (input->ptr >= input->end)
return myobj;
c = *input->ptr;
input->ptr++;
isargument = 0;
isindex = (c == '[');
if (!isindex && (c != '.')) {
SetError("Expected ., [, :, !, or }");
break;
}
}
if ((myobj == NULL) && isargument) {
/* XXX: include more useful error information, like which
* keyword not found or which index missing */
PyErr_Clear();
return SetError(isnumeric
? "Not enough positional arguments"
: "Keyword argument not found");
}
Py_XDECREF(myobj);
return NULL;
}
/************************************************************************/
/***************** Field rendering functions **************************/
/************************************************************************/
/*
render_field() is the main function in this section. It takes the
field object and field specification string generated by
get_field_and_spec, and renders the field into the output string.
format() does the actual calling of the objects __format__ method.
*/
/* returns fieldobj.__format__(format_spec) */
static PyObject *
format(PyObject *fieldobj, SubString *format_spec)
{
static PyObject *format_str = NULL;
PyObject *meth;
PyObject *spec = NULL;
PyObject *result = NULL;
/* Initialize cached value */
if (format_str == NULL) {
/* Initialize static variable needed by _PyType_Lookup */
format_str = PyUnicode_FromString("__format__");
if (format_str == NULL)
return NULL;
}
/* Make sure the type is initialized. float gets initialized late */
if (Py_Type(fieldobj)->tp_dict == NULL)
if (PyType_Ready(Py_Type(fieldobj)) < 0)
return NULL;
/* we need to create an object out of the pointers we have */
spec = SubString_new_object(format_spec);
if (spec == NULL)
goto done;
/* Find the (unbound!) __format__ method (a borrowed reference) */
meth = _PyType_Lookup(Py_Type(fieldobj), format_str);
if (meth == NULL) {
PyErr_Format(PyExc_TypeError,
"Type %.100s doesn't define __format__",
Py_Type(fieldobj)->tp_name);
goto done;
}
/* And call it, binding it to the value */
result = PyObject_CallFunctionObjArgs(meth, fieldobj, spec, NULL);
if (result == NULL)
goto done;
if (!STRINGLIB_CHECK(result)) {
PyErr_SetString(PyExc_TypeError,
"__format__ method did not return "
STRINGLIB_TYPE_NAME);
Py_DECREF(result);
result = NULL;
goto done;
}
done:
Py_XDECREF(spec);
return result;
}
/*
render_field calls fieldobj.__format__(format_spec) method, and
appends to the output.
*/
static int
render_field(PyObject *fieldobj, SubString *format_spec, OutputString *output)
{
int ok = 0;
PyObject *result = format(fieldobj, format_spec);
if (result == NULL)
goto done;
ok = output_data(output,
STRINGLIB_STR(result), STRINGLIB_LEN(result));
done:
Py_XDECREF(result);
return ok;
}
static int
parse_field(SubString *str, SubString *field_name, SubString *format_spec,
STRINGLIB_CHAR *conversion)
{
STRINGLIB_CHAR c = 0;
/* initialize these, as they may be empty */
*conversion = '\0';
SubString_init(format_spec, NULL, 0);
/* search for the field name. it's terminated by the end of the
string, or a ':' or '!' */
field_name->ptr = str->ptr;
while (str->ptr < str->end) {
switch (c = *(str->ptr++)) {
case ':':
case '!':
break;
default:
continue;
}
break;
}
if (c == '!' || c == ':') {
/* we have a format specifier and/or a conversion */
/* don't include the last character */
field_name->end = str->ptr-1;
/* the format specifier is the rest of the string */
format_spec->ptr = str->ptr;
format_spec->end = str->end;
/* see if there's a conversion specifier */
if (c == '!') {
/* there must be another character present */
if (format_spec->ptr >= format_spec->end) {
PyErr_SetString(PyExc_ValueError,
"end of format while looking for conversion "
"specifier");
return 0;
}
*conversion = *(format_spec->ptr++);
/* if there is another character, it must be a colon */
if (format_spec->ptr < format_spec->end) {
c = *(format_spec->ptr++);
if (c != ':') {
PyErr_SetString(PyExc_ValueError,
"expected ':' after format specifier");
return 0;
}
}
}
return 1;
} else {
/* end of string, there's no format_spec or conversion */
field_name->end = str->ptr;
return 1;
}
}
/************************************************************************/
/******* Output string allocation and escape-to-markup processing ******/
/************************************************************************/
/* MarkupIterator breaks the string into pieces of either literal
text, or things inside {} that need to be marked up. it is
designed to make it easy to wrap a Python iterator around it, for
use with the Formatter class */
typedef struct {
SubString str;
int in_markup;
} MarkupIterator;
static int
MarkupIterator_init(MarkupIterator *self, STRINGLIB_CHAR *ptr, Py_ssize_t len)
{
SubString_init(&self->str, ptr, len);
self->in_markup = 0;
return 1;
}
/* returns 0 on error, 1 on non-error termination, and 2 if it got a
string (or something to be expanded) */
static int
MarkupIterator_next(MarkupIterator *self, int *is_markup, SubString *literal,
SubString *field_name, SubString *format_spec,
STRINGLIB_CHAR *conversion,
int *format_spec_needs_expanding)
{
int at_end;
STRINGLIB_CHAR c = 0;
STRINGLIB_CHAR *start;
int count;
Py_ssize_t len;
*format_spec_needs_expanding = 0;
/* no more input, end of iterator */
if (self->str.ptr >= self->str.end)
return 1;
*is_markup = self->in_markup;
start = self->str.ptr;
if (self->in_markup) {
/* prepare for next iteration */
self->in_markup = 0;
/* this is markup, find the end of the string by counting nested
braces. note that this prohibits escaped braces, so that
format_specs cannot have braces in them. */
count = 1;
/* we know we can't have a zero length string, so don't worry
about that case */
while (self->str.ptr < self->str.end) {
switch (c = *(self->str.ptr++)) {
case '{':
/* the format spec needs to be recursively expanded.
this is an optimization, and not strictly needed */
*format_spec_needs_expanding = 1;
count++;
break;
case '}':
count--;
if (count <= 0) {
/* we're done. parse and get out */
literal->ptr = start;
literal->end = self->str.ptr-1;
if (parse_field(literal, field_name, format_spec,
conversion) == 0)
return 0;
/* success */
return 2;
}
break;
}
}
/* end of string while searching for matching '}' */
PyErr_SetString(PyExc_ValueError, "unmatched '{' in format");
return 0;
} else {
/* literal text, read until the end of string, an escaped { or },
or an unescaped { */
while (self->str.ptr < self->str.end) {
switch (c = *(self->str.ptr++)) {
case '{':
case '}':
self->in_markup = 1;
break;
default:
continue;
}
break;
}
at_end = self->str.ptr >= self->str.end;
len = self->str.ptr - start;
if ((c == '}') && (at_end || (c != *self->str.ptr)))
return (int)SetError("Single } encountered");
if (at_end && c == '{')
return (int)SetError("Single { encountered");
if (!at_end) {
if (c == *self->str.ptr) {
/* escaped } or {, skip it in the input */
self->str.ptr++;
self->in_markup = 0;
} else
len--;
}
/* this is just plain text, return it */
literal->ptr = start;
literal->end = start + len;
return 2;
}
}
/* do the !r or !s conversion on obj */
static PyObject *
do_conversion(PyObject *obj, STRINGLIB_CHAR conversion)
{
/* XXX in pre-3.0, do we need to convert this to unicode, since it
might have returned a string? */
switch (conversion) {
case 'r':
return PyObject_Repr(obj);
case 's':
return PyObject_Unicode(obj);
default:
PyErr_Format(PyExc_ValueError,
"Unknown converion specifier %c",
conversion);
return NULL;
}
}
/* given:
{field_name!conversion:format_spec}
compute the result and write it to output.
format_spec_needs_expanding is an optimization. if it's false,
just output the string directly, otherwise recursively expand the
format_spec string. */
static int
output_markup(SubString *field_name, SubString *format_spec,
int format_spec_needs_expanding, STRINGLIB_CHAR conversion,
OutputString *output, PyObject *args, PyObject *kwargs,
int *recursion_level)
{
PyObject *tmp = NULL;
PyObject *fieldobj = NULL;
SubString expanded_format_spec;
SubString *actual_format_spec;
int result = 0;
/* convert field_name to an object */
fieldobj = get_field_object(field_name, args, kwargs);
if (fieldobj == NULL)
goto done;
if (conversion != '\0') {
tmp = do_conversion(fieldobj, conversion);
if (tmp == NULL)
goto done;
/* do the assignment, transferring ownership: fieldobj = tmp */
Py_DECREF(fieldobj);
fieldobj = tmp;
tmp = NULL;
}
/* if needed, recurively compute the format_spec */
if (format_spec_needs_expanding) {
tmp = build_string(format_spec, args, kwargs, recursion_level);
if (tmp == NULL)
goto done;
/* note that in the case we're expanding the format string,
tmp must be kept around until after the call to
render_field. */
SubString_init(&expanded_format_spec,
STRINGLIB_STR(tmp), STRINGLIB_LEN(tmp));
actual_format_spec = &expanded_format_spec;
} else
actual_format_spec = format_spec;
if (render_field(fieldobj, actual_format_spec, output) == 0)
goto done;
result = 1;
done:
Py_XDECREF(fieldobj);
Py_XDECREF(tmp);
return result;
}
/*
do_markup is the top-level loop for the format() function. It
searches through the format string for escapes to markup codes, and
calls other functions to move non-markup text to the output,
and to perform the markup to the output.
*/
static int
do_markup(SubString *input, PyObject *args, PyObject *kwargs,
OutputString *output, int *recursion_level)
{
MarkupIterator iter;
int is_markup;
int format_spec_needs_expanding;
int result;
SubString str;
SubString field_name;
SubString format_spec;
STRINGLIB_CHAR conversion;
MarkupIterator_init(&iter, input->ptr, input->end - input->ptr);
while ((result = MarkupIterator_next(&iter, &is_markup, &str, &field_name,
&format_spec, &conversion,
&format_spec_needs_expanding)) == 2) {
if (is_markup) {
if (!output_markup(&field_name, &format_spec,
format_spec_needs_expanding, conversion, output,
args, kwargs, recursion_level))
return 0;
} else {
if (!output_data(output, str.ptr, str.end-str.ptr))
return 0;
}
}
return result;
}
/*
build_string allocates the output string and then
calls do_markup to do the heavy lifting.
*/
static PyObject *
build_string(SubString *input, PyObject *args, PyObject *kwargs,
int *recursion_level)
{
OutputString output;
PyObject *result = NULL;
Py_ssize_t count;
output.obj = NULL; /* needed so cleanup code always works */
/* check the recursion level */
(*recursion_level)--;
if (*recursion_level < 0) {
PyErr_SetString(PyExc_ValueError,
"Max string recursion exceeded");
goto done;
}
/* initial size is the length of the format string, plus the size
increment. seems like a reasonable default */
if (!output_initialize(&output,
input->end - input->ptr +
INITIAL_SIZE_INCREMENT))
goto done;
if (!do_markup(input, args, kwargs, &output, recursion_level)) {
goto done;
}
count = output.ptr - STRINGLIB_STR(output.obj);
if (STRINGLIB_RESIZE(&output.obj, count) < 0) {
goto done;
}
/* transfer ownership to result */
result = output.obj;
output.obj = NULL;
done:
(*recursion_level)++;
Py_XDECREF(output.obj);
return result;
}
/************************************************************************/
/*********** main routine ***********************************************/
/************************************************************************/
/* this is the main entry point */
static PyObject *
do_string_format(PyObject *self, PyObject *args, PyObject *kwargs)
{
SubString input;
/* PEP 3101 says only 2 levels, so that
"{0:{1}}".format('abc', 's') # works
"{0:{1:{2}}}".format('abc', 's', '') # fails
*/
int recursion_level = 2;
SubString_init(&input, STRINGLIB_STR(self), STRINGLIB_LEN(self));
return build_string(&input, args, kwargs, &recursion_level);
}

23
Objects/stringlib/stringdefs.h Normal file
View file

@ -0,0 +1,23 @@
#ifndef STRINGLIB_STRINGDEFS_H
#define STRINGLIB_STRINGDEFS_H
/* this is sort of a hack. there's at least one place (formatting
floats) where some stringlib code takes a different path if it's
compiled as unicode. */
#define STRINGLIB_IS_UNICODE 0
#define STRINGLIB_CHAR char
#define STRINGLIB_TYPE_NAME "string"
#define STRINGLIB_EMPTY string_empty
#define STRINGLIB_ISDECIMAL(x) ((x >= '0') && (x <= '9'))
#define STRINGLIB_TODECIMAL(x) (STRINGLIB_ISDECIMAL(x) ? (x - '0') : -1)
#define STRINGLIB_FILL memset
#define STRINGLIB_STR PyString_AS_STRING
#define STRINGLIB_LEN PyString_GET_SIZE
#define STRINGLIB_NEW PyString_FromStringAndSize
#define STRINGLIB_RESIZE _PyString_Resize
#define STRINGLIB_CHECK PyString_Check
#define STRINGLIB_CMP memcmp
#define STRINGLIB_TOSTR PyObject_Str
#endif /* !STRINGLIB_STRINGDEFS_H */

32
Objects/stringlib/unicodedefs.h Normal file
View file

@ -0,0 +1,32 @@
#ifndef STRINGLIB_UNICODEDEFS_H
#define STRINGLIB_UNICODEDEFS_H
/* this is sort of a hack. there's at least one place (formatting
floats) where some stringlib code takes a different path if it's
compiled as unicode. */
#define STRINGLIB_IS_UNICODE 1
#define STRINGLIB_CHAR Py_UNICODE
#define STRINGLIB_TYPE_NAME "unicode"
#define STRINGLIB_EMPTY unicode_empty
#define STRINGLIB_ISDECIMAL Py_UNICODE_ISDECIMAL
#define STRINGLIB_TODECIMAL Py_UNICODE_TODECIMAL
#define STRINGLIB_TOUPPER Py_UNICODE_TOUPPER
#define STRINGLIB_TOLOWER Py_UNICODE_TOLOWER
#define STRINGLIB_FILL Py_UNICODE_FILL
#define STRINGLIB_STR PyUnicode_AS_UNICODE
#define STRINGLIB_LEN PyUnicode_GET_SIZE
#define STRINGLIB_NEW PyUnicode_FromUnicode
#define STRINGLIB_RESIZE PyUnicode_Resize
#define STRINGLIB_CHECK PyUnicode_Check
#define STRINGLIB_TOSTR PyObject_Unicode
Py_LOCAL_INLINE(int)
STRINGLIB_CMP(const Py_UNICODE* str, const Py_UNICODE* other, Py_ssize_t len)
{
if (str[0] != other[0])
return 1;
return memcmp((void*) str, (void*) other, len * sizeof(Py_UNICODE));
}
#endif /* !STRINGLIB_UNICODEDEFS_H */