Merged revisions 62586 via svnmerge from

svn+ssh://pythondev@svn.python.org/python/trunk

........
  r62586 | eric.smith | 2008-04-29 21:09:30 -0400 (Tue, 29 Apr 2008) | 5 lines

  Issue 2526, float.__format__ 'n' specifier does not support thousands grouping.

  Implemented grouping, with tests.
  Cleaned up PyOS_ascii_formatd by breaking reformatting into smaller functions.
........

2 files changed, 240 insertions, 129 deletions

diff --git a/Lib/test/test_types.py b/Lib/test/test_types.py
index dae250e..1c7a8cd 100644
--- a/Lib/test/test_types.py
+++ b/Lib/test/test_types.py
@@ -1,8 +1,9 @@
 # Python test set -- part 6, built-in types
 
-from test.test_support import run_unittest
+from test.test_support import run_unittest, run_with_locale
 import unittest
 import sys
+import locale
 
 class TypesTests(unittest.TestCase):
 
@@ -407,6 +408,15 @@ class TypesTests(unittest.TestCase):
                 self.assertEqual(value.__format__(format_spec),
                                  float(value).__format__(format_spec))
 
+    @run_with_locale('LC_NUMERIC', 'en_US.UTF8')
+    def test_float__format__locale(self):
+        # test locale support for __format__ code 'n'
+
+        for i in range(-10, 10):
+            x = 1234567890.0 * (10.0 ** i)
+            self.assertEqual(locale.format('%g', x, grouping=True), format(x, 'n'))
+            self.assertEqual(locale.format('%.10g', x, grouping=True), format(x, '.10n'))
+
     def test_float__format__(self):
         # these should be rewritten to use both format(x, spec) and
         # x.__format__(spec)
diff --git a/Python/pystrtod.c b/Python/pystrtod.c
index 2ca8402..0912cec 100644
--- a/Python/pystrtod.c
+++ b/Python/pystrtod.c
@@ -187,6 +187,38 @@ PyOS_ascii_strtod(const char *nptr, char **endptr)
 	return val;
 }
 
+/* Given a string that may have a decimal point in the current
+   locale, change it back to a dot.  Since the string cannot get
+   longer, no need for a maximum buffer size parameter. */
+Py_LOCAL_INLINE(void)
+change_decimal_from_locale_to_dot(char* buffer)
+{
+	struct lconv *locale_data = localeconv();
+	const char *decimal_point = locale_data->decimal_point;
+
+	if (decimal_point[0] != '.' || decimal_point[1] != 0) {
+		size_t decimal_point_len = strlen(decimal_point);
+
+		if (*buffer == '+' || *buffer == '-')
+			buffer++;
+		while (isdigit(Py_CHARMASK(*buffer)))
+			buffer++;
+		if (strncmp(buffer, decimal_point, decimal_point_len) == 0) {
+			*buffer = '.';
+			buffer++;
+			if (decimal_point_len > 1) {
+				/* buffer needs to get smaller */
+				size_t rest_len = strlen(buffer +
+						     (decimal_point_len - 1));
+				memmove(buffer,
+					buffer + (decimal_point_len - 1),
+					rest_len);
+				buffer[rest_len] = 0;
+			}
+		}
+	}
+}
+
 
 /* From the C99 standard, section 7.19.6:
 The exponent always contains at least two digits, and only as many more digits
@@ -194,6 +226,189 @@ as necessary to represent the exponent.
 */
 #define MIN_EXPONENT_DIGITS 2
 
+/* Ensure that any exponent, if present, is at least MIN_EXPONENT_DIGITS
+   in length. */
+Py_LOCAL_INLINE(void)
+ensure_minumim_exponent_length(char* buffer, size_t buf_size)
+{
+	char *p = strpbrk(buffer, "eE");
+	if (p && (*(p + 1) == '-' || *(p + 1) == '+')) {
+		char *start = p + 2;
+		int exponent_digit_cnt = 0;
+		int leading_zero_cnt = 0;
+		int in_leading_zeros = 1;
+		int significant_digit_cnt;
+
+		/* Skip over the exponent and the sign. */
+		p += 2;
+
+		/* Find the end of the exponent, keeping track of leading
+		   zeros. */
+		while (*p && isdigit(Py_CHARMASK(*p))) {
+			if (in_leading_zeros && *p == '0')
+				++leading_zero_cnt;
+			if (*p != '0')
+				in_leading_zeros = 0;
+			++p;
+			++exponent_digit_cnt;
+		}
+
+		significant_digit_cnt = exponent_digit_cnt - leading_zero_cnt;
+		if (exponent_digit_cnt == MIN_EXPONENT_DIGITS) {
+			/* If there are 2 exactly digits, we're done,
+			   regardless of what they contain */
+		}
+		else if (exponent_digit_cnt > MIN_EXPONENT_DIGITS) {
+			int extra_zeros_cnt;
+
+			/* There are more than 2 digits in the exponent.  See
+			   if we can delete some of the leading zeros */
+			if (significant_digit_cnt < MIN_EXPONENT_DIGITS)
+				significant_digit_cnt = MIN_EXPONENT_DIGITS;
+			extra_zeros_cnt = exponent_digit_cnt -
+				significant_digit_cnt;
+
+			/* Delete extra_zeros_cnt worth of characters from the
+			   front of the exponent */
+			assert(extra_zeros_cnt >= 0);
+
+			/* Add one to significant_digit_cnt to copy the
+			   trailing 0 byte, thus setting the length */
+			memmove(start,
+				start + extra_zeros_cnt,
+				significant_digit_cnt + 1);
+		}
+		else {
+			/* If there are fewer than 2 digits, add zeros
+			   until there are 2, if there's enough room */
+			int zeros = MIN_EXPONENT_DIGITS - exponent_digit_cnt;
+			if (start + zeros + exponent_digit_cnt + 1
+			      < buffer + buf_size) {
+				memmove(start + zeros, start,
+					exponent_digit_cnt + 1);
+				memset(start, '0', zeros);
+			}
+		}
+	}
+}
+
+/* Ensure that buffer has a decimal point in it.  The decimal point
+   will not be in the current locale, it will always be '.' */
+Py_LOCAL_INLINE(void)
+ensure_decimal_point(char* buffer, size_t buf_size)
+{
+	int insert_count = 0;
+	char* chars_to_insert;
+
+	/* search for the first non-digit character */
+	char *p = buffer;
+	while (*p && isdigit(Py_CHARMASK(*p)))
+		++p;
+
+	if (*p == '.') {
+		if (isdigit(Py_CHARMASK(*(p+1)))) {
+			/* Nothing to do, we already have a decimal
+			   point and a digit after it */
+		}
+		else {
+			/* We have a decimal point, but no following
+			   digit.  Insert a zero after the decimal. */
+			++p;
+			chars_to_insert = "0";
+			insert_count = 1;
+		}
+	}
+	else {
+		chars_to_insert = ".0";
+		insert_count = 2;
+	}
+	if (insert_count) {
+		size_t buf_len = strlen(buffer);
+		if (buf_len + insert_count + 1 >= buf_size) {
+			/* If there is not enough room in the buffer
+			   for the additional text, just skip it.  It's
+			   not worth generating an error over. */
+		}
+		else {
+			memmove(p + insert_count, p,
+				buffer + strlen(buffer) - p + 1);
+			memcpy(p, chars_to_insert, insert_count);
+		}
+	}
+}
+
+/* Add the locale specific grouping characters to buffer.  Note
+   that any decimal point (if it's present) in buffer is already
+   locale-specific.  Return 0 on error, else 1. */
+Py_LOCAL_INLINE(int)
+add_thousands_grouping(char* buffer, size_t buf_size)
+{
+	struct lconv *locale_data = localeconv();
+	const char *grouping = locale_data->grouping;
+	const char *thousands_sep = locale_data->thousands_sep;
+	size_t thousands_sep_len = strlen(thousands_sep);
+	const char *decimal_point = locale_data->decimal_point;
+	char *pend = buffer + strlen(buffer); /* current end of buffer */
+	char *pmax = buffer + buf_size;       /* max of buffer */
+	char current_grouping;
+
+	/* Find the decimal point, if any.  We're only concerned
+	   about the characters to the left of the decimal when
+	   adding grouping. */
+	char *p = strstr(buffer, decimal_point);
+	if (!p) {
+		/* No decimal, use the entire string. */
+
+		/* If any exponent, adjust p. */
+		p = strpbrk(buffer, "eE");
+		if (!p)
+			/* No exponent and no decimal.  Use the entire
+			   string. */
+			p = pend;
+	}
+	/* At this point, p points just past the right-most character we
+	   want to format.  We need to add the grouping string for the
+	   characters between buffer and p. */
+
+	/* Starting at p and working right-to-left, keep track of
+	   what grouping needs to be added and insert that. */
+	current_grouping = *grouping++;
+
+	/* If the first character is 0, perform no grouping at all. */
+	if (current_grouping == 0)
+		return 1;
+
+	while (p - buffer > current_grouping) {
+		/* Always leave buffer and pend valid at the end of this
+		   loop, since we might leave with a return statement. */
+
+		/* Is there room to insert thousands_sep_len chars?. */
+		if (pmax - pend <= thousands_sep_len)
+			/* No room. */
+			return 0;
+
+		/* Move the rest of the string down. */
+		p -= current_grouping;
+		memmove(p + thousands_sep_len,
+			p,
+			pend - p + 1);
+		/* Adjust end pointer. */
+		pend += thousands_sep_len;
+		/* Copy the thousands_sep chars into the buffer. */
+		memcpy(p, thousands_sep, thousands_sep_len);
+
+		/* Move to the next grouping character, unless we're
+		   repeating (which is designated by a grouping of 0). */
+		if (*grouping != 0) {
+			current_grouping = *grouping++;
+			if (current_grouping == CHAR_MAX)
+				/* We're done. */
+				return 1;
+		}
+	}
+	return 1;
+}
+
 /* see FORMATBUFLEN in unicodeobject.c */
 #define FLOAT_FORMATBUFLEN 120
 
@@ -222,7 +437,6 @@ PyOS_ascii_formatd(char       *buffer,
 		   const char *format, 
 		   double      d)
 {
-	char *p;
 	char format_char;
 	size_t format_len = strlen(format);
 
@@ -277,144 +491,31 @@ PyOS_ascii_formatd(char       *buffer,
 	/* Have PyOS_snprintf do the hard work */
 	PyOS_snprintf(buffer, buf_size, format, d);
 
-	/* Get the current local, and find the decimal point character (or
-	   string?).  Convert that string back to a dot.  Do not do this if
-	   using the 'n' (number) format code. */
-	if (format_char != 'n') {
-		struct lconv *locale_data = localeconv();
-		const char *decimal_point = locale_data->decimal_point;
-		size_t decimal_point_len = strlen(decimal_point);
-		size_t rest_len;
-
-		assert(decimal_point_len != 0);
+	/* Do various fixups on the return string */
 
-		if (decimal_point[0] != '.' || decimal_point[1] != 0) {
-			p = buffer;
-
-			if (*p == '+' || *p == '-')
-				p++;
-
-			while (isdigit(Py_CHARMASK(*p)))
-				p++;
-
-			if (strncmp(p, decimal_point,
-				    decimal_point_len) == 0) {
-				*p = '.';
-				p++;
-				if (decimal_point_len > 1) {
-					rest_len = strlen(p +
-						      (decimal_point_len - 1));
-					memmove(p, p + (decimal_point_len - 1),
-						rest_len);
-					p[rest_len] = 0;
-				}
-			}
-		}
-	}
+	/* Get the current locale, and find the decimal point string.
+	   Convert that string back to a dot.  Do not do this if using the
+	   'n' (number) format code, since we want to keep the localized
+	   decimal point in that case. */
+	if (format_char != 'n')
+		change_decimal_from_locale_to_dot(buffer);
 
 	/* If an exponent exists, ensure that the exponent is at least
 	   MIN_EXPONENT_DIGITS digits, providing the buffer is large enough
 	   for the extra zeros.  Also, if there are more than
 	   MIN_EXPONENT_DIGITS, remove as many zeros as possible until we get
 	   back to MIN_EXPONENT_DIGITS */
-	p = strpbrk(buffer, "eE");
-	if (p && (*(p + 1) == '-' || *(p + 1) == '+')) {
-		char *start = p + 2;
-		int exponent_digit_cnt = 0;
-		int leading_zero_cnt = 0;
-		int in_leading_zeros = 1;
-		int significant_digit_cnt;
-
-		p += 2;
-		while (*p && isdigit(Py_CHARMASK(*p))) {
-			if (in_leading_zeros && *p == '0')
-				++leading_zero_cnt;
-			if (*p != '0')
-				in_leading_zeros = 0;
-			++p;
-			++exponent_digit_cnt;
-		}
-
-		significant_digit_cnt = exponent_digit_cnt - leading_zero_cnt;
-		if (exponent_digit_cnt == MIN_EXPONENT_DIGITS) {
-			/* If there are 2 exactly digits, we're done,
-			   regardless of what they contain */
-		}
-		else if (exponent_digit_cnt > MIN_EXPONENT_DIGITS) {
-			int extra_zeros_cnt;
-
-			/* There are more than 2 digits in the exponent.  See
-			   if we can delete some of the leading zeros */
-			if (significant_digit_cnt < MIN_EXPONENT_DIGITS)
-				significant_digit_cnt = MIN_EXPONENT_DIGITS;
-			extra_zeros_cnt = exponent_digit_cnt -
-				significant_digit_cnt;
-
-			/* Delete extra_zeros_cnt worth of characters from the
-			   front of the exponent */
-			assert(extra_zeros_cnt >= 0);
-
-			/* Add one to significant_digit_cnt to copy the
-			   trailing 0 byte, thus setting the length */
-			memmove(start,
-				start + extra_zeros_cnt,
-				significant_digit_cnt + 1);
-		}
-		else {
-			/* If there are fewer than 2 digits, add zeros
-			   until there are 2, if there's enough room */
-			int zeros = MIN_EXPONENT_DIGITS - exponent_digit_cnt;
-			if (start + zeros + exponent_digit_cnt + 1
-			      < buffer + buf_size) {
-				memmove(start + zeros, start,
-					exponent_digit_cnt + 1);
-				memset(start, '0', zeros);
-			}
-		}
-	}
+	ensure_minumim_exponent_length(buffer, buf_size);
 
 	/* If format_char is 'Z', make sure we have at least one character
 	   after the decimal point (and make sure we have a decimal point). */
-	if (format_char == 'Z') {
-		int insert_count = 0;
-		char* chars_to_insert;
-
-		/* search for the first non-digit character */
-		p = buffer;
-		while (*p && isdigit(Py_CHARMASK(*p)))
-			++p;
+	if (format_char == 'Z')
+		ensure_decimal_point(buffer, buf_size);
 
-		if (*p == '.') {
-			if (isdigit(Py_CHARMASK(*(p+1)))) {
-				/* Nothing to do, we already have a decimal
-				   point and a digit after it */
-			}
-			else {
-				/* We have a decimal point, but no following
-				   digit.  Insert a zero after the decimal. */
-				++p;
-				chars_to_insert = "0";
-				insert_count = 1;
-			}
-		}
-		else {
-			chars_to_insert = ".0";
-			insert_count = 2;
-		}
-		if (insert_count) {
-			size_t buf_len = strlen(buffer);
-			if (buf_len + insert_count + 1 >= buf_size) {
-				/* If there is not enough room in the buffer
-				   for the additional text, just skip it.  It's
-				   not worth generating an error over. */
-			}
-			else {
-				memmove(p + insert_count, p,
-					buffer + strlen(buffer) - p + 1);
-				memcpy(p, chars_to_insert, insert_count);
-			}
-		}
-	}
+	/* If format_char is 'n', add the thousands grouping. */
+	if (format_char == 'n')
+		if (!add_thousands_grouping(buffer, buf_size))
+			return NULL;
 
 	return buffer;
 }