bpo-45995: add "z" format specifer to coerce negative 0 to zero (GH-30049)

Add "z" format specifier to coerce negative 0 to zero.

See https://github.com/python/cpython/issues/90153 (originally https://bugs.python.org/issue45995) for discussion.
This covers `str.format()` and f-strings.  Old-style string interpolation is not supported.

Co-authored-by: Mark Dickinson <dickinsm@gmail.com>

3 files changed, 49 insertions, 2 deletions

diff --git a/Python/ast_opt.c b/Python/ast_opt.c
index 77ed29d0cdd..b1d807bcf10 100644
--- a/Python/ast_opt.c
+++ b/Python/ast_opt.c
@@ -310,6 +310,7 @@ simple_format_arg_parse(PyObject *fmt, Py_ssize_t *ppos,
             case ' ': *flags |= F_BLANK; continue;
             case '#': *flags |= F_ALT; continue;
             case '0': *flags |= F_ZERO; continue;
+            case 'z': *flags |= F_NO_NEG_0; continue;
         }
         break;
     }
diff --git a/Python/formatter_unicode.c b/Python/formatter_unicode.c
index a1e50e20c9d..04d37c0be28 100644
--- a/Python/formatter_unicode.c
+++ b/Python/formatter_unicode.c
@@ -130,6 +130,7 @@ typedef struct {
     Py_UCS4 fill_char;
     Py_UCS4 align;
     int alternate;
+    int no_neg_0;
     Py_UCS4 sign;
     Py_ssize_t width;
     enum LocaleType thousands_separators;
@@ -166,6 +167,7 @@ parse_internal_render_format_spec(PyObject *obj,
     format->fill_char = ' ';
     format->align = default_align;
     format->alternate = 0;
+    format->no_neg_0 = 0;
     format->sign = '\0';
     format->width = -1;
     format->thousands_separators = LT_NO_LOCALE;
@@ -193,6 +195,13 @@ parse_internal_render_format_spec(PyObject *obj,
         ++pos;
     }
 
+    /* If the next character is z, request coercion of negative 0.
+       Applies only to floats. */
+    if (end-pos >= 1 && READ_spec(pos) == 'z') {
+        format->no_neg_0 = 1;
+        ++pos;
+    }
+
     /* If the next character is #, we're in alternate mode.  This only
        applies to integers. */
     if (end-pos >= 1 && READ_spec(pos) == '#') {
@@ -779,6 +788,14 @@ format_string_internal(PyObject *value, const InternalFormatSpec *format,
         goto done;
     }
 
+    /* negative 0 coercion is not allowed on strings */
+    if (format->no_neg_0) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Negative zero coercion (z) not allowed in string format "
+                        "specifier");
+        goto done;
+    }
+
     /* alternate is not allowed on strings */
     if (format->alternate) {
         PyErr_SetString(PyExc_ValueError,
@@ -872,6 +889,13 @@ format_long_internal(PyObject *value, const InternalFormatSpec *format,
                         "Precision not allowed in integer format specifier");
         goto done;
     }
+    /* no negative zero coercion on integers */
+    if (format->no_neg_0) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Negative zero coercion (z) not allowed in integer"
+                        " format specifier");
+        goto done;
+    }
 
     /* special case for character formatting */
     if (format->type == 'c') {
@@ -1049,6 +1073,8 @@ format_float_internal(PyObject *value,
 
     if (format->alternate)
         flags |= Py_DTSF_ALT;
+    if (format->no_neg_0)
+        flags |= Py_DTSF_NO_NEG_0;
 
     if (type == '\0') {
         /* Omitted type specifier.  Behaves in the same way as repr(x)
@@ -1238,6 +1264,8 @@ format_complex_internal(PyObject *value,
 
     if (format->alternate)
         flags |= Py_DTSF_ALT;
+    if (format->no_neg_0)
+        flags |= Py_DTSF_NO_NEG_0;
 
     if (type == '\0') {
         /* Omitted type specifier. Should be like str(self). */
diff --git a/Python/pystrtod.c b/Python/pystrtod.c
index 1b27f0a3ad3..d77b846f040 100644
--- a/Python/pystrtod.c
+++ b/Python/pystrtod.c
@@ -916,6 +916,18 @@ char * PyOS_double_to_string(double val,
                       (flags & Py_DTSF_ALT ? "#" : ""), precision,
                       format_code);
         _PyOS_ascii_formatd(buf, bufsize, format, val, precision);
+
+        if (flags & Py_DTSF_NO_NEG_0 && buf[0] == '-') {
+            char *buf2 = buf + 1;
+            while (*buf2 == '0' || *buf2 == '.') {
+                ++buf2;
+            }
+            if (*buf2 == 0 || *buf2 == 'e') {
+                size_t len = buf2 - buf + strlen(buf2);
+                assert(buf[len] == 0);
+                memmove(buf, buf+1, len);
+            }
+        }
     }
 
     /* Add sign when requested.  It's convenient (esp. when formatting
@@ -995,8 +1007,8 @@ static char *
 format_float_short(double d, char format_code,
                    int mode, int precision,
                    int always_add_sign, int add_dot_0_if_integer,
-                   int use_alt_formatting, const char * const *float_strings,
-                   int *type)
+                   int use_alt_formatting, int no_negative_zero,
+                   const char * const *float_strings, int *type)
 {
     char *buf = NULL;
     char *p = NULL;
@@ -1022,6 +1034,11 @@ format_float_short(double d, char format_code,
     assert(digits_end != NULL && digits_end >= digits);
     digits_len = digits_end - digits;
 
+    if (no_negative_zero && sign == 1 &&
+            (digits_len == 0 || (digits_len == 1 && digits[0] == '0'))) {
+        sign = 0;
+    }
+
     if (digits_len && !Py_ISDIGIT(digits[0])) {
         /* Infinities and nans here; adapt Gay's output,
            so convert Infinity to inf and NaN to nan, and
@@ -1301,6 +1318,7 @@ char * PyOS_double_to_string(double val,
                               flags & Py_DTSF_SIGN,
                               flags & Py_DTSF_ADD_DOT_0,
                               flags & Py_DTSF_ALT,
+                              flags & Py_DTSF_NO_NEG_0,
                               float_strings, type);
 }
 #endif  // _PY_SHORT_FLOAT_REPR == 1