fmt: remove math package dependency and avoid float operations

Remove floating point comparisons and rely only on the information
directly provided by appendFloat.
Make restoring the zero padding flag explicit instead of using a defer.
Rearrange some case distinctions to remove duplicated code.
Add more test cases for zero padded floating point numbers with sign.

benchmark                   old ns/op     new ns/op     delta
BenchmarkSprintfFloat-4     187           180           -3.74%

Change-Id: Ifa2ae85257909f40b1b18118c92b516933271729
Reviewed-on: https://go-review.googlesource.com/19721
Reviewed-by: Rob Pike <r@golang.org>

diff --git a/src/fmt/fmt_test.go b/src/fmt/fmt_test.go
index 793f709a79055fa5d25a87747b7ceb8d713a35d3..8d7c36ceb1432ef76774cf798b2a32d646f3c17a 100644 (file)
--- a/src/fmt/fmt_test.go
+++ b/src/fmt/fmt_test.go
@@ -259,6 +259,12 @@ var fmtTests = []struct {
        {"%+.3F", float32(-1.0), "-1.000"},
        {"%+07.2f", 1.0, "+001.00"},
        {"%+07.2f", -1.0, "-001.00"},
+       {"%-07.2f", 1.0, "1.00   "},
+       {"%-07.2f", -1.0, "-1.00  "},
+       {"%+-07.2f", 1.0, "+1.00  "},
+       {"%+-07.2f", -1.0, "-1.00  "},
+       {"%-+07.2f", 1.0, "+1.00  "},
+       {"%-+07.2f", -1.0, "-1.00  "},
        {"%+10.2f", +1.0, "     +1.00"},
        {"%+10.2f", -1.0, "     -1.00"},
        {"% .3E", -1.0, "-1.000E+00"},

diff --git a/src/fmt/format.go b/src/fmt/format.go
index 517b18f7d4378e3c60bc3f9c123d10273d42a2ac..bf9d00bbc03b3ae39959a78901a6a2b56687bae6 100644 (file)
--- a/src/fmt/format.go
+++ b/src/fmt/format.go
@@ -5,7 +5,6 @@
 package fmt
 
 import (
-       "math"
        "strconv"
        "unicode/utf8"
 )
@@ -405,42 +404,34 @@ func doPrec(f *fmt, def int) int {
 // formatFloat formats a float64; it is an efficient equivalent to  f.pad(strconv.FormatFloat()...).
 func (f *fmt) formatFloat(v float64, verb byte, prec, n int) {
        // Format number, reserving space for leading + sign if needed.
-       num := strconv.AppendFloat(f.intbuf[0:1], v, verb, prec, n)
+       num := strconv.AppendFloat(f.intbuf[:1], v, verb, prec, n)
        if num[1] == '-' || num[1] == '+' {
                num = num[1:]
        } else {
                num[0] = '+'
        }
-       // Special handling for infinity, which doesn't look like a number so shouldn't be padded with zeros.
-       if math.IsInf(v, 0) {
-               if f.zero {
-                       defer func() { f.zero = true }()
-                       f.zero = false
-               }
-       }
-       // num is now a signed version of the number.
-       // If we're zero padding, want the sign before the leading zeros.
-       // Achieve this by writing the sign out and then padding the unsigned number.
-       if f.zero && f.widPresent && f.wid > len(num) {
-               if f.space && v >= 0 {
-                       f.buf.WriteByte(' ') // This is what C does: even with zero, f.space means space.
-                       f.wid--
-               } else if f.plus || v < 0 {
-                       f.buf.WriteByte(num[0])
-                       f.wid--
-               }
-               f.pad(num[1:])
-               return
-       }
        // f.space says to replace a leading + with a space.
        if f.space && num[0] == '+' {
                num[0] = ' '
+       }
+       // Special handling for "+Inf" and "-Inf",
+       // which don't look like a number so shouldn't be padded with zeros.
+       if num[1] == 'I' {
+               oldZero := f.zero
+               f.zero = false
                f.pad(num)
+               f.zero = oldZero
                return
        }
-       // Now we know the sign is attached directly to the number, if present at all.
-       // We want a sign if asked for, if it's negative, or if it's infinity (+Inf vs. -Inf).
-       if f.plus || num[0] == '-' || math.IsInf(v, 0) {
+       // We want a sign if asked for and if the sign is not positive.
+       if f.plus || num[0] != '+' {
+               // If we're zero padding we want the sign before the leading zeros.
+               // Achieve this by writing the sign out and then padding the unsigned number.
+               if f.zero && f.widPresent && f.wid > len(num) {
+                       f.buf.WriteByte(num[0])
+                       f.wid--
+                       num = num[1:]
+               }
                f.pad(num)
                return
        }