fmt: cleanup and optimize doPrintf for simple formats

Make a fast path for format strings that do not use
precision or width specifications or argument indices.

Only check and enforce the restriction to not pad left with zeros
in code paths that change either f.minus or f.zero.

Consolidate the if chains at the end of the main doPrintf loop
into a switch statement. Move error printing into extra
functions to reduce size of this switch statement.

name                             old time/op  new time/op  delta
SprintfPadding-2                  234ns ± 1%   233ns ± 1%   -0.54%  (p=0.010 n=19+19)
SprintfEmpty-2                   37.0ns ± 3%  39.1ns ±14%     ~     (p=0.501 n=17+20)
SprintfString-2                   112ns ± 1%   101ns ± 1%   -9.21%  (p=0.000 n=19+20)
SprintfTruncateString-2           139ns ± 1%   139ns ± 0%   +0.57%  (p=0.000 n=19+19)
SprintfQuoteString-2              402ns ± 0%   392ns ± 0%   -2.35%  (p=0.000 n=19+20)
SprintfInt-2                      114ns ± 1%   102ns ± 2%  -10.92%  (p=0.000 n=20+20)
SprintfIntInt-2                   177ns ± 2%   155ns ± 2%  -12.67%  (p=0.000 n=18+18)
SprintfPrefixedInt-2              260ns ± 3%   249ns ± 3%   -4.55%  (p=0.000 n=20+20)
SprintfFloat-2                    190ns ± 1%   178ns ± 2%   -6.54%  (p=0.000 n=20+20)
SprintfComplex-2                  533ns ± 1%   517ns ± 3%   -2.95%  (p=0.000 n=20+20)
SprintfBoolean-2                  102ns ± 1%    93ns ± 2%   -9.30%  (p=0.000 n=20+20)
SprintfHexString-2                176ns ± 0%   168ns ± 2%   -4.49%  (p=0.000 n=16+19)
SprintfHexBytes-2                 181ns ± 1%   174ns ± 2%   -4.27%  (p=0.000 n=20+20)
SprintfBytes-2                    326ns ± 1%   311ns ± 1%   -4.51%  (p=0.000 n=20+20)
ManyArgs-2                        540ns ± 2%   497ns ± 1%   -8.08%  (p=0.000 n=18+16)
FprintInt-2                       150ns ± 0%   149ns ± 0%   -0.33%  (p=0.000 n=20+18)
FprintfBytes-2                    185ns ± 0%   165ns ± 0%  -10.98%  (p=0.000 n=20+18)
FprintIntNoAlloc-2                113ns ± 0%   112ns ± 0%   -0.88%  (p=0.000 n=20+20)

Change-Id: I9ada8faa1f46aa67ea116a94ab3f4ad3e405c8fe
Reviewed-on: https://go-review.googlesource.com/20919
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>

src/fmt/print.go

@@ -943,11 +943,24 @@ func (p *pp) argNumber(argNum int, format string, i int, numArgs int) (newArgNum
 	return argNum, i + wid, ok
 }
 
+func (p *pp) badArgNum(verb rune) {
+	p.buf.WriteString(percentBangString)
+	p.buf.WriteRune(verb)
+	p.buf.WriteString(badIndexString)
+}
+
+func (p *pp) missingArg(verb rune) {
+	p.buf.WriteString(percentBangString)
+	p.buf.WriteRune(verb)
+	p.buf.WriteString(missingString)
+}
+
 func (p *pp) doPrintf(format string, a []interface{}) {
 	end := len(format)
 	argNum := 0         // we process one argument per non-trivial format
 	afterIndex := false // previous item in format was an index like [3].
 	p.reordered = false
+formatLoop:
 	for i := 0; i < end; {
 		p.goodArgNum = true
 		lasti := i
@@ -967,21 +980,40 @@ func (p *pp) doPrintf(format string, a []interface{}) {
 
 		// Do we have flags?
 		p.fmt.clearflags()
-	F:
+	simpleFormat:
 		for ; i < end; i++ {
-			switch format[i] {
+			c := format[i]
+			switch c {
 			case '#':
 				p.fmt.sharp = true
 			case '0':
-				p.fmt.zero = true
+				p.fmt.zero = !p.fmt.minus // Only allow zero padding to the left.
 			case '+':
 				p.fmt.plus = true
 			case '-':
 				p.fmt.minus = true
+				p.fmt.zero = false // Do not pad with zeros to the right.
 			case ' ':
 				p.fmt.space = true
 			default:
-				break F
+				// Fast path for common case of ascii lower case simple verbs
+				// without precision or width or argument indices.
+				if 'a' <= c && c <= 'z' && argNum < len(a) {
+					if c == 'v' {
+						// Go syntax
+						p.fmt.sharpV = p.fmt.sharp
+						p.fmt.sharp = false
+						// Struct-field syntax
+						p.fmt.plusV = p.fmt.plus
+						p.fmt.plus = false
+					}
+					p.printArg(a[argNum], rune(c))
+					argNum++
+					i++
+					continue formatLoop
+				}
+				// Format is more complex than simple flags and a verb or is malformed.
+				break simpleFormat
 			}
 		}
 
@@ -1002,6 +1034,7 @@ func (p *pp) doPrintf(format string, a []interface{}) {
 			if p.fmt.wid < 0 {
 				p.fmt.wid = -p.fmt.wid
 				p.fmt.minus = true
+				p.fmt.zero = false // Do not pad with zeros to the right.
 			}
 			afterIndex = false
 		} else {
@@ -1045,48 +1078,32 @@ func (p *pp) doPrintf(format string, a []interface{}) {
 
 		if i >= end {
 			p.buf.WriteString(noVerbString)
-			continue
+			break
 		}
-		c, w := utf8.DecodeRuneInString(format[i:])
+
+		verb, w := utf8.DecodeRuneInString(format[i:])
 		i += w
-		// percent is special - absorbs no operand
-		if c == '%' {
-			p.buf.WriteByte('%') // We ignore width and prec.
-			continue
-		}
-		if !p.goodArgNum {
-			p.buf.WriteString(percentBangString)
-			p.buf.WriteRune(c)
-			p.buf.WriteString(badIndexString)
-			continue
-		} else if argNum >= len(a) { // out of operands
-			p.buf.WriteString(percentBangString)
-			p.buf.WriteRune(c)
-			p.buf.WriteString(missingString)
-			continue
-		}
 
-		if c == 'v' {
-			if p.fmt.sharp {
-				// Go syntax. Set the flag in the fmt and clear the sharp flag.
+		switch {
+		case verb == '%': // Percent does not absorb operands and ignores f.wid and f.prec.
+			p.buf.WriteByte('%')
+		case !p.goodArgNum:
+			p.badArgNum(verb)
+		case argNum >= len(a): // No argument left over to print for the current verb.
+			p.missingArg(verb)
+		case verb == 'v':
+			// Go syntax
+			p.fmt.sharpV = p.fmt.sharp
 			p.fmt.sharp = false
-				p.fmt.sharpV = true
-			}
-			if p.fmt.plus {
-				// Struct-field syntax. Set the flag in the fmt and clear the plus flag.
+			// Struct-field syntax
+			p.fmt.plusV = p.fmt.plus
 			p.fmt.plus = false
-				p.fmt.plusV = true
-			}
-		}
-
-		// Use space padding instead of zero padding to the right.
-		if p.fmt.minus {
-			p.fmt.zero = false
-		}
-
-		p.printArg(a[argNum], c)
+			fallthrough
+		default:
+			p.printArg(a[argNum], verb)
 			argNum++
 		}
+	}
 
 	// Check for extra arguments unless the call accessed the arguments
 	// out of order, in which case it's too expensive to detect if they've all