codec: Use ContainerNext everywhere CheckBreak is used, and enable mo…

…re inlining in generated code

- remove direct calls to CheckBreak(); use ContainerNext instead.
  This cleans up the codebase dramatically, and allows us optimize ContainerNext as needed.
- updated genVersion to 26
- moved some Decoder fields from decRd into decoder
- moved some Encoder fields from encWr into Encoder
- removed genHelperEncoder.WriteStr (which wasn't inline'able) and add genHelperEncoder.EncWr()
  to give access to encWr.WriteStr directly
- Ensure WriteStr and StringZC are inline'able (to give optimal performance to codecgen)

codec/README.md

@@ -247,7 +247,7 @@ caveats. See Encode documentation.
 
 ```go
 const CborStreamBytes byte = 0x5f ...
-const GenVersion = 25
+const GenVersion = 26
 var SelfExt = &extFailWrapper{}
 var GoRpc goRpc
 var MsgpackSpecRpc msgpackSpecRpc

codec/build.sh

@@ -307,8 +307,10 @@ _usage() {
 
     cat <<EOF
 primary usage: $0 
-    -[tesow m n l d]   -> [t=tests (e=extra, s=short, o=cover, w=wait), m=make, n=inlining diagnostics, l=mid-stack inlining, d=race detector]
-    -v                 -> v=verbose
+    -t[esow]   -> t=tests [e=extra, s=short, o=cover, w=wait]
+    -[md]      -> [m=make, d=race detector]
+    -[n l i]   -> [n=inlining diagnostics, l=mid-stack inlining, i=check inlining for path (path)]
+    -v         -> v=verbose
 EOF
     if [[ "$(type -t _usage_run)" = "function" ]]; then _usage_run ; fi
 }
@@ -329,7 +331,7 @@ _main() {
     local gocmd=${MYGOCMD:-go}
 
     OPTIND=1
-    while getopts ":cetmnrgpfvldsowkxyzb:" flag
+    while getopts ":cetmnrgpfvldsowkxyzi" flag
     do
         case "x$flag" in
             'xo') zcover=1 ;;
@@ -341,7 +343,7 @@ _main() {
             'xl') zargs+=("-gcflags"); zargs+=("-l=4") ;;
             'xn') zargs+=("-gcflags"); zargs+=("-m=2") ;;
             'xd') zargs+=("-race") ;;
-            'xb') x='b'; zbenchflags=${OPTARG} ;;
+            # 'xi') x='i'; zbenchflags=${OPTARG} ;;
             x\?) _usage; return 1 ;;
             *) x=$flag ;;
         esac
@@ -359,7 +361,7 @@ _main() {
         'xy') _analyze_debug_types "$@" ;;
         'xz') _analyze_do_inlining_and_more "$@" ;;
         'xk') _go_compiler_validation_suite ;;
-        'xb') _bench "$@" ;;
+        'xi') _check_inlining_one "$@" ;;
     esac
     # unset zforce zargs zbenchflags
 }

codec/decode.go

@@ -16,9 +16,9 @@ import (
 const msgBadDesc = "unrecognized descriptor byte"
 
 const (
-	decDefMaxDepth         = 1024            // maximum depth
-	decDefChanCap          = 64              // should be large, as cap cannot be expanded
-	decScratchByteArrayLen = (8 + 2 + 2) * 8 // around cacheLineSize ie ~64, depending on Decoder size
+	decDefMaxDepth         = 1024                // maximum depth
+	decDefChanCap          = 64                  // should be large, as cap cannot be expanded
+	decScratchByteArrayLen = (8 + 2 + 2 + 1) * 8 // around cacheLineSize ie ~64, depending on Decoder size
 
 	// MARKER: massage decScratchByteArrayLen to ensure xxxDecDriver structs fit within cacheLine*N
 
@@ -150,13 +150,11 @@ type decDriver interface {
 	// If the format doesn't prefix the length, it returns containerLenUnknown.
 	// If the expected array was a nil in the stream, it returns containerLenNil.
 	ReadArrayStart() int
-	ReadArrayEnd()
 
 	// ReadMapStart will return the length of the array.
 	// If the format doesn't prefix the length, it returns containerLenUnknown.
 	// If the expected array was a nil in the stream, it returns containerLenNil.
 	ReadMapStart() int
-	ReadMapEnd()
 
 	reset()
 
@@ -186,6 +184,8 @@ type decDriverContainerTracker interface {
 	ReadArrayElem()
 	ReadMapElemKey()
 	ReadMapElemValue()
+	ReadArrayEnd()
+	ReadMapEnd()
 }
 
 type decNegintPosintFloatNumber interface {
@@ -202,11 +202,11 @@ func (x decDriverNoopNumberHelper) decFloat() (f float64, ok bool) { panic("decF
 
 type decDriverNoopContainerReader struct{}
 
-func (x decDriverNoopContainerReader) ReadArrayStart() (v int) { panic("ReadArrayStart unsupported") }
-func (x decDriverNoopContainerReader) ReadArrayEnd()           {}
-func (x decDriverNoopContainerReader) ReadMapStart() (v int)   { panic("ReadMapStart unsupported") }
-func (x decDriverNoopContainerReader) ReadMapEnd()             {}
-func (x decDriverNoopContainerReader) CheckBreak() (v bool)    { return }
+// func (x decDriverNoopContainerReader) ReadArrayStart() (v int) { panic("ReadArrayStart unsupported") }
+// func (x decDriverNoopContainerReader) ReadMapStart() (v int)   { panic("ReadMapStart unsupported") }
+func (x decDriverNoopContainerReader) ReadArrayEnd()        {}
+func (x decDriverNoopContainerReader) ReadMapEnd()          {}
+func (x decDriverNoopContainerReader) CheckBreak() (v bool) { return }
 
 // DecodeOptions captures configuration options during decode.
 type DecodeOptions struct {
@@ -729,38 +729,21 @@ func (d *Decoder) kStruct(f *codecFnInfo, rv reflect.Value) {
 		}
 		// Not much gain from doing it two ways for array.
 		// Arrays are not used as much for structs.
-		hasLen := containerLen >= 0
-		var checkbreak bool
 		tisfi := ti.sfi.source()
-		for j, si := range tisfi {
-			if hasLen {
-				if j == containerLen {
-					break
-				}
-			} else if d.checkBreak() {
-				checkbreak = true
-				break
-			}
+		hasLen := containerLen >= 0
+
+		// iterate all the items in the stream
+		// if mapped elem-wise to a field, handle it
+		// if more stream items than cap be mapped, error it
+		for j := 0; d.containerNext(j, containerLen, hasLen); j++ {
 			d.arrayElem()
-			d.kStructField(si, rv)
-		}
-		var proceed bool
-		if hasLen {
-			proceed = containerLen > len(tisfi)
-		} else {
-			proceed = !checkbreak
-		}
-		// if (hasLen && containerLen > len(tisfi)) || (!hasLen && !checkbreak) {
-		if proceed {
-			// read remaining values and throw away
-			for j := len(tisfi); ; j++ {
-				if !d.containerNext(j, containerLen, hasLen) {
-					break
-				}
-				d.arrayElem()
+			if j < len(tisfi) {
+				d.kStructField(tisfi[j], rv)
+			} else {
 				d.structFieldNotFound(j, "")
 			}
 		}
+
 		d.arrayEnd()
 	} else {
 		d.onerror(errNeedMapOrArrayDecodeToStruct)
@@ -1422,6 +1405,7 @@ func (d *Decoder) r() *decRd {
 
 func (d *Decoder) init(h Handle) {
 	initHandle(h)
+	d.cbreak = d.js || d.cbor
 	d.bytes = true
 	d.err = errDecoderNotInitialized
 	d.h = h.getBasicHandle()
@@ -1948,7 +1932,34 @@ func (d *Decoder) decodeFloat32() float32 {
 
 // MARKER: do not call mapEnd if mapStart returns containerLenNil.
 
+// MARKER: optimize decoding since all formats do not truly support all decDriver'ish operations.
+// - Read(Map|Array)Start is only supported by all formats.
+// - CheckBreak is only supported by json and cbor.
+// - Read(Map|Array)End is only supported by json.
+// - Read(Map|Array)Elem(Kay|Value) is only supported by json.
+// Honor these in the code, to reduce the number of interface calls (even if empty).
+
+func (d *Decoder) checkBreak() (v bool) {
+	// MARKER: jsonDecDriver.CheckBreak() cannot be inlined (over budget inlining cost).
+	// Consequently, there's no benefit in incurring the cost of this wrapping function.
+	// It is faster to just call the interface method directly.
+
+	// if d.js {
+	// 	return d.jsondriver().CheckBreak()
+	// }
+	// if d.cbor {
+	// 	return d.cbordriver().CheckBreak()
+	// }
+
+	if d.cbreak {
+		v = d.d.CheckBreak()
+	}
+	return
+}
+
 func (d *Decoder) containerNext(j, containerLen int, hasLen bool) bool {
+	// MARKER: keep in sync with gen-helper.go.tmpl
+
 	// return (hasLen && j < containerLen) || !(hasLen || slh.d.checkBreak())
 	if hasLen {
 		return j < containerLen
@@ -1979,7 +1990,10 @@ func (d *Decoder) mapElemValue() {
 }
 
 func (d *Decoder) mapEnd() {
-	d.d.ReadMapEnd()
+	if d.js {
+		d.jsondriver().ReadMapEnd()
+	}
+	// d.d.ReadMapEnd()
 	d.depthDecr()
 	d.c = 0
 }
@@ -2000,7 +2014,10 @@ func (d *Decoder) arrayElem() {
 }
 
 func (d *Decoder) arrayEnd() {
-	d.d.ReadArrayEnd()
+	if d.js {
+		d.jsondriver().ReadArrayEnd()
+	}
+	// d.d.ReadArrayEnd()
 	d.depthDecr()
 	d.c = 0
 }