Added a feature to run the parser without any of the built-in sanity checks (like loop checks). This improved performance, but at the risk of missing some runtime issues with the parser implementation.

This commit is contained in:
Maurice Makaay 2019-06-30 01:05:54 +00:00
parent 7ce12d1632
commit 4b0309453f
4 changed files with 74 additions and 19 deletions

5
Makefile Normal file
View File

@ -0,0 +1,5 @@
test:
@(cd read; go test | grep -v ^PASS)
@(cd tokenize; go test | grep -v ^PASS)
@(cd parse; go test | grep -v ^PASS)
@(cd examples; go test | grep -v ^PASS)

View File

@ -18,9 +18,22 @@ import (
type API struct {
tokenAPI *tokenize.API // the tokenize.API, used for communicating with tokenize.Handler functions
result *tokenize.Result // last tokenize.Handler result as produced by Accept() or Peek()
loopCheck map[string]bool // used for parser loop detection
sanityChecksEnabled bool // whether or not runtime sanity checks are enabled
loopCheck map[filepos]bool // used for parser loop detection
err error // parse error, retrieved by Error(), using API methods is denied when set
stopped bool // a boolean set to true by Stop(), using API methods is denied when true
stopped bool // a boolean set to true by Stop()
}
// DisableSanityChecks disables the built-in parser implementation sanity checks,
// which detects parser implementation errors like loops and continuing parsing
// after an error or invoking Stop().
//
// These tests do cause a performance hit. When your parser has to handle a lot
// of input data and is fairly complex, you might want to disable the sanity
// checks. When you're not sure, You probably don't want to use this method,
// and enjoy the added safety of the built-in checks.
func (p *API) DisableSanityChecks() {
p.sanityChecksEnabled = true
}
// Peek checks if the upcoming input data matches the provided tokenize.Handler.
@ -54,7 +67,7 @@ func (p *API) Accept(tokenHandler tokenize.Handler) bool {
forkedAPI.Merge()
p.result = p.tokenAPI.Result()
forkedAPI.Dispose()
if p.tokenAPI.FlushInput() {
if p.sanityChecksEnabled && p.tokenAPI.FlushInput() {
p.initLoopCheck()
}
}
@ -62,11 +75,13 @@ func (p *API) Accept(tokenHandler tokenize.Handler) bool {
}
func (p *API) invokeHandler(name string, tokenHandler tokenize.Handler) (*tokenize.API, bool) {
if p.sanityChecksEnabled {
p.panicWhenStoppedOrInError(name)
p.checkForLoops(name)
if tokenHandler == nil {
callerPanic(name, "parsekit.parse.API.{name}(): {name}() called with nil tokenHandler argument at {caller}")
}
}
p.result = nil
p.tokenAPI.Reset()
@ -108,7 +123,7 @@ func (p *API) IsStoppedOrInError() bool {
// When Accept() is called, and the parser moved forward in the input data,
// this method is called to reset the map for the new read cursor position.
func (p *API) initLoopCheck() {
p.loopCheck = map[string]bool{}
p.loopCheck = make(map[filepos]bool)
}
// checkForLoops checks if the line of code from which Accept() or Peek()
@ -152,9 +167,13 @@ func (p *API) Result() *tokenize.Result {
// of this method, because it performs some sanity checks and it will return
// an easy to use boolean indicating whether the parser can continue or not.
func (p *API) Handle(parseHandler ...Handler) bool {
if p.sanityChecksEnabled {
p.panicWhenStoppedOrInError("Handle")
}
for _, handler := range parseHandler {
if p.sanityChecksEnabled {
p.panicWhenHandlerNil("Handle", handler)
}
handler(p)
if p.IsStoppedOrInError() {
return false
@ -204,7 +223,9 @@ func (p *API) Error(format string, data ...interface{}) {
// will be stopped through Stop(). Otherwise, the unexpected input is reported
// using Expected("end of file").
func (p *API) ExpectEndOfFile() {
if p.sanityChecksEnabled {
p.panicWhenStoppedOrInError("ExpectEndofFile")
}
if p.Peek(tokenize.A.EndOfFile) {
p.Stop()
} else {
@ -226,7 +247,9 @@ func (p *API) ExpectEndOfFile() {
//
// • there was an error while reading the input.
func (p *API) Expected(expected string) {
if p.sanityChecksEnabled {
p.panicWhenStoppedOrInError("Expected")
}
_, err := p.tokenAPI.NextRune()
switch {
case err == nil:

View File

@ -24,10 +24,19 @@ func callerBefore(name string) string {
}
}
func callerFilepos(depth int) string {
type filepos struct {
file string
line int
}
func (pos *filepos) String() string {
return fmt.Sprintf("%s:%d", pos.file, pos.line)
}
func callerFilepos(depth int) filepos {
// No error handling, because we call this method ourselves with safe depth values.
_, file, line, _ := runtime.Caller(depth + 1)
return fmt.Sprintf("%s:%d", file, line)
return filepos{file, line}
}
func callerPanic(name, f string, data ...interface{}) {

View File

@ -26,13 +26,31 @@ type Func func(interface{}) error
// against the provided input data. For an overview of allowed inputs, take a
// look at the documentation for parsekit.read.New().
func New(startHandler Handler) Func {
return new(startHandler, true)
}
// NewWithoutSanityChecks instantiates a new parser, which does not have
// parsekit's built-in sanith checks enabled (e.g. checks for loops or
// or calls to parse.API methods after an error or Stop()).
//
// Disabling sanity checks does improve parsing performance, but for
// most use cases this is not an issue. Only disable sanity checks when
// you really need the extra performance.
// You can of course create a normal sanity-checked parser that is used
// during development / unit testing, and an unchecked one for production.
func NewWithoutSanityChecks(startHandler Handler) Func {
return new(startHandler, false)
}
func new(startHandler Handler, sanityChecksEnabled bool) Func {
if startHandler == nil {
callerPanic("New", "parsekit.parse.{name}(): {name}() called with nil input at {caller}")
}
return func(input interface{}) error {
api := &API{
tokenAPI: tokenize.NewAPI(input),
loopCheck: map[string]bool{},
loopCheck: make(map[filepos]bool),
sanityChecksEnabled: sanityChecksEnabled,
}
if api.Handle(startHandler) {
// Handle returned true, indicating that parsing could still continue.