mirror of
https://github.com/hanzoai/traces.git
synced 2026-10-30 18:20:41 +00:00
* Upgrade open-telemetry/opentelemetry-collector-contrib to v0.89.0 * Update CHANGELOG * Fix SIGSEGV due to missing ReportComponentStatus facility
133 lines
4.6 KiB
Go
133 lines
4.6 KiB
Go
package participle
|
|
|
|
import (
|
|
"fmt"
|
|
"io"
|
|
"reflect"
|
|
|
|
"github.com/alecthomas/participle/v2/lexer"
|
|
)
|
|
|
|
// MaxLookahead can be used with UseLookahead to get pseudo-infinite
|
|
// lookahead without the risk of pathological cases causing a stack
|
|
// overflow.
|
|
const MaxLookahead = 99999
|
|
|
|
// An Option to modify the behaviour of the Parser.
|
|
type Option func(p *parserOptions) error
|
|
|
|
// Lexer is an Option that sets the lexer to use with the given grammar.
|
|
func Lexer(def lexer.Definition) Option {
|
|
return func(p *parserOptions) error {
|
|
p.lex = def
|
|
return nil
|
|
}
|
|
}
|
|
|
|
// UseLookahead allows branch lookahead up to "n" tokens.
|
|
//
|
|
// If parsing cannot be disambiguated before "n" tokens of lookahead, parsing will fail.
|
|
//
|
|
// Note that increasing lookahead has a minor performance impact, but also
|
|
// reduces the accuracy of error reporting.
|
|
//
|
|
// If "n" is negative, it will be treated as "infinite" lookahead.
|
|
// This can have a large impact on performance, and does not provide any
|
|
// protection against stack overflow during parsing.
|
|
// In most cases, using MaxLookahead will achieve the same results in practice,
|
|
// but with a concrete upper bound to prevent pathological behavior in the parser.
|
|
// Using infinite lookahead can be useful for testing, or for parsing especially
|
|
// ambiguous grammars. Use at your own risk!
|
|
func UseLookahead(n int) Option {
|
|
return func(p *parserOptions) error {
|
|
p.useLookahead = n
|
|
return nil
|
|
}
|
|
}
|
|
|
|
// CaseInsensitive allows the specified token types to be matched case-insensitively.
|
|
//
|
|
// Note that the lexer itself will also have to be case-insensitive; this option
|
|
// just controls whether literals in the grammar are matched case insensitively.
|
|
func CaseInsensitive(tokens ...string) Option {
|
|
return func(p *parserOptions) error {
|
|
for _, token := range tokens {
|
|
p.caseInsensitive[token] = true
|
|
}
|
|
return nil
|
|
}
|
|
}
|
|
|
|
// ParseTypeWith associates a custom parsing function with some interface type T.
|
|
// When the parser encounters a value of type T, it will use the given parse function to
|
|
// parse a value from the input.
|
|
//
|
|
// The parse function may return anything it wishes as long as that value satisfies the interface T.
|
|
// However, only a single function can be defined for any type T.
|
|
// If you want to have multiple parse functions returning types that satisfy the same interface, you'll
|
|
// need to define new wrapper types for each one.
|
|
//
|
|
// This can be useful if you want to parse a DSL within the larger grammar, or if you want
|
|
// to implement an optimized parsing scheme for some portion of the grammar.
|
|
func ParseTypeWith[T any](parseFn func(*lexer.PeekingLexer) (T, error)) Option {
|
|
return func(p *parserOptions) error {
|
|
parseFnVal := reflect.ValueOf(parseFn)
|
|
parseFnType := parseFnVal.Type()
|
|
if parseFnType.Out(0).Kind() != reflect.Interface {
|
|
return fmt.Errorf("ParseTypeWith: T must be an interface type (got %s)", parseFnType.Out(0))
|
|
}
|
|
prodType := parseFnType.Out(0)
|
|
p.customDefs = append(p.customDefs, customDef{prodType, parseFnVal})
|
|
return nil
|
|
}
|
|
}
|
|
|
|
// Union associates several member productions with some interface type T.
|
|
// Given members X, Y, Z, and W for a union type U, then the EBNF rule is:
|
|
//
|
|
// U = X | Y | Z | W .
|
|
//
|
|
// When the parser encounters a field of type T, it will attempt to parse each member
|
|
// in sequence and take the first match. Because of this, the order in which the
|
|
// members are defined is important. You must be careful to order your members appropriately.
|
|
//
|
|
// An example of a bad parse that can happen if members are out of order:
|
|
//
|
|
// If the first member matches A, and the second member matches A B,
|
|
// and the source string is "AB", then the parser will only match A, and will not
|
|
// try to parse the second member at all.
|
|
func Union[T any](members ...T) Option {
|
|
return func(p *parserOptions) error {
|
|
var t T
|
|
unionType := reflect.TypeOf(&t).Elem()
|
|
if unionType.Kind() != reflect.Interface {
|
|
return fmt.Errorf("union: union type must be an interface (got %s)", unionType)
|
|
}
|
|
memberTypes := make([]reflect.Type, 0, len(members))
|
|
for _, m := range members {
|
|
memberTypes = append(memberTypes, reflect.TypeOf(m))
|
|
}
|
|
p.unionDefs = append(p.unionDefs, unionDef{unionType, memberTypes})
|
|
return nil
|
|
}
|
|
}
|
|
|
|
// ParseOption modifies how an individual parse is applied.
|
|
type ParseOption func(p *parseContext)
|
|
|
|
// Trace the parse to "w".
|
|
func Trace(w io.Writer) ParseOption {
|
|
return func(p *parseContext) {
|
|
p.trace = w
|
|
}
|
|
}
|
|
|
|
// AllowTrailing tokens without erroring.
|
|
//
|
|
// That is, do not error if a full parse completes but additional tokens remain.
|
|
func AllowTrailing(ok bool) ParseOption {
|
|
return func(p *parseContext) {
|
|
p.allowTrailing = ok
|
|
}
|
|
}
|