package parser import "git.tebibyte.media/arf/arf/lexer" import "git.tebibyte.media/arf/arf/infoerr" // TODO: // (parser *ParsingOperation) parseDefaultValues // (parser *ParsingOperation) parseDefaultMemberValues (return tree of new members and a tree of member values) // (parser *ParsingOperation) parseDefaultArrayValues // (parser *ParsingOperation) parseDefaultMemberValue // (parser *ParsingOperation) parseMemberDeclaration // parseInitializationValues starts on the line after a data section, or a set // phrase. It checks for an indent greater than the indent of the aforementioned // data section or set phrase (passed through baseIndent), and if there is, // it parses initialization values. func (parser *ParsingOperation) parseInitializationValues ( baseIndent int, ) ( initializationArgument Argument, err error, ) { // check if line is indented one more than baseIndent if !parser.token.Is(lexer.TokenKindIndent) { return } if parser.token.Value().(int) != baseIndent + 1 { return } initializationArgument.location = parser.token.Location() err = parser.nextToken() if err != nil { return } if parser.token.Is(lexer.TokenKindDot) { // object initialization parser.previousToken() var initializationValues ObjectInitializationValues initializationValues, err = parser.parseObjectInitializationValues() initializationArgument.kind = ArgumentKindObjectInitializationValues initializationArgument.value = initializationValues } else { // array initialization parser.previousToken() var initializationValues ArrayInitializationValues initializationValues, err = parser.parseArrayInitializationValues() initializationArgument.kind = ArgumentKindArrayInitializationValues initializationArgument.value = initializationValues } return } // parseObjectInitializationValues parses a list of object initialization // values until the indentation level drops. func (parser *ParsingOperation) parseObjectInitializationValues () ( initializationValues ObjectInitializationValues, err error, ) { initializationValues.attributes = make(map[string] Argument) baseIndent := 0 begin := true for { // if there is no indent we can just stop parsing if !parser.token.Is(lexer.TokenKindIndent) { break} indent := parser.token.Value().(int) if begin == true { initializationValues.location = parser.token.Location() baseIndent = indent begin = false } // do not parse any further if the indent has changed if indent != baseIndent { break } // move on to the beginning of the line, which must contain // a member initialization value err = parser.nextToken(lexer.TokenKindDot) if err != nil { return } err = parser.nextToken(lexer.TokenKindName) if err != nil { return } name := parser.token.Value().(string) // if the member has already been listed, throw an error _, exists := initializationValues.attributes[name] if exists { err = parser.token.NewError ( "duplicate member \"" + name + "\" in object " + "member initialization", infoerr.ErrorKindError) return } // parse the argument determining the member initialization // value err = parser.nextToken() if err != nil { return } var value Argument if parser.token.Is(lexer.TokenKindNewline) { // recurse err = parser.nextToken(lexer.TokenKindIndent) if err != nil { return } value, err = parser.parseInitializationValues(baseIndent) initializationValues.attributes[name] = value if err != nil { return } } else { // parse as normal argument value, err = parser.parseArgument() initializationValues.attributes[name] = value if err != nil { return } err = parser.expect(lexer.TokenKindNewline) if err != nil { return } err = parser.nextToken() if err != nil { return } } } return } // parseArrayInitializationValues parses a list of array initialization values // until the indentation lexel drops. func (parser *ParsingOperation) parseArrayInitializationValues () ( initializationValues ArrayInitializationValues, err error, ) { baseIndent := 0 begin := true for { // if there is no indent we can just stop parsing if !parser.token.Is(lexer.TokenKindIndent) { break} indent := parser.token.Value().(int) if begin == true { initializationValues.location = parser.token.Location() baseIndent = indent begin = false } // do not parse any further if the indent has changed if indent != baseIndent { break } // move on to the beginning of the line, which must contain // arguments err = parser.nextToken(validArgumentStartTokens...) if err != nil { return } for { // stop parsing this line and go on to the next if a // newline token is encountered if parser.token.Is(lexer.TokenKindNewline) { err = parser.nextToken() if err != nil { return } break } // otherwise, parse the argument var argument Argument argument, err = parser.parseArgument() if err != nil { return } initializationValues.values = append ( initializationValues.values, argument) } } return }