package pkglint

import "fmt"

func parseShellProgram(line *Line, program string) (*MkShList, error) {
	if trace.Tracing {
		defer trace.Call(program)()
	}

	tokens, rest := splitIntoShellTokens(line, program)
	lexer := NewShellLexer(tokens, rest)
	parser := shyyParserImpl{}

	zeroMeansSuccess := parser.Parse(lexer)

	switch {
	case zeroMeansSuccess == 0 && lexer.error == "":
		return lexer.result, nil
	case zeroMeansSuccess == 0:
		return nil, fmt.Errorf("splitIntoShellTokens couldn't parse %q", rest)
	default:
		return nil, &ParseError{append([]string{lexer.current}, lexer.remaining...)}
	}
}

type ParseError struct {
	RemainingTokens []string
}

func (e *ParseError) Error() string {
	return sprintf("parse error at %#v", e.RemainingTokens)
}

// ShellLexer categorizes tokens for shell commands, providing
// the lexer required by the yacc-generated parser.
//
// The main work of tokenizing is done in ShellTokenizer though.
//
// Example:
//  while :; do var=$$other; done
// =>
//  while
//  space " "
//  word ":"
//  semicolon
//  space " "
//  do
//  space " "
//  assign "var=$$other"
//  semicolon
//  space " "
//  done
//
// See splitIntoShellTokens and ShellTokenizer.
type ShellLexer struct {
	current        string
	ioRedirect     string
	remaining      []string
	atCommandStart bool
	sinceFor       int
	sinceCase      int
	inCasePattern  bool // true inside (pattern1|pattern2|pattern3); works only for simple cases
	error          string
	result         *MkShList
}

func NewShellLexer(tokens []string, rest string) *ShellLexer {
	return &ShellLexer{
		remaining:      tokens,
		atCommandStart: true,
		error:          rest}
}

func (lex *ShellLexer) Lex(lval *shyySymType) (ttype int) {
	if len(lex.remaining) == 0 {
		return 0
	}

	if trace.Tracing {
		defer func() {
			if ttype == 0 {
				trace.Stepf("lex EOF because of a comment")
				return
			}
			tname := shyyTokname(int(shyyTok2[ttype-shyyPrivate]))
			switch ttype {
			case tkWORD, tkASSIGNMENT_WORD:
				trace.Stepf("lex %v %q", tname, lval.Word.MkText)
			case tkIO_NUMBER:
				trace.Stepf("lex %v %v", tname, lval.IONum)
			default:
				trace.Stepf("lex %v", tname)
			}
		}()
	}

	token := lex.ioRedirect
	lex.ioRedirect = ""
	if token == "" {
		token = lex.remaining[0]
		lex.current = token
		lex.remaining = lex.remaining[1:]
	}

	switch token {
	case ";":
		lex.atCommandStart = true
		return tkSEMI
	case ";;":
		lex.atCommandStart = true
		lex.inCasePattern = true
		return tkSEMISEMI
	case "\n":
		lex.atCommandStart = true
		return tkNEWLINE
	case "&":
		lex.atCommandStart = true
		return tkBACKGROUND
	case "|":
		lex.atCommandStart = !lex.inCasePattern
		return tkPIPE
	case "(":
		lex.atCommandStart = !lex.inCasePattern
		return tkLPAREN
	case ")":
		lex.atCommandStart = true
		lex.inCasePattern = false
		return tkRPAREN
	case "&&":
		lex.atCommandStart = true
		return tkAND
	case "||":
		lex.atCommandStart = true
		return tkOR

	case ">":
		lex.atCommandStart = false
		return tkGT
	case ">&":
		lex.atCommandStart = false
		return tkGTAND
	case "<":
		lex.atCommandStart = false
		return tkLT
	case "<&":
		lex.atCommandStart = false
		return tkLTAND
	case "<>":
		lex.atCommandStart = false
		return tkLTGT
	case ">>":
		lex.atCommandStart = false
		return tkGTGT
	case "<<":
		lex.atCommandStart = false
		return tkLTLT
	case "<<-":
		lex.atCommandStart = false
		return tkLTLTDASH
	case ">|":
		lex.atCommandStart = false
		return tkGTPIPE
	}

	if m, fdstr, op := match2(token, `^(\d+)(<<-|<<|<>|<&|>>|>&|>\||<|>)$`); m {
		fd := toInt(fdstr, -1)
		lval.IONum = fd
		lex.ioRedirect = op
		return tkIO_NUMBER
	}

	if lex.atCommandStart {
		lex.sinceCase = -1
		lex.sinceFor = -1
		switch token {
		case "if":
			return tkIF
		case "then":
			return tkTHEN
		case "elif":
			return tkELIF
		case "else":
			return tkELSE
		case "fi":
			return tkFI
		case "for":
			lex.atCommandStart = false
			lex.sinceFor = 0
			return tkFOR
		case "while":
			return tkWHILE
		case "until":
			return tkUNTIL
		case "do":
			return tkDO
		case "done":
			// lex.atCommandStart must stay true here because further "done" or "fi"
			// may follow directly, without any semicolon.
			// Ideally lex.atCommandStart would be a tri-state variable: yes, no, partly.
			return tkDONE
		case "in":
			lex.atCommandStart = false
			return tkIN
		case "case":
			lex.atCommandStart = false
			lex.sinceCase = 0
			return tkCASE
		case "{":
			return tkLBRACE
		case "}":
			// See the comment at the "done" case above.
			return tkRBRACE
		case "!":
			return tkEXCLAM
		}
	}

	if lex.sinceFor >= 0 {
		lex.sinceFor++
	}
	if lex.sinceCase >= 0 {
		lex.sinceCase++
	}

	switch {
	case lex.sinceFor == 2 && token == "in":
		ttype = tkIN
		lex.atCommandStart = false
	case lex.sinceFor == 2 && token == "do":
		ttype = tkDO
		lex.atCommandStart = true
	case lex.sinceCase == 2 && token == "in":
		ttype = tkIN
		lex.atCommandStart = false
		lex.inCasePattern = true
	case (lex.atCommandStart || lex.sinceCase == 3) && token == "esac":
		ttype = tkESAC
		lex.atCommandStart = false
	case lex.atCommandStart && matches(token, `^[A-Za-z_]\w*=`):
		ttype = tkASSIGNMENT_WORD
		p := NewShTokenizer(nil, token)
		lval.Word = p.ShToken()
	case hasPrefix(token, "#"):
		// This doesn't work for multiline shell programs.
		// Since pkglint only processes single lines, that's ok.
		return 0
	default:
		ttype = tkWORD
		p := NewShTokenizer(nil, token)
		lval.Word = p.ShToken()
		lex.atCommandStart = false

		// Inside of a case statement, ${PATTERNS:@p@ (${p}) continue ;; @} expands to
		// a list of case-items, and after this list a new command starts.
		// This is necessary to return a following "esac" as tkESAC instead of a
		// simple word.
		if lex.sinceCase >= 0 && len(lval.Word.Atoms) == 1 {
			if varUse := lval.Word.Atoms[0].VarUse(); varUse != nil {
				if len(varUse.modifiers) > 0 {
					lastModifier := varUse.modifiers[len(varUse.modifiers)-1]
					if lastModifier.HasPrefix("@") || lastModifier.HasPrefix("=") {
						lex.atCommandStart = true
					}
				}
			}
		}
	}

	return ttype
}

func (lex *ShellLexer) Error(s string) {
	lex.error = s
}