Sketching out the parser.
This commit is contained in:
parent
9ba796161d
commit
d129ff285c
5 changed files with 258 additions and 18 deletions
55
README.md
Normal file
55
README.md
Normal file
|
|
@ -0,0 +1,55 @@
|
|||
|
||||
# Mk
|
||||
|
||||
Mk is a reboot of the Plan 9 mk command, which itself is a replacement for make.
|
||||
This tool is for anyone who loves make, but hates all its stupid bullshit.
|
||||
|
||||
# Why Plan 9 mk is better than make
|
||||
|
||||
Plan 9 mk blows make out of the water. Yet tragically, few use or have even heard
|
||||
of it. Put simply, mk takes make, keeps its simple direct syntax, but fixes
|
||||
basically everything that's annoyed you over the years. To name a few things:
|
||||
|
||||
1. Recipes are delimited by any indentation, not tab characters in particular.
|
||||
2. Phony targets are handled separately from file targets. Your mkfile won't
|
||||
be broken by having a file named 'clean'.
|
||||
2. Attributes instead of weird special targets like `.SECONDARY:`.
|
||||
5. Special variables like `$target`, `$prereq`, and `$stem` in place of
|
||||
make's pointlessly cryptic `$@`, `$^`, and `$*`.
|
||||
3. In addition to suffix rules (e.g. `%.o: %.c`), mk has more powerful regular
|
||||
expression rules.
|
||||
4. Sane handling of rules with multiple targets.
|
||||
5. An optional attribute to delete targets when a recipe fails, so you aren't
|
||||
left with corrupt output.
|
||||
6. Plan 9 mkfiles can not only include other mkfiles, but pipe in the output of
|
||||
recipes. Your mkfile can configure itself by doing something like
|
||||
`<|sh config.sh`.
|
||||
7. A generalized mechanism to determine if a target is out of date, for when
|
||||
timestamps won't cut it.
|
||||
|
||||
And much more! For more, read the original mk paper: ["Mk: a successor to
|
||||
make"](#).
|
||||
|
||||
# Improvements over Plan 9 mk
|
||||
|
||||
This mk stays mostly faithful to Plan 9, but makes a few minor (in my opinion)
|
||||
improvements.
|
||||
|
||||
1. Allow blank lines in recipes. A recipe is any indented block of text, and
|
||||
continues until a non-indented character or the end of the file.
|
||||
2. Add an 'S' attribute to execute recipes with programs other than sh. This
|
||||
way, you don't have to separate your six line python script into its own
|
||||
file. Just stick it in the mkfile.
|
||||
3. Use a perl-compatible regular expressions. The original mk used plan9
|
||||
regex, which few people know or care to learn.
|
||||
4. A clean, modern implementation in go, that doesn't depend on the whole plan
|
||||
9 for userspace stack.
|
||||
|
||||
Most Plan 9 mkfiles should remain backwards compatible, but strict backwards
|
||||
compatibility isn't the goal.
|
||||
|
||||
# Current State
|
||||
|
||||
Totally non-functional. Check back later!
|
||||
|
||||
|
||||
29
lex.go
29
lex.go
|
|
@ -1,4 +1,6 @@
|
|||
|
||||
// TODO: Backquoted strings.
|
||||
|
||||
package main
|
||||
|
||||
import (
|
||||
|
|
@ -15,7 +17,8 @@ const (
|
|||
tokenError tokenType = iota
|
||||
tokenBareString
|
||||
tokenQuotedString
|
||||
tokenInclude
|
||||
tokenPipeInclude
|
||||
tokenRedirInclude
|
||||
tokenColon
|
||||
tokenAssign
|
||||
tokenRecipe
|
||||
|
|
@ -27,7 +30,8 @@ func (typ tokenType) String() string {
|
|||
case tokenError: return "[Error]"
|
||||
case tokenBareString: return "[BareString]"
|
||||
case tokenQuotedString: return "[QuotedString]"
|
||||
case tokenInclude: return "[Include]"
|
||||
case tokenPipeInclude: return "[PipeInclude]"
|
||||
case tokenRedirInclude: return "[RedirInclude]"
|
||||
case tokenColon: return "[Colon]"
|
||||
case tokenAssign: return "[Assign]"
|
||||
case tokenRecipe: return "[Recipe]"
|
||||
|
|
@ -144,6 +148,17 @@ func (l *lexer) accept(valid string) bool {
|
|||
}
|
||||
|
||||
|
||||
// Skip the next rune if it is in the valid string. Return true if it was
|
||||
// skipped.
|
||||
func (l *lexer) ignore(valid string) bool {
|
||||
if strings.IndexRune(valid, l.peek()) >= 0 {
|
||||
l.skip()
|
||||
return true
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
|
||||
// Consume characters from the valid string until the next is not.
|
||||
func (l *lexer) acceptRun(valid string) int {
|
||||
prevpos := l.pos
|
||||
|
|
@ -256,9 +271,15 @@ func lexComment (l* lexer) lexerStateFun {
|
|||
|
||||
func lexInclude (l* lexer) lexerStateFun {
|
||||
l.skip() // '<'
|
||||
var typ tokenType
|
||||
if l.ignore("|") {
|
||||
typ = tokenPipeInclude
|
||||
} else {
|
||||
typ = tokenRedirInclude
|
||||
}
|
||||
|
||||
l.skipRun(" \t\n\r")
|
||||
l.acceptUntil("\n\r")
|
||||
l.emit(tokenInclude)
|
||||
l.emit(typ)
|
||||
return lexTopLevel
|
||||
}
|
||||
|
||||
|
|
|
|||
14
mk.go
14
mk.go
|
|
@ -2,23 +2,9 @@
|
|||
package main
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"os"
|
||||
"io/ioutil"
|
||||
)
|
||||
|
||||
func main() {
|
||||
input, _ := ioutil.ReadAll(os.Stdin)
|
||||
l, tokens := lex(string(input))
|
||||
|
||||
for t := range tokens {
|
||||
if t.typ == tokenError {
|
||||
fmt.Printf("Error: %s", l.errmsg)
|
||||
break
|
||||
}
|
||||
|
||||
fmt.Println(t.String())
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
121
parse.go
Normal file
121
parse.go
Normal file
|
|
@ -0,0 +1,121 @@
|
|||
|
||||
package main
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"os"
|
||||
"os/exec"
|
||||
)
|
||||
|
||||
/* Grammar, to the best of my knowledge:
|
||||
|
||||
Should we deviate at all from mk?
|
||||
|
||||
Yes! I want to simplify things by saying recipes have nonzero indentation and
|
||||
everything else has zero.
|
||||
|
||||
rule ::= targets ':' attributes ':' prereqs NEWLINE RECIPE |
|
||||
targets ':' prereqs NEWLINE RECIPE
|
||||
|
||||
targets ::= string | string "," targets
|
||||
|
||||
attributes ::= SCALAR | SCALAR attributes
|
||||
|
||||
prereqs ::= string | string "," prereqs
|
||||
|
||||
include ::= '<' string NEWLINE
|
||||
|
||||
string ::= SCALAR | QSTRING
|
||||
|
||||
assignment ::= SCALAR '=' string
|
||||
|
||||
How do we handle escaping new lines?
|
||||
Is newline a token that's emitted?
|
||||
|
||||
*/
|
||||
|
||||
|
||||
// The parser for mk files is terribly simple. There are only three sorts of
|
||||
// statements in mkfiles: variable assignments, rules (possibly with
|
||||
// accompanying recipes), and includes.
|
||||
|
||||
|
||||
|
||||
//
|
||||
// Maybe this is the wrong way to organize things.
|
||||
// We should perhaps have a type for a parsed mkfile that includes every
|
||||
// assignment as well as every rule.
|
||||
//
|
||||
// Rule order should not matter.
|
||||
//
|
||||
// Includes are tricky. If they were straight up includes, the could be
|
||||
// evaluated in place, but they could contain shell script, etc.
|
||||
//
|
||||
// No...we still have to evaluate them in place. That means figuring out how to
|
||||
// spawn shells from go.
|
||||
//
|
||||
|
||||
|
||||
type parser struct {
|
||||
l *lexer // underlying lexer
|
||||
tokenbuf []token // tokens consumed on the current statement
|
||||
rules *ruleSet // current ruleSet
|
||||
}
|
||||
|
||||
|
||||
// A parser state function takes a parser and the next token and returns a new
|
||||
// state function, or nil if there was a parse error.
|
||||
type parserStateFun func (*parser, token) parserStateFun
|
||||
|
||||
|
||||
// Parse a mkfile, returning a new ruleSet.
|
||||
func parse(input string) *ruleSet {
|
||||
rules := &ruleSet{}
|
||||
parseInto(input, rules)
|
||||
return rules
|
||||
}
|
||||
|
||||
|
||||
// Parse a mkfile inserting rules and variables into a given ruleSet.
|
||||
func parseInto(input string, rules *ruleSet) {
|
||||
l, tokens := lex(input)
|
||||
p := &parser{l, []token{}, rules}
|
||||
state := parseTopLevel
|
||||
for t := range tokens {
|
||||
if t.typ == tokenError {
|
||||
// TODO: fancier error messages
|
||||
fmt.Fprintf(os.Stderr, "Error: %s", l.errmsg)
|
||||
break
|
||||
}
|
||||
|
||||
state = state(p, t)
|
||||
}
|
||||
|
||||
// TODO: Handle the case when state is not top level.
|
||||
}
|
||||
|
||||
|
||||
func parseTopLevel(p *parser, t token) parserStateFun {
|
||||
switch t.typ {
|
||||
case tokenPipeInclude: return parsePipeInclude(p, t)
|
||||
// TODO: all others
|
||||
}
|
||||
|
||||
return parseTopLevel
|
||||
}
|
||||
|
||||
|
||||
func parsePipeInclude(p *parser, t token) parserStateFun {
|
||||
// TODO: We need to split this up into arguments so we can feed it into
|
||||
// executeRecipe.
|
||||
return parseTopLevel
|
||||
}
|
||||
|
||||
|
||||
func parseRedirInclude(p *parser, t token) parserStateFun {
|
||||
// TODO: Open the file, read its context, call parseInto recursively.
|
||||
return parseTopLevel
|
||||
}
|
||||
|
||||
|
||||
|
||||
57
recipe.go
Normal file
57
recipe.go
Normal file
|
|
@ -0,0 +1,57 @@
|
|||
|
||||
package main
|
||||
|
||||
|
||||
import (
|
||||
"os/exec"
|
||||
"os"
|
||||
"io"
|
||||
)
|
||||
|
||||
|
||||
// A monolithic function for executing recipes.
|
||||
func executeRecipe(program string,
|
||||
args []string,
|
||||
input string,
|
||||
echo_out bool,
|
||||
echo_err bool,
|
||||
capture_out bool) string {
|
||||
cmd := exec.Command(program, args...)
|
||||
|
||||
if echo_out {
|
||||
cmdout, err := cmd.StdoutPipe()
|
||||
if err != nil {
|
||||
go io.Copy(os.Stdout, cmdout)
|
||||
}
|
||||
}
|
||||
|
||||
if echo_err {
|
||||
cmderr, err := cmd.StdoutPipe()
|
||||
if err != nil {
|
||||
go io.Copy(os.Stderr, cmderr)
|
||||
}
|
||||
}
|
||||
|
||||
if len(input) > 0 {
|
||||
cmdin, err := cmd.StdinPipe()
|
||||
go func () { cmdin.Write([]byte(input)) }()
|
||||
}
|
||||
|
||||
|
||||
output := ""
|
||||
var err error
|
||||
if capture_out {
|
||||
output, err = cmd.Output()
|
||||
} else {
|
||||
err = cmd.Run()
|
||||
}
|
||||
|
||||
if err != nil {
|
||||
// TODO: better error output
|
||||
log.Fatal("Recipe failed")
|
||||
}
|
||||
|
||||
return output
|
||||
}
|
||||
|
||||
|
||||
Loading…
Add table
Add a link
Reference in a new issue