[ragel-users] -G2 -Z generating an incorrect FSM

[Karl Stenerud]

I have a ragel machine that generates passing Go code when built with -G1,
but failing code when built with -G2 (using ragel 6.10 on ubuntu).

Given the following machine definition:

-------------------------------------------------------------------------
package safe16l

import (
    "fmt"
)

%%{
    machine safe16;
    access this.;

    numeric_hi = [0-9] @{
        accumulator = (fc - '0') << 4
    };
    af_hi = [a-f] @{
        accumulator = (fc - 'a' + 10) << 4
    };
    AF_hi = [A-F] @{
        accumulator = (fc - 'A' + 10) << 4
    };
    numeric_lo = [0-9] @{
        accumulator |= fc - '0'
    };
    af_lo = [a-f] @{
        accumulator |= fc - 'a' + 10
    };
    AF_lo = [A-F] @{
        accumulator |= fc - 'A' + 10
    };

    sequence = space* (numeric_hi | af_hi | AF_hi) space* (numeric_lo |
af_lo | AF_lo) %{
        dst[dstIndex] = accumulator
        dstIndex++
    };


    length_nocontinue = [0-7] @{
        this.length = this.length << 3 + int(fc - '0')
    };
    length_numeric = [8-9] @{
        this.length = this.length << 3 + int(fc - '0')
    };
    length_af = [a-f] @{
        this.length = this.length << 3 + int(fc - 'a' + 10 - 8)
    };
    length_AF = [A-F] @{
        this.length = this.length << 3 + int(fc - 'A' + 10 - 8)
    };

    length = (length_numeric | length_af | length_AF)* length_nocontinue;

    sequence_counted = sequence %{
        fmt.Printf("### End of sequence\n")
        this.length--
        if this.length == 0 {
            this.isComplete = true
            fbreak;
        }
    };

    main := length sequence_counted* @/{
        err = fmt.Errorf("Incomplete document")
    };
}%%


%% write data;

type Parser struct {
    cs int // Current Ragel state
    data []byte
    length int
    isComplete bool
}

func (this *Parser) Init() {
}

func NewParser() *Parser {
    this := new(Parser)
    return this
}

func (this *Parser) Parse(src []byte, dst []byte) (bytesWritten int,
isComplete bool, err error) {
    this.data = src
    p := 0 // Position: current
    pe := len(this.data) // Position: end of buffer
    // TODO: Change to -1 and check for end of file
    eof := pe // Position: end of file
    accumulator := byte(0)
    dstIndex := 0

    _ = eof

    %%{
        write init;
        write exec;
    }%%

    if this.cs == safe16_error {
        err = fmt.Errorf("Parse error at %v", p)
    }

    return dstIndex, this.isComplete, err
}
-------------------------------------------------------------------------

And the following test code:

-------------------------------------------------------------------------
package safe16l

import (
    "bytes"
    "testing"
)

func testSafe16L(t *testing.T, src string, expected []byte) {
    dst := make([]byte, len(src))
    parser := NewParser()

    bytesWritten, isComplete, err := parser.Parse([]byte(src), dst)
    if err != nil {
        t.Error(err)
    }
    if !isComplete {
        t.Errorf("Sequence [%v]: Incomplete parse", src)
    }

    actual := dst[:bytesWritten]
    if !bytes.Equal(actual, expected) {
        t.Errorf("Sequence [%v]: Expected %v but got %v", src, expected,
actual)
    }
}

func TestSafe16L(t *testing.T) {
    testSafe16L(t, "100", []byte{0x00})
}
-------------------------------------------------------------------------

With G1:

$ ragel -Z -G1 safe16l.rl && go test
### End of sequence
PASS
ok   github.com/kstenerud/go-safe16 0.001s

With G2:

$ ragel -Z -G2 safe16l.rl && go test
### End of sequence
--- FAIL: TestSafe16L (0.00s)
    parser_test.go:14: Parse error at 4
FAIL
exit status 1
FAIL github.com/kstenerud/go-safe16 0.001s

-------------------------------------------------------------------------

Looking at the generated code, I see this in the -G2 version:

//line safe16l.rl:51

        fmt.Printf("### End of sequence\n")
        this.length--
        if this.length == 0 {
            this.isComplete = true
            {p++;  this.cs = 0; goto _out }
        }

I'm not sure why cs would be set to 0 here. This isn't an error state as
far as I can tell...


Cheers,
Karl
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.colm.net/pipermail/ragel-users/attachments/20191108/53e2c0c6/attachment.html>