bug: using struct-types with Capture interface is broken

[torbenschinke]

In setField nodes.go:592 the type is asserted to a string to assemble the slice parameter for the Capture method. That is wrong according to the following test:

type Boxes struct {
	Pos          lexer.Position
	Boxes Box `@@`
}

type Box struct{
	Pos          lexer.Position
	Val string `@Ident`
}

func (b *Box) Capture(values []string) error {
	b.Val = values[0]
	return nil
}

func TestBoxedCapture(t *testing.T) {
	lex := stateful.MustSimple([]stateful.Rule{
		{"Ident", `[a-zA-Z](\w|\.|/|:|-)*`, nil},
		{"whitespace", `\s+`, nil},
	})

	parser := participle.MustBuild(&Boxes{},
		participle.Lexer(lex),
		participle.UseLookahead(2),
	)

	boxed:=&Boxes{}
	if err:=parser.ParseString("test", "abc::cdef.abc", boxed);err!=nil{
		t.Fatal(err)
	}
}

This causes a type assertion panic:

panic: interface conversion: interface {} is parser.Box, not string [recovered]
	panic: interface conversion: interface {} is parser.Box, not string

goroutine 6 [running]:
testing.tRunner.func1.2(0x1166b80, 0xc00006c810)
	/usr/local/go/src/testing/testing.go:1144 +0x332
testing.tRunner.func1(0xc000001380)
	/usr/local/go/src/testing/testing.go:1147 +0x4b6
panic(0x1166b80, 0xc00006c810)
	/usr/local/go/src/runtime/panic.go:965 +0x1b9
github.com/alecthomas/participle/v2.setField(0xc000022880, 0x1, 0x1, 0x116f320, 0xc0000585a0, 0x199, 0x114c743, 0x5, 0x0, 0x0, ...)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/nodes.go:582 +0x23e5
github.com/alecthomas/participle/v2.(*parseContext).Apply(0xc0000102d0, 0xc000022880, 0x1)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/context.go:53 +0x150
github.com/alecthomas/participle/v2.(*strct).Parse(0xc000114d20, 0xc0000102d0, 0x116f320, 0xc0000584e0, 0x199, 0x199, 0xc00007fcc8, 0x113635e, 0x1168200, 0xc00006c4b0)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/nodes.go:121 +0x5cc
github.com/alecthomas/participle/v2.(*Parser).parseInto(0xc000114c40, 0xc0000102d0, 0x1156840, 0xc0000584e0, 0x16, 0x11c23c8, 0x128e940)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/parser.go:239 +0xd0
github.com/alecthomas/participle/v2.(*Parser).parseOne(0xc000114c40, 0xc0000102d0, 0x1156840, 0xc0000584e0, 0x16, 0xc000022800, 0x1)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/parser.go:222 +0x77
github.com/alecthomas/participle/v2.(*Parser).ParseFromLexer(0xc000114c40, 0xc000114e70, 0x1156840, 0xc0000584e0, 0x0, 0x0, 0x0, 0x0, 0x0)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/parser.go:152 +0x4c5
github.com/alecthomas/participle/v2.(*Parser).parse(0xc000114c40, 0x11c1a28, 0xc000058540, 0x1156840, 0xc0000584e0, 0x0, 0x0, 0x0, 0x0, 0xc0000584e0)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/parser.go:160 +0xe5
github.com/alecthomas/participle/v2.(*Parser).ParseString(0xc000114c40, 0x118ff92, 0x4, 0x11918f3, 0xd, 0x1156840, 0xc0000584e0, 0x0, 0x0, 0x0, ...)
	/Users/tschinke/go/pkg/mod/github.com/alecthomas/participle/[email protected]/parser.go:189 +0xf5
github.com/golangee/architecture/adl/parser.TestBoxedCapture(0xc000001380)
	/Users/tschinke/git/github.com/golangee/architecture.git/adl/parser/parser_test.go:73 +0x1fa
testing.tRunner(0xc000001380, 0x119ad10)
	/usr/local/go/src/testing/testing.go:1194 +0xef
created by testing.(*T).Run
	/usr/local/go/src/testing/testing.go:1239 +0x2b3

btw, just closing #139 "as incorrect" was a bit offensive.

[alecthomas]

Definitely looks like a bug.

btw, just closing #139 "as incorrect" was a bit offensive.

Given the information you provided in the previous ticket, it was incorrect - it does work with other types including bool, ints, etc., just not structs. If you'd provided the above panic there, then I would have had enough information to determine that there was a bug.

[alecthomas]

I've finally had a chance to look at this. This is a bug, but not quite the one expected.

Altering the grammar to the following works as expected:

-	Boxes Box `@@`
+ 	Boxes *Box `@Ident`

The Capture interface:

transform captured tokens into field values

What this means in practice is that you have to capture tokens explicitly as above with @Ident rather than @@. If you want to use @@ you should implement that Parseable interface.

However there are two bugs in Participle here:

• @@ should not panic in this case.
• Box shouldn't have to be a pointer to work.

[torbenschinke]

cool, many thanks for the follow up.