Addons for a nom parser.
The complete code can be found as [examples/example1.rs].
```rust // := ( a | b )* fn parseabstar(input: ESpan<'>) -> EResult<'_, AstABstar> { Context.enter(EABstar, input);
let mut loop_rest = input;
let mut res = AstABstar {
a: vec![],
b: vec![],
};
let mut err = None;
loop {
let rest2 = loop_rest;
let rest2 = match parse_a(rest2) {
Ok((rest3, a)) => {
res.a.push(a);
rest3
}
Err(e) => match parse_b(rest2) {
Ok((rest3, b)) => {
res.b.push(b);
rest3
}
Err(e2) => {
err.append(e)?;
err.append(e2)?;
rest2
}
},
};
if let Some(err) = err {
return Context.err(err);
}
if rest2.is_empty() {
break;
}
loop_rest = rest2;
}
Context.ok(loop_rest, input, res)
} ```
Define the error code enum. The error codes are used in actual error reporting and as a marker when tracing the execution of the parser.
All the nom errorkind are mapped to one parser error and it's kept as extra info.
```rust
pub enum ECode { ENomError,
ETagA,
ETagB,
ENumber,
EAthenB,
EAoptB,
EAstarB,
EABstar,
EAorB,
EABNum,
}
impl Code for ECode { const NOM_ERROR: Self = Self::ENomError; } ```
This crate is very heavy on type variables. The following type aliases are recommended.
```rust
pub type ESpan<'s> = TrackSpan<'s, ECode, &'s str>;
pub type EResult<'s, O> = TrackResult
```
Define your parsers output as you wish. No constraints here.
```rust
struct AstNumber<'s> { pub number: u32, pub span: ESpan<'s>, } ```
Parser functions are the same as with a plain nom parser, just using different input and error types
rust
fn token_number(i: ESpan<'_>) -> EResult<'_, AstNumber<'_>> {
match nom_number(i) {
Ok((rest, (tok, val))) => Ok((
rest,
AstNumber {
number: val,
span: tok,
},
)),
Err(e) => Err(e.with_code(ENumber)),
}
}
ParserError implements nom::error::ParseError, it can be used instead of nom::error::Error.
The trait WithSpan is used to convert an external error to a ParserError and add an error code and a span at the same time.
rust
impl<'s, C: Code, Y: Copy> WithSpan<'s, C, nom::Err<ParserError<'s, C, Y>>>
for std::num::ParseIntError
{
fn with_span(self, code: C, span: Span<'s, C>) -> nom::Err<ParserError<'s, C, Y>>
{
nom::Err::Failure(ParserError::new(code, span))
}
}
With the combinator transform(...)
this can be integrated in the
parser.
rust
fn nom_number(i: ESpan<'_>) -> EResult<'_, (ESpan<'_>, u32)> {
consumed(transform(
terminated(digit1, nom_ws),
|v| (*v).parse::<u32>(),
ENumber,
))(i)
}
The trait WithCode allows altering the error code. The previous error code is kept as a hint.
rust
fn token_number(i: ESpan<'_>) -> EResult<'_, AstNumber<'_>> {
match nom_number(i) {
Ok((rest, (tok, val))) => Ok((
rest,
AstNumber {
number: val,
span: tok,
},
)),
Err(e) => Err(e.with_code(ENumber)),
}
}
The tracker is added as the LocatedSpan.extra field, this way no extra parameters are needed.
To access the tracker the Context struct is used.
```rust fn parsea(input: ESpan<'>) -> EResult<', AstA> { Context.enter(ETagA, input); let (rest, tok) = nomparse_a(input).track()?;
if false {
return Context.err(EParserError::new(EAorB, input));
}
Context.ok(rest, tok, AstA { span: tok })
} ```
enter() and ok() and err() capture the normal control flow of the parser.
track() acts on Result to allow easy error propagation.
Note: There are trackas() and trackok() too.
Create a StdTracker and call the parser with an annotated span.
```rust fn main() { for txt in env::args() { let trk = StdTracker::new(); let span = trk.span(txt.as_str());
match parse_a_b_star(span) {
Ok((rest, val)) => {}
Err(e) => {
println!("{:?}", trk.results());
println!("{:?}", e);
}
}
}
} ```
Tracking only works if a TrackSpan is used in the parser.
If the type alias points to a &str, a &[u8] or any LocatedSpan The call to StdTracker::results() returns the tracking data. The test module has several functions to run a test for one parser function
and to evaluate the result. track_parse() runs the parser and returns a Test struct with a variety of
builder like functions to check the results. If any check went wrong the
q() call reports this as failed test. q() takes one parameter that defines the actual report done.
CheckTrace is one of them, it dumps the trace and the error and panics. ```rust fn test1() {
trackparse(&mut None, "", parseab).errany().q(CheckTrace);
trackparse(&mut None, "ab", parseab)
.okany()
.q(CheckTrace);
trackparse(&mut None, "aba", parse_ab)
.rest("a")
.q(CheckTrace);
}
``` The result looks like this. ```txt
FAIL: Expected ok, but was an error. when parsing LocatedSpan { offset: 0, line: 1, fragment: "aabc", extra: } in 43.4µs =>
trace
(A | B): enter with "aabc"
a: enter with "aabc"
a: ok -> [ "a", "abc" ]
a: enter with "abc"
a: ok -> [ "a", "bc" ]
a: enter with "bc"
a: err ENomError errorkind Tag for span LocatedSpan { offset: 2, line: 1, fragment: "bc", extra: }
b: enter with "bc"
b: ok -> [ "b", "c" ]
a: enter with "c"
a: err ENomError errorkind Tag for span LocatedSpan { offset: 3, line: 1, fragment: "c", extra: }
b: enter with "c"
b: err ENomError errorkind Tag for span LocatedSpan { offset: 3, line: 1, fragment: "c", extra: }
(A | B): err ENomError errorkind Tag for span LocatedSpan { offset: 3, line: 1, fragment: "c", extra: } error
ParserError nom for LocatedSpan { offset: 3, line: 1, fragment: "c", extra: }
errorkind=Tag
``` Just some things I have been missing. Combines parsing and value conversion. If the external error type implements
WithSpan this looks quite smooth. Change the error_code of a partial parser. Runs a condition function on the input and only runs the parser function
if it succeeds. There is nom::cond(), but it's not the the same. This trait is kind of a undo of parsing. It takes two output spans and
can create a span that covers both of them and anything between. nom has consumed() and recognize() for this, which work fine too. "Ok, so now I got the error, but what was the context?" SpanLines can help. It contains the complete parser input and can
find the text lines surrounding any given span returned by the error. SpanBytes does the same with &[u8]. Expect some overhead when tracking is enabled.
When disabled with a different Span type the calls to Context etc boil down
to no-ops, so there should be no difference to a equivalent nom-only parser. The size of ParserError is a bit larger than nom::error::Error, the difference
is the size of the Vec used to store all the extra data. That still amounts
only to 48 bytes on x64, so it's not too worrisome.Getting the tracking data
Testing the parser
[test]
Combinators
transform()
error_code()
conditional()
Error reporting
SpanExt
SpanLines and SpanBytes
Performance