1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
//! Extensions to the parsing API with niche applicability. use super::*; /// Extensions to the `ParseStream` API to support speculative parsing. pub trait Speculative { /// Advance this parse stream to the position of a forked parse stream. /// /// This is the opposite operation to [`ParseStream::fork`]. You can fork a /// parse stream, perform some speculative parsing, then join the original /// stream to the fork to "commit" the parsing from the fork to the main /// stream. /// /// If you can avoid doing this, you should, as it limits the ability to /// generate useful errors. That said, it is often the only way to parse /// syntax of the form `A* B*` for arbitrary syntax `A` and `B`. The problem /// is that when the fork fails to parse an `A`, it's impossible to tell /// whether that was because of a syntax error and the user meant to provide /// an `A`, or that the `A`s are finished and its time to start parsing /// `B`s. Use with care. /// /// Also note that if `A` is a subset of `B`, `A* B*` can be parsed by /// parsing `B*` and removing the leading members of `A` from the /// repetition, bypassing the need to involve the downsides associated with /// speculative parsing. /// /// [`ParseStream::fork`]: ../struct.ParseBuffer.html#method.fork /// /// # Example /// /// There has been chatter about the possibility of making the colons in the /// turbofish syntax like `path::to::<T>` no longer required by accepting /// `path::to<T>` in expression position. Specifically, according to [RFC /// 2544], [`PathSegment`] parsing should always try to consume a following /// `<` token as the start of generic arguments, and reset to the `<` if /// that fails (e.g. the token is acting as a less-than operator). /// /// This is the exact kind of parsing behavior which requires the "fork, /// try, commit" behavior that [`ParseStream::fork`] discourages. With /// `advance_to`, we can avoid having to parse the speculatively parsed /// content a second time. /// /// This change in behavior can be implemented in syn by replacing just the /// `Parse` implementation for `PathSegment`: /// /// ```edition2018 /// # use syn::ext::IdentExt; /// use syn::parse::discouraged::Speculative; /// # use syn::parse::{Parse, ParseStream}; /// # use syn::{Ident, PathArguments, Result, Token}; /// /// pub struct PathSegment { /// pub ident: Ident, /// pub arguments: PathArguments, /// } /// # /// # impl<T> From<T> for PathSegment /// # where /// # T: Into<Ident>, /// # { /// # fn from(ident: T) -> Self { /// # PathSegment { /// # ident: ident.into(), /// # arguments: PathArguments::None, /// # } /// # } /// # } /// /// impl Parse for PathSegment { /// fn parse(input: ParseStream) -> Result<Self> { /// if input.peek(Token![super]) /// || input.peek(Token![self]) /// || input.peek(Token![Self]) /// || input.peek(Token![crate]) /// || input.peek(Token![extern]) /// { /// let ident = input.call(Ident::parse_any)?; /// return Ok(PathSegment::from(ident)); /// } /// /// let ident = input.parse()?; /// if input.peek(Token![::]) && input.peek3(Token![<]) { /// return Ok(PathSegment { /// ident: ident, /// arguments: PathArguments::AngleBracketed(input.parse()?), /// }); /// } /// if input.peek(Token![<]) && !input.peek(Token![<=]) { /// let fork = input.fork(); /// if let Ok(arguments) = fork.parse() { /// input.advance_to(&fork); /// return Ok(PathSegment { /// ident: ident, /// arguments: PathArguments::AngleBracketed(arguments), /// }); /// } /// } /// Ok(PathSegment::from(ident)) /// } /// } /// /// # syn::parse_str::<PathSegment>("a<b,c>").unwrap(); /// ``` /// /// # Drawbacks /// /// The main drawback of this style of speculative parsing is in error /// presentation. Even if the lookahead is the "correct" parse, the error /// that is shown is that of the "fallback" parse. To use the same example /// as the turbofish above, take the following unfinished "turbofish": /// /// ```text /// let _ = f<&'a fn(), for<'a> serde::>(); /// ``` /// /// If this is parsed as generic arguments, we can provide the error message /// /// ```text /// error: expected identifier /// --> src.rs:L:C /// | /// L | let _ = f<&'a fn(), for<'a> serde::>(); /// | ^ /// ``` /// /// but if parsed using the above speculative parsing, it falls back to /// assuming that the `<` is a less-than when it fails to parse the generic /// arguments, and tries to interpret the `&'a` as the start of a labelled /// loop, resulting in the much less helpful error /// /// ```text /// error: expected `:` /// --> src.rs:L:C /// | /// L | let _ = f<&'a fn(), for<'a> serde::>(); /// | ^^ /// ``` /// /// This can be mitigated with various heuristics (two examples: show both /// forks' parse errors, or show the one that consumed more tokens), but /// when you can control the grammar, sticking to something that can be /// parsed LL(3) and without the LL(*) speculative parsing this makes /// possible, displaying reasonable errors becomes much more simple. /// /// [RFC 2544]: https://github.com/rust-lang/rfcs/pull/2544 /// [`PathSegment`]: ../../struct.PathSegment.html /// /// # Performance /// /// This method performs a cheap fixed amount of work that does not depend /// on how far apart the two streams are positioned. /// /// # Panics /// /// The forked stream in the argument of `advance_to` must have been /// obtained by forking `self`. Attempting to advance to any other stream /// will cause a panic. fn advance_to(&self, fork: &Self); } impl<'a> Speculative for ParseBuffer<'a> { fn advance_to(&self, fork: &Self) { if !private::same_scope(self.cursor(), fork.cursor()) { panic!("Fork was not derived from the advancing parse stream"); } // See comment on `cell` in the struct definition. self.cell .set(unsafe { mem::transmute::<Cursor, Cursor<'static>>(fork.cursor()) }) } }