Merge branch 'main' into cellulite

crates/filter-parser/src/condition.rs

@@ -7,11 +7,22 @@
 
 use nom::branch::alt;
 use nom::bytes::complete::tag;
+use nom::character::complete::char;
+use nom::character::complete::multispace0;
 use nom::character::complete::multispace1;
 use nom::combinator::cut;
+use nom::combinator::map;
+use nom::combinator::value;
+use nom::sequence::preceded;
 use nom::sequence::{terminated, tuple};
 use Condition::*;
 
+use crate::error::IResultExt;
+use crate::value::parse_vector_value;
+use crate::value::parse_vector_value_cut;
+use crate::Error;
+use crate::ErrorKind;
+use crate::VectorFilter;
 use crate::{parse_value, FilterCondition, IResult, Span, Token};
 
 #[derive(Debug, Clone, PartialEq, Eq)]
@@ -113,6 +124,83 @@ pub fn parse_not_exists(input: Span) -> IResult<FilterCondition> {
     Ok((input, FilterCondition::Not(Box::new(FilterCondition::Condition { fid: key, op: Exists }))))
 }
 
+fn parse_vectors(input: Span) -> IResult<(Token, Option<Token>, VectorFilter<'_>)> {
+    let (input, _) = multispace0(input)?;
+    let (input, fid) = tag("_vectors")(input)?;
+
+    if let Ok((input, _)) = multispace1::<_, crate::Error>(input) {
+        return Ok((input, (Token::from(fid), None, VectorFilter::None)));
+    }
+
+    let (input, _) = char('.')(input)?;
+
+    // From this point, we are certain this is a vector filter, so our errors must be final.
+    // We could use nom's `cut` but it's better to be explicit about the errors
+
+    if let Ok((_, space)) = tag::<_, _, ()>(" ")(input) {
+        return Err(crate::Error::failure_from_kind(space, ErrorKind::VectorFilterMissingEmbedder));
+    }
+
+    let (input, embedder_name) =
+        parse_vector_value_cut(input, ErrorKind::VectorFilterInvalidEmbedder)?;
+
+    let (input, filter) = alt((
+        map(
+            preceded(tag(".fragments"), |input| {
+                let (input, _) = tag(".")(input).map_cut(ErrorKind::VectorFilterMissingFragment)?;
+                parse_vector_value_cut(input, ErrorKind::VectorFilterInvalidFragment)
+            }),
+            VectorFilter::Fragment,
+        ),
+        value(VectorFilter::UserProvided, tag(".userProvided")),
+        value(VectorFilter::DocumentTemplate, tag(".documentTemplate")),
+        value(VectorFilter::Regenerate, tag(".regenerate")),
+        value(VectorFilter::None, nom::combinator::success("")),
+    ))(input)?;
+
+    if let Ok((input, point)) = tag::<_, _, ()>(".")(input) {
+        let opt_value = parse_vector_value(input).ok().map(|(_, v)| v);
+        let value =
+            opt_value.as_ref().map(|v| v.value().to_owned()).unwrap_or_else(|| point.to_string());
+        let context = opt_value.map(|v| v.original_span()).unwrap_or(point);
+        let previous_kind = match filter {
+            VectorFilter::Fragment(_) => Some("fragments"),
+            VectorFilter::DocumentTemplate => Some("documentTemplate"),
+            VectorFilter::UserProvided => Some("userProvided"),
+            VectorFilter::Regenerate => Some("regenerate"),
+            VectorFilter::None => None,
+        };
+        return Err(Error::failure_from_kind(
+            context,
+            ErrorKind::VectorFilterUnknownSuffix(previous_kind, value),
+        ));
+    }
+
+    let (input, _) = multispace1(input).map_cut(ErrorKind::VectorFilterLeftover)?;
+
+    Ok((input, (Token::from(fid), Some(embedder_name), filter)))
+}
+
+/// vectors_exists          = vectors ("EXISTS" | ("NOT" WS+ "EXISTS"))
+pub fn parse_vectors_exists(input: Span) -> IResult<FilterCondition> {
+    let (input, (fid, embedder, filter)) = parse_vectors(input)?;
+
+    // Try parsing "EXISTS" first
+    if let Ok((input, _)) = tag::<_, _, ()>("EXISTS")(input) {
+        return Ok((input, FilterCondition::VectorExists { fid, embedder, filter }));
+    }
+
+    // Try parsing "NOT EXISTS"
+    if let Ok((input, _)) = tuple::<_, _, (), _>((tag("NOT"), multispace1, tag("EXISTS")))(input) {
+        return Ok((
+            input,
+            FilterCondition::Not(Box::new(FilterCondition::VectorExists { fid, embedder, filter })),
+        ));
+    }
+
+    Err(crate::Error::failure_from_kind(input, ErrorKind::VectorFilterOperation))
+}
+
 /// contains        = value "CONTAINS" value
 pub fn parse_contains(input: Span) -> IResult<FilterCondition> {
     let (input, (fid, contains, value)) =

crates/filter-parser/src/error.rs

@@ -42,6 +42,23 @@ pub fn cut_with_err<'a, O>(
     }
 }
 
+pub trait IResultExt<'a> {
+    fn map_cut(self, kind: ErrorKind<'a>) -> Self;
+}
+
+impl<'a, T> IResultExt<'a> for IResult<'a, T> {
+    fn map_cut(self, kind: ErrorKind<'a>) -> Self {
+        self.map_err(move |e: nom::Err<Error<'a>>| {
+            let input = match e {
+                nom::Err::Incomplete(_) => return e,
+                nom::Err::Error(e) => *e.context(),
+                nom::Err::Failure(e) => *e.context(),
+            };
+            Error::failure_from_kind(input, kind)
+        })
+    }
+}
+
 #[derive(Debug)]
 pub struct Error<'a> {
     context: Span<'a>,
@@ -61,6 +78,14 @@ pub enum ErrorKind<'a> {
     GeoBoundingBox,
     MisusedGeoRadius,
     MisusedGeoBoundingBox,
+    VectorFilterLeftover,
+    VectorFilterInvalidQuotes,
+    VectorFilterMissingEmbedder,
+    VectorFilterInvalidEmbedder,
+    VectorFilterMissingFragment,
+    VectorFilterInvalidFragment,
+    VectorFilterUnknownSuffix(Option<&'static str>, String),
+    VectorFilterOperation,
     InvalidPrimary,
     InvalidEscapedNumber,
     ExpectedEof,
@@ -91,6 +116,10 @@ impl<'a> Error<'a> {
         Self { context, kind }
     }
 
+    pub fn failure_from_kind(context: Span<'a>, kind: ErrorKind<'a>) -> nom::Err<Self> {
+        nom::Err::Failure(Self::new_from_kind(context, kind))
+    }
+
     pub fn new_from_external(context: Span<'a>, error: impl std::error::Error) -> Self {
         Self::new_from_kind(context, ErrorKind::External(error.to_string()))
     }
@@ -128,6 +157,20 @@ impl Display for Error<'_> {
         // first line being the diagnostic and the second line being the incriminated filter.
         let escaped_input = input.escape_debug();
 
+        fn key_suggestion<'a>(key: &str, keys: &[&'a str]) -> Option<&'a str> {
+            let typos =
+                levenshtein_automata::LevenshteinAutomatonBuilder::new(2, true).build_dfa(key);
+            for key in keys.iter() {
+                match typos.eval(key) {
+                    levenshtein_automata::Distance::Exact(_) => {
+                        return Some(key);
+                    }
+                    levenshtein_automata::Distance::AtLeast(_) => continue,
+                }
+            }
+            None
+        }
+
         match &self.kind {
             ErrorKind::ExpectedValue(_) if input.trim().is_empty() => {
                 writeln!(f, "Was expecting a value but instead got nothing.")?
@@ -169,6 +212,44 @@ impl Display for Error<'_> {
             ErrorKind::MisusedGeoBoundingBox => {
                 writeln!(f, "The `_geoBoundingBox` filter is an operation and can't be used as a value.")?
             }
+            ErrorKind::VectorFilterLeftover => {
+                writeln!(f, "The vector filter has leftover tokens.")?
+            }
+            ErrorKind::VectorFilterUnknownSuffix(_, value) if value.as_str() == "." => {
+                writeln!(f, "Was expecting one of `.fragments`, `.userProvided`, `.documentTemplate`, `.regenerate` or nothing, but instead found a point without a valid value.")?;
+            }
+            ErrorKind::VectorFilterUnknownSuffix(None, value) if ["fragments", "userProvided", "documentTemplate", "regenerate"].contains(&value.as_str()) => {
+                // This will happen with "_vectors.rest.\"userProvided\"" for instance
+                writeln!(f, "Was expecting this part to be unquoted.")?
+            }
+            ErrorKind::VectorFilterUnknownSuffix(None, value) => {
+                if let Some(suggestion) = key_suggestion(value, &["fragments", "userProvided", "documentTemplate", "regenerate"]) {
+                    writeln!(f, "Was expecting one of `fragments`, `userProvided`, `documentTemplate`, `regenerate` or nothing, but instead found `{value}`. Did you mean `{suggestion}`?")?;
+                } else {
+                    writeln!(f, "Was expecting one of `fragments`, `userProvided`, `documentTemplate`, `regenerate` or nothing, but instead found `{value}`.")?;
+                }
+            }
+            ErrorKind::VectorFilterUnknownSuffix(Some(previous_filter_kind), value) => {
+                writeln!(f, "Vector filter can only accept one of `fragments`, `userProvided`, `documentTemplate` or `regenerate`, but found both `{previous_filter_kind}` and `{value}`.")?
+            },
+            ErrorKind::VectorFilterInvalidFragment => {
+                writeln!(f, "The vector filter's fragment name is invalid.")?
+            }
+            ErrorKind::VectorFilterMissingFragment => {
+                writeln!(f, "The vector filter is missing a fragment name.")?
+            }
+            ErrorKind::VectorFilterMissingEmbedder => {
+                writeln!(f, "Was expecting embedder name but found nothing.")?
+            }
+            ErrorKind::VectorFilterInvalidEmbedder => {
+                writeln!(f, "The vector filter's embedder name is invalid.")?
+            }
+            ErrorKind::VectorFilterOperation => {
+                writeln!(f, "Was expecting an operation like `EXISTS` or `NOT EXISTS` after the vector filter.")?
+            }
+            ErrorKind::VectorFilterInvalidQuotes => {
+                writeln!(f, "The quotes in one of the values are inconsistent.")?
+            }
             ErrorKind::ReservedKeyword(word) => {
                 writeln!(f, "`{word}` is a reserved keyword and thus cannot be used as a field name unless it is put inside quotes. Use \"{word}\" or \'{word}\' instead.")?
             }

crates/filter-parser/src/lib.rs

@@ -66,6 +66,9 @@ use nom_locate::LocatedSpan;
 pub(crate) use value::parse_value;
 use value::word_exact;
 
+use crate::condition::parse_vectors_exists;
+use crate::error::IResultExt;
+
 pub type Span<'a> = LocatedSpan<&'a str, &'a str>;
 
 type IResult<'a, Ret> = nom::IResult<Span<'a>, Ret, Error<'a>>;
@@ -137,6 +140,15 @@ impl<'a> From<&'a str> for Token<'a> {
     }
 }
 
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum VectorFilter<'a> {
+    Fragment(Token<'a>),
+    DocumentTemplate,
+    UserProvided,
+    Regenerate,
+    None,
+}
+
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub enum FilterCondition<'a> {
     Not(Box<Self>),
@@ -144,6 +156,7 @@ pub enum FilterCondition<'a> {
     In { fid: Token<'a>, els: Vec<Token<'a>> },
     Or(Vec<Self>),
     And(Vec<Self>),
+    VectorExists { fid: Token<'a>, embedder: Option<Token<'a>>, filter: VectorFilter<'a> },
     GeoLowerThan { point: [Token<'a>; 2], radius: Token<'a> },
     GeoBoundingBox { top_right_point: [Token<'a>; 2], bottom_left_point: [Token<'a>; 2] },
     GeoPolygon { points: Vec<[Token<'a>; 2]> },
@@ -167,18 +180,34 @@ impl<'a> FilterCondition<'a> {
                 | Condition::Exists
                 | Condition::LowerThan(_)
                 | Condition::LowerThanOrEqual(_)
-                | Condition::Between { .. } => None,
-                Condition::Contains { keyword, word: _ }
-                | Condition::StartsWith { keyword, word: _ } => Some(keyword),
+                | Condition::Between { .. }
+                | Condition::StartsWith { .. } => None,
+                Condition::Contains { keyword, word: _ } => Some(keyword),
             },
             FilterCondition::Not(this) => this.use_contains_operator(),
             FilterCondition::Or(seq) | FilterCondition::And(seq) => {
                 seq.iter().find_map(|filter| filter.use_contains_operator())
             }
+            FilterCondition::VectorExists { .. }
+            | FilterCondition::GeoLowerThan { .. }
+            | FilterCondition::GeoBoundingBox { .. }
+            | FilterCondition::GeoPolygon { .. }
+            | FilterCondition::In { .. } => None,
+        }
+    }
+
+    pub fn use_vector_filter(&self) -> Option<&Token> {
+        match self {
+            FilterCondition::Condition { .. } => None,
+            FilterCondition::Not(this) => this.use_vector_filter(),
+            FilterCondition::Or(seq) | FilterCondition::And(seq) => {
+                seq.iter().find_map(|filter| filter.use_vector_filter())
+            }
             FilterCondition::GeoLowerThan { .. }
             | FilterCondition::GeoBoundingBox { .. }
             | FilterCondition::GeoPolygon { .. }
             | FilterCondition::In { .. } => None,
+            FilterCondition::VectorExists { fid, .. } => Some(fid),
         }
     }
 
@@ -266,10 +295,7 @@ fn parse_in_body(input: Span) -> IResult<Vec<Token>> {
     let (input, _) = ws(word_exact("IN"))(input)?;
 
     // everything after `IN` can be a failure
-    let (input, _) =
-        cut_with_err(tag("["), |_| Error::new_from_kind(input, ErrorKind::InOpeningBracket))(
-            input,
-        )?;
+    let (input, _) = tag("[")(input).map_cut(ErrorKind::InOpeningBracket)?;
 
     let (input, content) = cut(parse_value_list)(input)?;
 
@@ -415,7 +441,7 @@ fn parse_geo_bounding_box(input: Span) -> IResult<FilterCondition> {
     let (input, args) = parsed?;
 
     if args.len() != 2 || args[0].len() != 2 || args[1].len() != 2 {
-        return Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::GeoBoundingBox)));
+        return Err(Error::failure_from_kind(input, ErrorKind::GeoBoundingBox));
     }
 
     let res = FilterCondition::GeoBoundingBox {
@@ -468,7 +494,7 @@ fn parse_geo_point(input: Span) -> IResult<FilterCondition> {
     ))(input)
     .map_err(|e| e.map(|_| Error::new_from_kind(input, ErrorKind::ReservedGeo("_geoPoint"))))?;
     // if we succeeded we still return a `Failure` because geoPoints are not allowed
-    Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::ReservedGeo("_geoPoint"))))
+    Err(Error::failure_from_kind(input, ErrorKind::ReservedGeo("_geoPoint")))
 }
 
 /// geoPoint      = WS* "_geoDistance(float WS* "," WS* float WS* "," WS* float)
@@ -482,7 +508,7 @@ fn parse_geo_distance(input: Span) -> IResult<FilterCondition> {
     ))(input)
     .map_err(|e| e.map(|_| Error::new_from_kind(input, ErrorKind::ReservedGeo("_geoDistance"))))?;
     // if we succeeded we still return a `Failure` because `geoDistance` filters are not allowed
-    Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::ReservedGeo("_geoDistance"))))
+    Err(Error::failure_from_kind(input, ErrorKind::ReservedGeo("_geoDistance")))
 }
 
 /// geo      = WS* "_geo(float WS* "," WS* float WS* "," WS* float)
@@ -496,7 +522,7 @@ fn parse_geo(input: Span) -> IResult<FilterCondition> {
     ))(input)
     .map_err(|e| e.map(|_| Error::new_from_kind(input, ErrorKind::ReservedGeo("_geo"))))?;
     // if we succeeded we still return a `Failure` because `_geo` filter is not allowed
-    Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::ReservedGeo("_geo"))))
+    Err(Error::failure_from_kind(input, ErrorKind::ReservedGeo("_geo")))
 }
 
 fn parse_error_reserved_keyword(input: Span) -> IResult<FilterCondition> {
@@ -535,8 +561,7 @@ fn parse_primary(input: Span, depth: usize) -> IResult<FilterCondition> {
         parse_is_not_null,
         parse_is_empty,
         parse_is_not_empty,
-        parse_exists,
-        parse_not_exists,
+        alt((parse_vectors_exists, parse_exists, parse_not_exists)),
         parse_to,
         parse_contains,
         parse_not_contains,
@@ -592,6 +617,22 @@ impl std::fmt::Display for FilterCondition<'_> {
                 }
                 write!(f, "]")
             }
+            FilterCondition::VectorExists { fid: _, embedder, filter: inner } => {
+                write!(f, "_vectors")?;
+                if let Some(embedder) = embedder {
+                    write!(f, ".{:?}", embedder.value())?;
+                }
+                match inner {
+                    VectorFilter::Fragment(fragment) => {
+                        write!(f, ".fragments.{:?}", fragment.value())?
+                    }
+                    VectorFilter::DocumentTemplate => write!(f, ".documentTemplate")?,
+                    VectorFilter::UserProvided => write!(f, ".userProvided")?,
+                    VectorFilter::Regenerate => write!(f, ".regenerate")?,
+                    VectorFilter::None => (),
+                }
+                write!(f, " EXISTS")
+            }
             FilterCondition::GeoLowerThan { point, radius } => {
                 write!(f, "_geoRadius({}, {}, {})", point[0], point[1], radius)
             }
@@ -672,6 +713,9 @@ pub mod tests {
         insta::assert_snapshot!(p(r"title = 'foo\\\\\\\\'"), @r#"{title} = {foo\\\\}"#);
         // but it also works with other sequences
         insta::assert_snapshot!(p(r#"title = 'foo\x20\n\t\"\'"'"#), @"{title} = {foo \n\t\"\'\"}");
+
+        insta::assert_snapshot!(p(r#"_vectors." valid.name  ".fragments."also.. valid! " EXISTS"#), @r#"_vectors." valid.name  ".fragments."also.. valid! " EXISTS"#);
+        insta::assert_snapshot!(p("_vectors.\"\n\t\r\\\"\" EXISTS"), @r#"_vectors."\n\t\r\"" EXISTS"#);
     }
 
     #[test]
@@ -734,6 +778,18 @@ pub mod tests {
         insta::assert_snapshot!(p("NOT subscribers IS NOT EMPTY"), @"{subscribers} IS EMPTY");
         insta::assert_snapshot!(p("subscribers  IS   NOT   EMPTY"), @"NOT ({subscribers} IS EMPTY)");
 
+        // Test _vectors EXISTS + _vectors NOT EXITS
+        insta::assert_snapshot!(p("_vectors EXISTS"), @"_vectors EXISTS");
+        insta::assert_snapshot!(p("_vectors.embedderName EXISTS"), @r#"_vectors."embedderName" EXISTS"#);
+        insta::assert_snapshot!(p("_vectors.embedderName.documentTemplate EXISTS"), @r#"_vectors."embedderName".documentTemplate EXISTS"#);
+        insta::assert_snapshot!(p("_vectors.embedderName.regenerate EXISTS"), @r#"_vectors."embedderName".regenerate EXISTS"#);
+        insta::assert_snapshot!(p("_vectors.embedderName.regenerate EXISTS"), @r#"_vectors."embedderName".regenerate EXISTS"#);
+        insta::assert_snapshot!(p("_vectors.embedderName.fragments.fragmentName EXISTS"), @r#"_vectors."embedderName".fragments."fragmentName" EXISTS"#);
+        insta::assert_snapshot!(p("  _vectors.embedderName.fragments.fragmentName   EXISTS"), @r#"_vectors."embedderName".fragments."fragmentName" EXISTS"#);
+        insta::assert_snapshot!(p("NOT _vectors EXISTS"), @"NOT (_vectors EXISTS)");
+        insta::assert_snapshot!(p(" NOT  _vectors   EXISTS"), @"NOT (_vectors EXISTS)");
+        insta::assert_snapshot!(p("  _vectors  NOT  EXISTS"), @"NOT (_vectors EXISTS)");
+
         // Test EXISTS + NOT EXITS
         insta::assert_snapshot!(p("subscribers EXISTS"), @"{subscribers} EXISTS");
         insta::assert_snapshot!(p("NOT subscribers EXISTS"), @"NOT ({subscribers} EXISTS)");
@@ -988,6 +1044,71 @@ pub mod tests {
         "###
         );
 
+        insta::assert_snapshot!(p(r#"_vectors _vectors EXISTS"#), @r"
+        Was expecting an operation like `EXISTS` or `NOT EXISTS` after the vector filter.
+        10:25 _vectors _vectors EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors. embedderName EXISTS"#), @r"
+        Was expecting embedder name but found nothing.
+        10:11 _vectors. embedderName EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors .embedderName EXISTS"#), @r"
+        Was expecting an operation like `EXISTS` or `NOT EXISTS` after the vector filter.
+        10:30 _vectors .embedderName EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName. EXISTS"#), @r"
+        Was expecting one of `.fragments`, `.userProvided`, `.documentTemplate`, `.regenerate` or nothing, but instead found a point without a valid value.
+        22:23 _vectors.embedderName. EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors."embedderName EXISTS"#), @r#"
+        The quotes in one of the values are inconsistent.
+        10:30 _vectors."embedderName EXISTS
+        "#);
+        insta::assert_snapshot!(p(r#"_vectors."embedderNam"e EXISTS"#), @r#"
+        The vector filter has leftover tokens.
+        23:31 _vectors."embedderNam"e EXISTS
+        "#);
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.documentTemplate. EXISTS"#), @r"
+        Was expecting one of `.fragments`, `.userProvided`, `.documentTemplate`, `.regenerate` or nothing, but instead found a point without a valid value.
+        39:40 _vectors.embedderName.documentTemplate. EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.fragments EXISTS"#), @r"
+        The vector filter is missing a fragment name.
+        32:39 _vectors.embedderName.fragments EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.fragments. EXISTS"#), @r"
+        The vector filter's fragment name is invalid.
+        33:40 _vectors.embedderName.fragments. EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.fragments.test test EXISTS"#), @r"
+        Was expecting an operation like `EXISTS` or `NOT EXISTS` after the vector filter.
+        38:49 _vectors.embedderName.fragments.test test EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.fragments. test EXISTS"#), @r"
+        The vector filter's fragment name is invalid.
+        33:45 _vectors.embedderName.fragments. test EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName .fragments. test EXISTS"#), @r"
+        Was expecting an operation like `EXISTS` or `NOT EXISTS` after the vector filter.
+        23:46 _vectors.embedderName .fragments. test EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName .fragments.test EXISTS"#), @r"
+        Was expecting an operation like `EXISTS` or `NOT EXISTS` after the vector filter.
+        23:45 _vectors.embedderName .fragments.test EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.fargments.test EXISTS"#), @r"
+        Was expecting one of `fragments`, `userProvided`, `documentTemplate`, `regenerate` or nothing, but instead found `fargments`. Did you mean `fragments`?
+        23:32 _vectors.embedderName.fargments.test EXISTS
+        ");
+        insta::assert_snapshot!(p(r#"_vectors.embedderName."userProvided" EXISTS"#), @r#"
+        Was expecting this part to be unquoted.
+        24:36 _vectors.embedderName."userProvided" EXISTS
+        "#);
+        insta::assert_snapshot!(p(r#"_vectors.embedderName.userProvided.fragments.test EXISTS"#), @r"
+        Vector filter can only accept one of `fragments`, `userProvided`, `documentTemplate` or `regenerate`, but found both `userProvided` and `fragments`.
+        36:45 _vectors.embedderName.userProvided.fragments.test EXISTS
+        ");
+
         insta::assert_snapshot!(p(r#"NOT OR EXISTS AND EXISTS NOT EXISTS"#), @r###"
         Was expecting a value but instead got `OR`, which is a reserved keyword. To use `OR` as a field name or a value, surround it by quotes.
         5:7 NOT OR EXISTS AND EXISTS NOT EXISTS

crates/filter-parser/src/value.rs

@@ -80,6 +80,51 @@ pub fn word_exact<'a, 'b: 'a>(tag: &'b str) -> impl Fn(Span<'a>) -> IResult<'a,
     }
 }
 
+/// vector_value          = ( non_dot_word | singleQuoted | doubleQuoted)
+pub fn parse_vector_value(input: Span) -> IResult<Token> {
+    pub fn non_dot_word(input: Span) -> IResult<Token> {
+        let (input, word) = take_while1(|c| is_value_component(c) && c != '.')(input)?;
+        Ok((input, word.into()))
+    }
+
+    let (input, value) = alt((
+        delimited(char('\''), cut(|input| quoted_by('\'', input)), cut(char('\''))),
+        delimited(char('"'), cut(|input| quoted_by('"', input)), cut(char('"'))),
+        non_dot_word,
+    ))(input)?;
+
+    match unescaper::unescape(value.value()) {
+        Ok(content) => {
+            if content.len() != value.value().len() {
+                Ok((input, Token::new(value.original_span(), Some(content))))
+            } else {
+                Ok((input, value))
+            }
+        }
+        Err(unescaper::Error::IncompleteStr(_)) => Err(nom::Err::Incomplete(nom::Needed::Unknown)),
+        Err(unescaper::Error::ParseIntError { .. }) => Err(nom::Err::Error(Error::new_from_kind(
+            value.original_span(),
+            ErrorKind::InvalidEscapedNumber,
+        ))),
+        Err(unescaper::Error::InvalidChar { .. }) => Err(nom::Err::Error(Error::new_from_kind(
+            value.original_span(),
+            ErrorKind::MalformedValue,
+        ))),
+    }
+}
+
+pub fn parse_vector_value_cut<'a>(input: Span<'a>, kind: ErrorKind<'a>) -> IResult<'a, Token<'a>> {
+    parse_vector_value(input).map_err(|e| match e {
+        nom::Err::Failure(e) => match e.kind() {
+            ErrorKind::Char(c) if *c == '"' || *c == '\'' => {
+                crate::Error::failure_from_kind(input, ErrorKind::VectorFilterInvalidQuotes)
+            }
+            _ => crate::Error::failure_from_kind(input, kind),
+        },
+        _ => crate::Error::failure_from_kind(input, kind),
+    })
+}
+
 /// value          = WS* ( word | singleQuoted | doubleQuoted) WS+
 pub fn parse_value(input: Span) -> IResult<Token> {
     // to get better diagnostic message we are going to strip the left whitespaces from the input right now
@@ -99,31 +144,21 @@ pub fn parse_value(input: Span) -> IResult<Token> {
     }
 
     match parse_geo_radius(input) {
-        Ok(_) => {
-            return Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::MisusedGeoRadius)))
-        }
+        Ok(_) => return Err(Error::failure_from_kind(input, ErrorKind::MisusedGeoRadius)),
         // if we encountered a failure it means the user badly wrote a _geoRadius filter.
         // But instead of showing them how to fix his syntax we are going to tell them they should not use this filter as a value.
         Err(e) if e.is_failure() => {
-            return Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::MisusedGeoRadius)))
+            return Err(Error::failure_from_kind(input, ErrorKind::MisusedGeoRadius))
         }
         _ => (),
     }
 
     match parse_geo_bounding_box(input) {
-        Ok(_) => {
-            return Err(nom::Err::Failure(Error::new_from_kind(
-                input,
-                ErrorKind::MisusedGeoBoundingBox,
-            )))
-        }
+        Ok(_) => return Err(Error::failure_from_kind(input, ErrorKind::MisusedGeoBoundingBox)),
         // if we encountered a failure it means the user badly wrote a _geoBoundingBox filter.
         // But instead of showing them how to fix his syntax we are going to tell them they should not use this filter as a value.
         Err(e) if e.is_failure() => {
-            return Err(nom::Err::Failure(Error::new_from_kind(
-                input,
-                ErrorKind::MisusedGeoBoundingBox,
-            )))
+            return Err(Error::failure_from_kind(input, ErrorKind::MisusedGeoBoundingBox))
         }
         _ => (),
     }