feat(targets): complex targets sets expressions

allows intersection and complementary

src/nix/node_filter.rs

@@ -2,7 +2,7 @@
 
 use std::collections::HashSet;
 use std::convert::AsRef;
-use std::iter::{FromIterator, Iterator};
+use std::iter::Iterator;
 use std::str::FromStr;
 
 use clap::Args;
@@ -28,22 +28,26 @@ The list is comma-separated and globs are supported. To match tags, prepend the
     pub on: Option<NodeFilter>,
 }
 
-/// A node filter containing a list of rules.
-#[derive(Clone, Debug)]
-pub struct NodeFilter {
-    rules: Vec<Rule>,
-}
-
 /// A filter rule.
-///
-/// The filter rules are OR'd together.
 #[derive(Debug, Clone, Eq, PartialEq)]
-enum Rule {
+pub enum NodeFilter {
     /// Matches a node's attribute name.
     MatchName(GlobPattern),
 
     /// Matches a node's `deployment.tags`.
     MatchTag(GlobPattern),
+
+    /// Matches an Union
+    Union(Vec<Box<NodeFilter>>),
+
+    /// Matches an Intersection
+    Inter(Vec<Box<NodeFilter>>),
+
+    /// Matches the complementary
+    Not(Box<NodeFilter>),
+
+    /// Empty
+    Empty,
 }
 
 impl FromStr for NodeFilter {
@@ -53,7 +57,169 @@ impl FromStr for NodeFilter {
     }
 }
 
+#[inline]
+fn end_delimiter(c: char) -> bool {
+    [',', '&', ')'].contains(&c)
+}
+
 impl NodeFilter {
+    fn and(a: Self, b: Self) -> Self {
+        match (a, b) {
+            (Self::Inter(mut av), Self::Inter(mut bv)) => {
+                av.append(&mut bv);
+                Self::Inter(av)
+            }
+            (Self::Inter(mut av), b) => {
+                av.push(Box::new(b));
+                Self::Inter(av)
+            }
+            (a, Self::Inter(mut bv)) => {
+                bv.push(Box::new(a));
+                Self::Inter(bv)
+            }
+            (a, b) => Self::Inter(vec![Box::new(a), Box::new(b)]),
+        }
+    }
+
+    fn or(a: Self, b: Self) -> Self {
+        match (a, b) {
+            (Self::Union(mut av), Self::Union(mut bv)) => {
+                av.append(&mut bv);
+                Self::Union(av)
+            }
+            (Self::Union(mut av), b) => {
+                av.push(Box::new(b));
+                Self::Union(av)
+            }
+            (a, Self::Union(mut bv)) => {
+                bv.push(Box::new(a));
+                Self::Union(bv)
+            }
+            (a, b) => Self::Union(vec![Box::new(a), Box::new(b)]),
+        }
+    }
+
+    fn not(a: Self) -> Self {
+        if let Self::Not(ae) = a {
+            *ae
+        } else {
+            Self::Not(Box::new(a))
+        }
+    }
+
+    /// Parses an elementary expression,
+    /// that is base tags and name, with expression between parentheses
+    /// Negations are also parsed here as the most prioritary operation
+    ///
+    /// It returns the unparsed text that follows
+    fn parse_expr0(unparsed: &str) -> ColmenaResult<(Self, &str)> {
+        let unparsed = unparsed.trim_start();
+        // Negation
+        if let Some(negated_expr) = unparsed.strip_prefix('!') {
+            let (negated, unparsed) = Self::parse_expr0(negated_expr)?;
+            Ok((Self::not(negated), unparsed))
+        } else
+        // parentheses
+        if let Some(parenthesed_expr) = unparsed.strip_prefix('(') {
+            let (interior, unparsed) = Self::parse_expr2(parenthesed_expr)?;
+            Ok((
+                interior,
+                unparsed.strip_prefix(')').ok_or(ColmenaError::Unknown {
+                    message: format!("Expected a closing parenthesis at {:?}.", unparsed),
+                })?,
+            ))
+        } else
+        // tag
+        if let Some(tag_expr) = unparsed.strip_prefix('@') {
+            match tag_expr
+                .find(end_delimiter)
+                .map(|idx| tag_expr.split_at(idx))
+                .map(|(tag, end)| (tag.trim_end(), end))
+            {
+                Some((tag, unparsed)) => {
+                    if tag.is_empty() {
+                        return Err(ColmenaError::EmptyFilterRule);
+                    } else {
+                        Ok((Self::MatchTag(GlobPattern::new(tag).unwrap()), unparsed))
+                    }
+                }
+                None => {
+                    let tag_expr = tag_expr.trim_end();
+                    if tag_expr.is_empty() {
+                        Err(ColmenaError::EmptyFilterRule)
+                    } else {
+                        Ok((Self::MatchTag(GlobPattern::new(tag_expr).unwrap()), ""))
+                    }
+                }
+            }
+        } else
+        //node name
+        {
+            match unparsed
+                .find(end_delimiter)
+                .map(|idx| unparsed.split_at(idx))
+                .map(|(tag, end)| (tag.trim_end(), end))
+            {
+                Some((name, unparsed)) => {
+                    if name.is_empty() {
+                        Err(ColmenaError::EmptyFilterRule)
+                    } else {
+                        Ok((Self::MatchName(GlobPattern::new(name).unwrap()), unparsed))
+                    }
+                }
+                None => {
+                    let unparsed = unparsed.trim_end();
+                    if unparsed.is_empty() {
+                        Err(ColmenaError::EmptyFilterRule)
+                    } else {
+                        Ok((Self::MatchName(GlobPattern::new(unparsed).unwrap()), ""))
+                    }
+                }
+            }
+        }
+    }
+
+    /// Parses the union operations between elementary expression.
+    ///
+    /// It returns the unparsed text that follows
+    fn parse_op1(acc: Self, unparsed: &str) -> ColmenaResult<(Self, &str)> {
+        let unparsed = unparsed.trim_start();
+        if let Some(unions) = unparsed.strip_prefix(',') {
+            let (base_expr, unparsed) = Self::parse_expr0(unions)?;
+            Self::parse_op1(Self::or(acc, base_expr), unparsed)
+        } else {
+            Ok((acc, unparsed))
+        }
+    }
+
+    /// Parses elementary expression and their unions.
+    ///
+    /// It returns the unparsed text that follows
+    fn parse_expr1(unparsed: &str) -> ColmenaResult<(Self, &str)> {
+        let (base_expr, unparsed) = Self::parse_expr0(unparsed)?;
+        Self::parse_op1(base_expr, unparsed)
+    }
+
+    /// Parses the intersection operations between unions.
+    ///
+    /// It returns the unparsed text that follows
+    fn parse_op2(acc: Self, unparsed: &str) -> ColmenaResult<(Self, &str)> {
+        if let Some(intersections) = unparsed.strip_prefix('&') {
+            let (union, unparsed) = Self::parse_expr1(intersections)?;
+            Self::parse_op2(Self::and(acc, union), unparsed)
+        } else {
+            Ok((acc, unparsed))
+        }
+    }
+
+    /// Parses a complete expression
+    ///
+    /// It returns the unparsed text that follows
+    fn parse_expr2(unparsed: &str) -> ColmenaResult<(Self, &str)> {
+        let (union, unparsed) = Self::parse_expr1(unparsed)?;
+        Self::parse_op2(union, unparsed)
+    }
+
     /// Creates a new filter using an expression passed using `--on`.
     pub fn new<S: AsRef<str>>(filter: S) -> ColmenaResult<Self> {
         let filter = filter.as_ref();
@@ -62,29 +228,16 @@ impl NodeFilter {
         if trimmed.is_empty() {
             log::warn!("Filter \"{}\" is blank and will match nothing", filter);
 
-            return Ok(Self { rules: Vec::new() });
-        }
-
-        let rules = trimmed
-            .split(',')
-            .map(|pattern| {
-                let pattern = pattern.trim();
-
-                if pattern.is_empty() {
-                    return Err(ColmenaError::EmptyFilterRule);
-                }
-
-                if let Some(tag_pattern) = pattern.strip_prefix('@') {
-                    Ok(Rule::MatchTag(GlobPattern::new(tag_pattern).unwrap()))
-                } else {
-                    Ok(Rule::MatchName(GlobPattern::new(pattern).unwrap()))
+            return Ok(Self::Empty);
         }
+        let (target_filter, unparsed) = Self::parse_expr2(trimmed)?;
+        if unparsed != "" {
+            Err(ColmenaError::Unknown {
+                message: format!("Found garbage {:?} when parsing the node filter.", unparsed),
             })
-            .collect::<Vec<ColmenaResult<Rule>>>();
-
-        let rules = Result::from_iter(rules)?;
-
-        Ok(Self { rules })
+        } else {
+            Ok(target_filter)
+        }
     }
 
     /// Returns whether the filter has any rule matching NodeConfig information.
@@ -93,7 +246,31 @@ impl NodeFilter {
     /// especially when its values (e.g., tags) depend on other parts of
     /// the configuration.
     pub fn has_node_config_rules(&self) -> bool {
-        self.rules.iter().any(|rule| rule.matches_node_config())
+        match self {
+            Self::MatchName(_) => false,
+            Self::MatchTag(_) => true,
+            Self::Union(v) => v.iter().any(|e| e.has_node_config_rules()),
+            Self::Inter(v) => v.iter().any(|e| e.has_node_config_rules()),
+            Self::Not(e) => e.has_node_config_rules(),
+            Self::Empty => false,
+        }
+    }
+
+    /// Decides whether a node is accepted by the filter or not.
+    /// panic if the filter depends on tags and config is None
+    fn is_accepted(&self, name: &NodeName, config: Option<&NodeConfig>) -> bool {
+        match self {
+            Self::MatchName(pat) => pat.matches(name.as_str()),
+            Self::MatchTag(pat) => config
+                .unwrap()
+                .tags()
+                .iter()
+                .any(|tag| pat.matches(tag.as_str())),
+            Self::Union(v) => v.iter().any(|e| e.is_accepted(name, config)),
+            Self::Inter(v) => v.iter().all(|e| e.is_accepted(name, config)),
+            Self::Not(e) => !e.is_accepted(name, config),
+            Self::Empty => false,
+        }
     }
 
     /// Runs the filter against a set of NodeConfigs and returns the matched ones.
@@ -101,30 +278,17 @@ impl NodeFilter {
     where
         I: Iterator<Item = (&'a NodeName, &'a NodeConfig)>,
     {
-        if self.rules.is_empty() {
+        if self == &Self::Empty {
             return HashSet::new();
         }
 
         nodes
             .filter_map(|(name, node)| {
-                for rule in self.rules.iter() {
-                    match rule {
-                        Rule::MatchName(pat) => {
-                            if pat.matches(name.as_str()) {
-                                return Some(name);
-                            }
-                        }
-                        Rule::MatchTag(pat) => {
-                            for tag in node.tags() {
-                                if pat.matches(tag) {
-                                    return Some(name);
-                                }
-                            }
-                        }
-                    }
-                }
-
+                if self.is_accepted(name, Some(node)) {
+                    Some(name)
+                } else {
                     None
+                }
             })
             .cloned()
             .collect()
@@ -132,32 +296,24 @@ impl NodeFilter {
 
     /// Runs the filter against a set of node names and returns the matched ones.
     pub fn filter_node_names(&self, nodes: &[NodeName]) -> ColmenaResult<HashSet<NodeName>> {
-        nodes.iter().filter_map(|name| -> Option<ColmenaResult<NodeName>> {
-            for rule in self.rules.iter() {
-                match rule {
-                    Rule::MatchName(pat) => {
-                        if pat.matches(name.as_str()) {
-                            return Some(Ok(name.clone()));
-                        }
-                    }
-                    _ => {
-                        return Some(Err(ColmenaError::Unknown {
-                            message: format!("Not enough information to run rule {:?} - We only have node names", rule),
-                        }));
-                    }
-                }
-            }
+        if self.has_node_config_rules() {
+            Err(ColmenaError::Unknown {
+                message: format!(
+                    "Not enough information to run rule {:?} - We only have node names",
+                    self
+                ),
+            })
+        } else {
+            Ok(nodes
+                .iter()
+                .filter_map(|name| {
+                    if self.is_accepted(name, None) {
+                        Some(name.clone())
+                    } else {
                         None
-        }).collect()
                     }
-}
-
-impl Rule {
-    /// Returns whether the rule matches against the NodeConfig (i.e., `config.deployment`).
-    pub fn matches_node_config(&self) -> bool {
-        match self {
-            Self::MatchTag(_) => true,
-            Self::MatchName(_) => false,
+                })
+                .collect())
         }
     }
 }
@@ -177,13 +333,13 @@ mod tests {
     #[test]
     fn test_empty_filter() {
         let filter = NodeFilter::new("").unwrap();
-        assert_eq!(0, filter.rules.len());
+        assert_eq!(NodeFilter::Empty, filter);
 
         let filter = NodeFilter::new("\t").unwrap();
-        assert_eq!(0, filter.rules.len());
+        assert_eq!(NodeFilter::Empty, filter);
 
         let filter = NodeFilter::new("    ").unwrap();
-        assert_eq!(0, filter.rules.len());
+        assert_eq!(NodeFilter::Empty, filter);
     }
 
     #[test]
@@ -197,21 +353,73 @@ mod tests {
     fn test_filter_rule_mixed() {
         let filter = NodeFilter::new("@router,gamma-*").unwrap();
         assert_eq!(
-            vec![
-                Rule::MatchTag(GlobPattern::new("router").unwrap()),
-                Rule::MatchName(GlobPattern::new("gamma-*").unwrap()),
-            ],
-            filter.rules,
+            NodeFilter::Union(vec![
+                Box::new(NodeFilter::MatchTag(GlobPattern::new("router").unwrap())),
+                Box::new(NodeFilter::MatchName(GlobPattern::new("gamma-*").unwrap())),
+            ]),
+            filter,
         );
 
         let filter = NodeFilter::new("a, \t@b ,    c-*").unwrap();
         assert_eq!(
-            vec![
-                Rule::MatchName(GlobPattern::new("a").unwrap()),
-                Rule::MatchTag(GlobPattern::new("b").unwrap()),
-                Rule::MatchName(GlobPattern::new("c-*").unwrap()),
-            ],
-            filter.rules,
+            NodeFilter::Union(vec![
+                Box::new(NodeFilter::MatchName(GlobPattern::new("a").unwrap())),
+                Box::new(NodeFilter::MatchTag(GlobPattern::new("b").unwrap())),
+                Box::new(NodeFilter::MatchName(GlobPattern::new("c-*").unwrap())),
+            ]),
+            filter,
+        );
+
+        let filter = NodeFilter::new("a  & \t@b ,    c-*").unwrap();
+        assert_eq!(
+            NodeFilter::Inter(vec![
+                Box::new(NodeFilter::MatchName(GlobPattern::new("a").unwrap())),
+                Box::new(NodeFilter::Union(vec![
+                    Box::new(NodeFilter::MatchTag(GlobPattern::new("b").unwrap())),
+                    Box::new(NodeFilter::MatchName(GlobPattern::new("c-*").unwrap())),
+                ])),
+            ]),
+            filter,
+        );
+
+        let filter = NodeFilter::new("( a  & \t@b ) ,  c-*").unwrap();
+        assert_eq!(
+            NodeFilter::Union(vec![
+                Box::new(NodeFilter::Inter(vec![
+                    Box::new(NodeFilter::MatchName(GlobPattern::new("a").unwrap())),
+                    Box::new(NodeFilter::MatchTag(GlobPattern::new("b").unwrap())),
+                ])),
+                Box::new(NodeFilter::MatchName(GlobPattern::new("c-*").unwrap())),
+            ]),
+            filter,
+        );
+
+        let filter = NodeFilter::new("( a  & \t@b ) ,  !  c-*").unwrap();
+        assert_eq!(
+            NodeFilter::Union(vec![
+                Box::new(NodeFilter::Inter(vec![
+                    Box::new(NodeFilter::MatchName(GlobPattern::new("a").unwrap())),
+                    Box::new(NodeFilter::MatchTag(GlobPattern::new("b").unwrap())),
+                ])),
+                Box::new(NodeFilter::Not(Box::new(NodeFilter::MatchName(
+                    GlobPattern::new("c-*").unwrap()
+                )))),
+            ]),
+            filter,
+        );
+
+        let filter = NodeFilter::new("( a  & \t@b ) ,  !!!  c-*").unwrap();
+        assert_eq!(
+            NodeFilter::Union(vec![
+                Box::new(NodeFilter::Inter(vec![
+                    Box::new(NodeFilter::MatchName(GlobPattern::new("a").unwrap())),
+                    Box::new(NodeFilter::MatchTag(GlobPattern::new("b").unwrap())),
+                ])),
+                Box::new(NodeFilter::Not(Box::new(NodeFilter::MatchName(
+                    GlobPattern::new("c-*").unwrap()
+                )))),
+            ]),
+            filter,
         );
     }
 
@@ -250,6 +458,7 @@ mod tests {
             privilege_escalation_command: vec![],
             extra_ssh_options: vec![],
             keys: HashMap::new(),
+            system_type: None,
         };
 
         let mut nodes = HashMap::new();
@@ -315,5 +524,26 @@ mod tests {
                 .unwrap()
                 .filter_node_configs(nodes.iter()),
         );
+
+        assert_eq!(
+            &HashSet::from_iter([]),
+            &NodeFilter::new("@router&@controller")
+                .unwrap()
+                .filter_node_configs(nodes.iter()),
+        );
+
+        assert_eq!(
+            &HashSet::from_iter([node!("beta")]),
+            &NodeFilter::new("@router&@infra-*")
+                .unwrap()
+                .filter_node_configs(nodes.iter()),
+        );
+
+        assert_eq!(
+            &HashSet::from_iter([node!("alpha")]),
+            &NodeFilter::new("!@router&@infra-*")
+                .unwrap()
+                .filter_node_configs(nodes.iter()),
+        );
     }
 }