Make smart rule exclusion cover rule chains of any depth

src/config.rs

@@ -18,5 +18,9 @@ fn default_folders() -> Vec<String> {
 pub fn read_config() -> Config {
     let contents =
         fs::read_to_string("makeultra.toml").expect("ERROR: Could not read makeultra.toml.");
-    toml::from_str(&contents).unwrap()
+    let mut config: Config = toml::from_str(&contents).unwrap();
+    for rule in &mut config.rules {
+        rule.compile_output_shape();
+    }
+    config
 }

src/main.rs

@@ -355,16 +355,37 @@ fn generate_children(path: String) {
                     new_file = rule.get_output(&path).to_string();
 
                     // "Smart" rule exclusion.
-                    // TODO: Make recursive instead of only evaluating the child rules
                     // Imagine we have a rule that would match *.js, turning it into *.min.js.
                     // Now imagine we have another rule that does something to *.min.js files.
                     // Given the file a.min.js, we need to be able to determine that only the
-                    // *.min.js rule should run:
-                    if matching_rules.len() > 1 {
-                        let new_file_rules = get_matching_rules(&new_file);
-                        if new_file_rules.eq(matching_rules) {
+                    // *.min.js rule should run.
+                    //
+                    // Rules that rewrite a file in-place are exempt: they are
+                    // stored as self-loops in the graph and cannot grow a rule
+                    // chain, so there is never anything to exclude.
+                    if new_file != *path {
+                        // If `path` already looks like an output of this rule,
+                        // the rule was already applied somewhere along the
+                        // chain that produced `path`, and applying it again
+                        // would only pile up endless derivatives (a.min.min.js,
+                        // a.min.min.min.js, ...). Because every file of a chain
+                        // is checked against the output shapes of all of its
+                        // matching rules, this covers chains of any depth
+                        // (e.g. *.js -> *.min.js -> *.pkg.min.js).
+                        if rule.matches_own_output(path) {
                             return;
                         }
+
+                        // Fall back to comparing rule sets, which also covers
+                        // `to` templates that only rewrite part of a path: if
+                        // the output file matches exactly the same rules as
+                        // `path`, applying the rule makes no progress.
+                        if matching_rules.len() > 1 {
+                            let new_file_rules = get_matching_rules(&new_file);
+                            if new_file_rules.eq(matching_rules) {
+                                return;
+                            }
+                        }
                     }
 
                     let mut files = FILES.write();

src/rule.rs

@@ -10,6 +10,11 @@ pub struct Rule {
     #[serde(default)]
     pub exclude: Option<Regex>,
     pub command: String,
+    /// Regex matching every path that looks like an output of this rule,
+    /// derived from `to`. Compiled up-front by `compile_output_shape`;
+    /// `matches_own_output` falls back to compiling on the fly when unset.
+    #[serde(skip)]
+    output_shape: Option<Regex>,
 }
 
 impl Rule {
@@ -28,6 +33,82 @@ impl Rule {
         }
         false
     }
+
+    /// Compiles and caches the regex used by `matches_own_output`. Called
+    /// once per rule after the config is read.
+    pub fn compile_output_shape(&mut self) {
+        self.output_shape = Some(output_shape_pattern(&self.to));
+    }
+
+    /// Returns `true` if `path` looks like a file this rule already
+    /// produced, i.e. it matches the shape of the rule's `to` template.
+    /// Given `to = '$name.min.js'`, this matches `a.min.js`, `a.pkg.min.js`,
+    /// and `a.min.min.js`, but not `a.js` or `a.min.js.br`.
+    pub fn matches_own_output(&self, path: &str) -> bool {
+        match &self.output_shape {
+            Some(shape) => shape.is_match(path),
+            None => output_shape_pattern(&self.to).is_match(path),
+        }
+    }
+}
+
+/// Converts a rule's `to` replacement template into a regex matching every
+/// path the template could have produced: capture group references
+/// (`$name`, `${name}`, `$1`) become `.*`, `$$` becomes a literal `$`, and
+/// everything else is matched literally. The pattern is anchored at both
+/// ends, so templates that only rewrite part of a path produce a shape
+/// that simply never matches.
+fn output_shape_pattern(to: &str) -> Regex {
+    let mut pattern = String::from("^");
+    let mut chars = to.chars().peekable();
+    while let Some(c) = chars.next() {
+        if c != '$' {
+            pattern.push_str(&regex::escape(&c.to_string()));
+            continue;
+        }
+        match chars.peek() {
+            // `$$` is an escaped literal `$`.
+            Some('$') => {
+                chars.next();
+                pattern.push_str("\\$");
+            }
+            // `${name}`
+            Some('{') => {
+                chars.next();
+                let mut name = String::new();
+                let mut closed = false;
+                for c in chars.by_ref() {
+                    if c == '}' {
+                        closed = true;
+                        break;
+                    }
+                    name.push(c);
+                }
+                if closed {
+                    pattern.push_str(".*");
+                } else {
+                    // The regex crate treats an unclosed brace literally.
+                    pattern.push_str(&regex::escape(&format!("${{{}", name)));
+                }
+            }
+            // `$name`, where the name is the longest run of
+            // `[0-9A-Za-z_]` characters (same as the regex crate).
+            Some(c) if c.is_ascii_alphanumeric() || *c == '_' => {
+                while let Some(c) = chars.peek() {
+                    if c.is_ascii_alphanumeric() || *c == '_' {
+                        chars.next();
+                    } else {
+                        break;
+                    }
+                }
+                pattern.push_str(".*");
+            }
+            // A `$` followed by anything else is literal.
+            _ => pattern.push_str("\\$"),
+        }
+    }
+    pattern.push('$');
+    Regex::new(&pattern).expect("Output shape patterns should always be valid.")
 }
 
 impl PartialEq for Rule {
@@ -40,32 +121,97 @@ impl PartialEq for Rule {
 
 #[cfg(test)]
 mod tests {
+    use crate::rule::Rule;
+    use regex::Regex;
+
+    fn minify_rule() -> Rule {
+        Rule {
+            from: Regex::new("(?P<name>.*)\\.js$").unwrap(),
+            to: String::from("$name.min.js"),
+            command: "terser $i -o $o".to_string(),
+            exclude: None,
+            output_shape: None,
+        }
+    }
+
+    fn brotli_rule() -> Rule {
+        Rule {
+            from: Regex::new("(?P<name>.*)\\.min\\.js$").unwrap(),
+            to: String::from("$name.min.js.br"),
+            command: "brotli -f $i".to_string(),
+            exclude: None,
+            output_shape: None,
+        }
+    }
+
     #[test]
     fn generate_rules() {
         #![allow(clippy::trivial_regex)]
         use std::{fs::File, io::Write};
 
-        use crate::rule::Rule;
         use maplit::hashmap;
-        use regex::Regex;
 
         let rules = hashmap! {
             String::from("minify")=> Rule {
-                from: Regex::new("(?P<name>.*)\\.js$").unwrap(),
-                to: String::from("$name.min.js"),
-                command: "terser $i -o $o".to_string(),
                 exclude: Some(Regex::new("\\.min\\.js$").unwrap()),
+                ..minify_rule()
             },
-            String::from("brotli") => Rule {
-                from: Regex::new("(?P<name>.*)\\.min\\.js$").unwrap(),
-                to: String::from("$name.min.js.br"),
-                command: "brotli -f $i".to_string(),
-                exclude: None,
-            }
+            String::from("brotli") => brotli_rule(),
         };
         File::create("rules.toml")
             .unwrap()
             .write_all(toml::to_string_pretty(&rules).unwrap().as_bytes())
             .unwrap();
     }
+
+    #[test]
+    fn matches_own_output_single_level() {
+        let rule = minify_rule();
+        assert!(rule.matches_own_output("a.min.js"));
+        assert!(rule.matches_own_output("dir/a.min.js"));
+        assert!(!rule.matches_own_output("a.js"));
+        assert!(!rule.matches_own_output("a.min.js.br"));
+    }
+
+    #[test]
+    fn matches_own_output_any_depth() {
+        // *.js -> *.min.js -> *.pkg.min.js: files produced further down a
+        // rule chain still look like outputs of the first rule.
+        let rule = minify_rule();
+        assert!(rule.matches_own_output("a.pkg.min.js"));
+        assert!(rule.matches_own_output("a.min.min.js"));
+    }
+
+    #[test]
+    fn matches_own_output_ignores_other_rules_outputs() {
+        let rule = brotli_rule();
+        assert!(!rule.matches_own_output("a.min.js"));
+        assert!(rule.matches_own_output("a.min.js.br"));
+    }
+
+    #[test]
+    fn matches_own_output_uses_precompiled_shape() {
+        let mut rule = minify_rule();
+        rule.compile_output_shape();
+        assert!(rule.matches_own_output("a.min.js"));
+        assert!(!rule.matches_own_output("a.js"));
+    }
+
+    #[test]
+    fn output_shape_braced_and_numbered_references() {
+        let shape = super::output_shape_pattern("${name}.min.$1");
+        assert!(shape.is_match("a.min.js"));
+        assert!(!shape.is_match("a.js"));
+    }
+
+    #[test]
+    fn output_shape_escapes_literals() {
+        let shape = super::output_shape_pattern("$$cache.$name.js");
+        assert!(shape.is_match("$cache.foo.js"));
+        assert!(!shape.is_match("Xcache.foo.js"));
+
+        // `.` in the template is literal, not a regex wildcard.
+        let shape = super::output_shape_pattern("$name.min.js");
+        assert!(!shape.is_match("a.minxjs"));
+    }
 }