parsanol 1.3.75 → 1.3.77

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ca6a45b629235f8dc4976c046341a91d159a699f2784c564b08f73f31ed7cb49
-  data.tar.gz: f0fcbe3350657d533a298e67384eb54861fd5289a4ec049d98425d6db689012b
+  metadata.gz: d50458b2aaf5e7227e08f447b64dcc2197fd3df69e02ad7d9c332f87fdad15ec
+  data.tar.gz: 07e542379ad8ffaf5a9c0c317c88f8f2b1a3f9c9bff9d1b54983b8e6bbb70d14
 SHA512:
-  metadata.gz: 720729fb5363a237a40bf29ca97c21d977c45a0b3f30c58025abc32b940edfbc2ee8bcbe791697180d9272ea21370df0b59469a984b0cc815177f0989ca2e044
-  data.tar.gz: 66a5dd4694fd3ee9d8791e2fa94b68c6103762b2cd2f43322b508eb6dbee2d7bf1cb4c345403de1eecbfa9ce50ed5f5161441f9f6f2f958a550bdf6e7c893c05
+  metadata.gz: 4be52cd42ea4ad9773e0d7a0e8bcf718cd485c880594291fcf4c798255d74c1619ed238a82ca81d83f2b1a74392ffb384583e190a2a6a308ae93c1b38cd425ba
+  data.tar.gz: 1160ef2ee557c591c7c28894bf24a59570e1ba9f565744cb5cb839d7f33af59a1deeae1ef1906c1211d1cc1fe9c388c0f16e16d4e0aa003442fa15bff74c2c97

data/Cargo.lock

@@ -307,9 +307,9 @@ checksum = "9f7c3e4beb33f85d45ae3e3a1792185706c8e16d043238c593331cc7cd313b50"
 
 [[package]]
 name = "parsanol"
-version = "0.15.0"
+version = "0.17.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "01c2f6b018d28cdf676aa117f37747b227476b3f5fad5b577b51abb01609e6b5"
+checksum = "0a96694c55219bc77b62cc838883850952ad4f883cca10c11312a76a380da58d"
 dependencies = [
  "ahash",
  "getrandom 0.3.4",
@@ -326,9 +326,9 @@ dependencies = [
 
 [[package]]
 name = "parsanol-derive"
-version = "0.15.0"
+version = "0.17.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0aff970b9bcf36471356270512814045b011f30bb93f46382529ee94b3d8e11a"
+checksum = "07917f2daf91d216927237719ae459f068dd29bef53669a2d12dc4b28ad6a1d9"
 dependencies = [
  "proc-macro2",
  "quote",

data/ext/parsanol_cdylib/Cargo.toml

@@ -12,5 +12,5 @@ crate-type = ["cdylib"]
 doc = false
 
 [dependencies]
-# 0.15.0 carries the Constant and Lookbehind atoms
-parsanol = { version = "0.15.0" }
+# 0.17.0 carries LookSpec + the widened opaque writeback
+parsanol = { version = "0.17.0" }

data/ext/parsanol_native/Cargo.toml

@@ -29,9 +29,10 @@ magnus = { version = "0.9" }
 
 # parsanol parser library (crates.io; 0.12 carries the rs#174 native
 # backtracking fix: dynamic-dependent memoization + outcome cache)
-# parsanol parser library (crates.io; 0.15.0 carries the Constant and
-# Lookbehind atoms: the coradoc-markdown native gap, rs#137 follow-up)
-parsanol = { version = "0.15.0", features = ["ruby"] }
+# parsanol parser library (crates.io; 0.17.0 carries the LookSpec regex
+# lookbehind (rs#163 gap 1) and the widened opaque writeback: the
+# lazy-continuation fix, parsanol-ruby#160)
+parsanol = { version = "0.17.0", features = ["ruby"] }
 
 # Logging
 log = "0.4"

data/lib/parsanol/atoms/lookbehind.rb

@@ -2,30 +2,60 @@
 
 module Parsanol
   module Atoms
-    # Inspects the byte window behind the current position: the
-    # previous +count+ bytes must equal +pattern+ (positive) or
-    # differ (negative). Consumes nothing, yields nil — the
-    # precedes?/does_not_precede? guard shape (coradoc-markdown
-    # parity, rs#137 follow-up).
+    # Inspects the text behind the current position and consumes
+    # nothing, yielding nil.
+    #
+    # Two forms (rs#137/#163):
+    # - Literal: the previous +count+ bytes must equal +pattern+
+    #   (the precedes?/does_not_precede? fixed-window guard).
+    # - Regex: +source+ is searched in the preceding text and must
+    #   end at the position — the CommonMark flanking form, which is
+    #   class-based, multibyte and variable-length.
     class Lookbehind < Base
-      attr_reader :count, :pattern, :positive
+      attr_reader :count, :pattern, :regex_source, :positive
 
       def initialize(count, pattern, positive: true)
         super()
         @count = count
         @pattern = pattern
+        @regex_source = nil
+        @compiled = nil
         @positive = positive
       end
 
+      # Class-based flanking guard: +source+ must match the text
+      # ending at the position.
+      def self.regex(source, positive: true)
+        atom = new(0, "", positive: positive)
+        atom.instance_variable_set(:@regex_source, source)
+        atom
+      end
+
+      def regex?
+        !@regex_source.nil?
+      end
+
       def try(source, _context, _consume_all)
         start = source.bytepos
-        behind = start >= @count ? source.input.byteslice(start - @count, @count) : nil
-        matched = !behind.nil? && behind == @pattern.b
+        if regex?
+          @compiled ||= Regexp.new("(?:#{@regex_source})\\z")
+          matched = @compiled.match?(source.input.byteslice(0, start))
+        else
+          behind =
+            if start >= @count
+              source.input.byteslice(start - @count, @count)
+            end
+          matched = !behind.nil? && behind == @pattern.b
+        end
         [matched == @positive, nil]
       end
 
       def to_s_inner(_prec)
-        "lookbehind(#{@count}, #{@pattern.inspect}, #{@positive ? '+' : '-'})"
+        if regex?
+          "lookbehind(/#{@regex_source}/, #{@positive ? '+' : '-'})"
+        else
+          "lookbehind(#{@count}, #{@pattern.inspect}, #{@positive ? '+' : '-'})"
+        end
       end
     end
   end

data/lib/parsanol/native/serializer.rb

@@ -103,15 +103,17 @@ module Parsanol
                      # atoms — empty match yielding a wire constant.
                      { "Constant" => { "value" => serialize_constant_value(atom.value) } }
                    when Parsanol::Atoms::Lookbehind
-                     # rs#137 follow-up: the precedes?/does_not_precede?
-                     # guards — inspect the bytes behind the position.
-                     {
-                       "Lookbehind" => {
-                         "count" => atom.count,
-                         "pattern" => atom.pattern,
-                         "positive" => atom.positive,
-                       },
-                     }
+                     # rs#137/#163: the behind-window test — a fixed
+                     # literal window or an end-anchored regex (the
+                     # CommonMark flanking form).
+                     look = if atom.regex?
+                              { "Regex" => { "source" => atom.regex_source } }
+                            else
+                              { "Literal" => { "count" => atom.count,
+                                               "pattern" => atom.pattern } }
+                            end
+                     { "Lookbehind" => { "look" => look,
+                                         "positive" => atom.positive } }
                    when Parsanol::Atoms::Ignored
                      serialize_ignored(atom)
                    else

data/lib/parsanol/parg/artifact.rb

@@ -65,8 +65,16 @@ module Parsanol
       def parse(entry_name, input, mode: :native)
         case mode
         when :native
-          if !envelope["dynamic"] && Native.available?
-            return Native.parse(JSON.generate(entry(entry_name).fetch("grammar")), input)
+          if Native.available? && !envelope["dynamic"]
+            # Fast lane (parsanol-ruby#162): the per-instance atom
+            # tree rides the DSL serialization/registration caches
+            # (GRAMMAR_CACHE / HANDLE_CACHE keyed on the stable root
+            # atom), so an artifact parses at compiled speed after
+            # the first call. The old path re-serialized the whole
+            # grammar JSON and re-registered it per parse (~8x).
+            # Dynamic artifacts carry ruby-tier atoms their wire
+            # cannot express — they keep the interpreter.
+            return root_atom(entry_name).parse(input, mode: :native)
           end
 
           root_atom(entry_name).parse(input)

data/lib/parsanol/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Parsanol
-  VERSION = "1.3.75"
+  VERSION = "1.3.77"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: parsanol
 version: !ruby/object:Gem::Version
-  version: 1.3.75
+  version: 1.3.77
 platform: ruby
 authors:
 - Ribose Inc.