parsanol 1.3.68 → 1.3.70

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 5917e9971751a101c27984a0ccec6aa6254813b77d18a45eea7f6a5538ead517
-  data.tar.gz: 9513f097d858d7c394a10c8002e372523212bf11629813867ec59a91dacbc41d
+  metadata.gz: 46a9a34dd51e63ba24ee155131eb13326947d08cf08a9814a6d600046aead88a
+  data.tar.gz: 679d459fa6c46a6b27816c042af13b0c6e9f15ccae847786140c473731b25ebd
 SHA512:
-  metadata.gz: 65cf06e07d16879100effb50a7ed8012e95774ed6b8c86f97c997d3b6ce7a41525a07e3943d4670dc3c6f65e91d587953e719668decba35f90610bdf29a7c011
-  data.tar.gz: 84cd53317231f028e6510d90c7c426379566e47570594e960b022766ae45fffb8305afdcc5f044339fbbdb3a30c74e44e25ce03a735c7e3672221f70f26b0ab1
+  metadata.gz: 5bddc1e0b863c8f50be6573f29e09b17091c95dbfd0db57c660aea348d090b681bb7d99e3870276f3b05f32fcc1944098f71777bdaf4db2b34f348c1e847fbaa
+  data.tar.gz: bbb0ce39775917e628fcc19a04d15e2f252437d6fdcfe9b9b69f33711e096b2cb9a2b9a40908177068f66e54c3dc9ceb4d4ba25396b4cfc0ac465de34667b65d

data/Cargo.lock

@@ -310,13 +310,31 @@ name = "parsanol"
 version = "0.10.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "dd43c104276def3f79a54ce6e8d26642ea330c13f97d66380bd94fc057f0a849"
+dependencies = [
+ "ahash",
+ "getrandom 0.3.4",
+ "getrandom 0.4.3",
+ "hashbrown",
+ "parsanol-derive 0.10.2",
+ "regex",
+ "regex-syntax",
+ "serde",
+ "serde_json",
+ "sha2",
+]
+
+[[package]]
+name = "parsanol"
+version = "0.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b3754baf8f7f9761748bac497cd60efd1bab93ff54c4212e899e27cf3f09100a"
 dependencies = [
  "ahash",
  "getrandom 0.3.4",
  "getrandom 0.4.3",
  "hashbrown",
  "magnus",
- "parsanol-derive",
+ "parsanol-derive 0.12.0",
  "regex",
  "regex-syntax",
  "serde",
@@ -335,11 +353,22 @@ dependencies = [
  "syn 3.0.6",
 ]
 
+[[package]]
+name = "parsanol-derive"
+version = "0.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7cdc8438988a014ec5a7cb6d7891c3c467154de134b353e8790bd42d4ae2ddd5"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 3.0.6",
+]
+
 [[package]]
 name = "parsanol_cdylib"
 version = "1.3.52"
 dependencies = [
- "parsanol",
+ "parsanol 0.10.2",
 ]
 
 [[package]]
@@ -348,7 +377,7 @@ version = "1.0.2"
 dependencies = [
  "log",
  "magnus",
- "parsanol",
+ "parsanol 0.12.0",
  "rb-sys",
 ]
 

data/ext/parsanol_native/Cargo.toml

@@ -27,8 +27,9 @@ rb-sys = { version = "0.9.124", features = ["global-allocator"] }
 # magnus for Ruby bindings (from crates.io, patched locally for Ruby 4.0 support)
 magnus = { version = "0.9" }
 
-# parsanol parser library (crates.io; 0.10 carries the class fast-path fix)
-parsanol = { version = "0.10.2", features = ["ruby"] }
+# parsanol parser library (crates.io; 0.12 carries the rs#174 native
+# backtracking fix: dynamic-dependent memoization + outcome cache)
+parsanol = { version = "0.12.0", features = ["ruby"] }
 
 # Logging
 log = "0.4"

data/lib/parsanol/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Parsanol
-  VERSION = "1.3.68"
+  VERSION = "1.3.70"
 end

data/lib/parsanol/vm.rb

@@ -84,6 +84,14 @@ module Parsanol
     STEP_BUDGET_FIXED = 10_000
     MEMO_BUDGET_FACTOR = 4_096
     MEMO_BUDGET_CELLS_FACTOR = 16
+    # Wall-clock ceiling for the memoized retry pass. Step budgets assume
+    # sub-microsecond steps; pathological grammars make each step
+    # expensive enough that a step budget alone mispredicts wall time by
+    # orders of magnitude (parsanol-ruby#147: a 6-line skip-injected
+    # document ground ~146s inside its memo budget where the interpreter
+    # answers in milliseconds). Bounded wall time surrenders the grammar
+    # to the interpreter, the source of truth, instead of burning CPU.
+    MEMO_WALL_BUDGET_SECONDS = 5.0
 
     # Compiled-program and heavy-flag caches hold at most this many
     # grammars. A process rarely uses more than a handful concurrently;
@@ -130,7 +138,10 @@ module Parsanol
           end
         cache[root] = program
         trim_cache(cache)
-        program.equal?(:fallback) ? nil : program
+        # :fallback = VM-incompatible; :oversize even non-inlined means
+        # the grammar is beyond the program cap — the interpreter
+        # handles both.
+        program.equal?(:fallback) || program.equal?(:oversize) ? nil : program
       end
 
       # Marks a grammar as VM-incompatible after a runtime failure.
@@ -197,6 +208,10 @@ module Parsanol
         end
 
         compiler.to_program
+      rescue Compiler::ProgramTooLarge
+        # Emission aborted at the cap: discard and let the caller
+        # decide (compile retries non-inlined).
+        :oversize
       end
 
       # Returns violation descriptions for this atom's ordered-choice
@@ -679,7 +694,19 @@ module Parsanol
 
       private
 
+      # Raised mid-emission once the program passes MAX_PROGRAM: abort
+      # the build immediately instead of materializing millions of
+      # instructions only to discover the overflow in to_program
+      # (parsanol-ruby#147: an inline pass on a skip-injected grammar
+      # reached 6.3M instructions / 16M with subroutines before the
+      # post-hoc check discarded them — minutes of compile time and
+      # gigabytes of transient allocations for a program that is
+      # thrown away and rebuilt non-inlined anyway).
+      class ProgramTooLarge < StandardError; end
+
       def emit(op, a = nil, b = nil, c = nil) # rubocop:disable Naming/MethodParameterName -- opcode operand slots
+        raise ProgramTooLarge if @ops.size > MAX_PROGRAM
+
         @ops << [op, a, b, c]
         @ops.size - 1
       end
@@ -902,6 +929,9 @@ module Parsanol
                    (STEP_BUDGET_FACTOR * n) + STEP_BUDGET_FIXED
                  end
         steps = 0
+        wall_deadline = if @memoize
+                          Process.clock_gettime(Process::CLOCK_MONOTONIC) + MEMO_WALL_BUDGET_SECONDS
+                        end
 
         pc = 0
         pos = 0
@@ -917,6 +947,13 @@ module Parsanol
         # rubocop:disable Metrics/BlockLength -- the dispatch loop IS execute
         loop do
           steps += 1
+          # Checked every 1024 steps: a monotonic clock read per
+          # millisecond of ordinary work costs nothing measurable.
+          if wall_deadline && steps.nobits?(0x3FF) &&
+              Process.clock_gettime(Process::CLOCK_MONOTONIC) > wall_deadline
+            return BUDGET
+          end
+
           if steps > budget
             if trace
               warn "BUDGET at steps=#{steps} pc=#{pc} pos=#{pos} rstack=#{rstack.size} bt=#{bt.size} calls=#{calls.size}"

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: parsanol
 version: !ruby/object:Gem::Version
-  version: 1.3.68
+  version: 1.3.70
 platform: ruby
 authors:
 - Ribose Inc.