shrinkray 25.12.27.0__py3-none-any.whl → 25.12.27.2__py3-none-any.whl

shrinkray/__main__.py

@@ -75,12 +75,14 @@ async def run_shrink_ray(
 )
 @click.option(
     "--timeout",
-    default=1,
+    default=None,
     type=click.FLOAT,
     help=(
-        "Time out subprocesses after this many seconds. If set to <= 0 then "
-        "no timeout will be used. Any commands that time out will be treated "
-        "as failing the test"
+        "Time out subprocesses after this many seconds. If not specified, "
+        "runs the interestingness test once and sets timeout to 10x the "
+        "measured time (capped at 5 minutes). If set to <= 0 then no timeout "
+        "will be used. Any commands that time out will be treated as failing "
+        "the test"
     ),
 )
 @click.option(
@@ -212,7 +214,7 @@ def main(
     backup: str,
     filename: str,
     test: list[str],
-    timeout: float,
+    timeout: float | None,
     in_place: bool,
     parallelism: int,
     seed: int,
@@ -225,7 +227,7 @@ def main(
     ui_type: UIType,
     theme: str,
 ) -> None:
-    if timeout <= 0:
+    if timeout is not None and timeout <= 0:
         timeout = float("inf")
 
     if not os.access(test[0], os.X_OK):

shrinkray/passes/bytes.py

@@ -24,8 +24,8 @@ from collections.abc import Sequence
 
 from attrs import define
 
-from shrinkray.passes.definitions import Format, ReductionProblem
 from shrinkray.passes.patching import Cuts, Patches, apply_patches
+from shrinkray.problem import Format, ReductionProblem
 
 
 @define(frozen=True)
@@ -739,12 +739,13 @@ async def line_sorter(problem: ReductionProblem[bytes]):
     while i < len(lines):
         j = i
         while j > 0:
-            u = lines[j - 1]
-            v = lines[j]
-            if v + u < u + v:
-                attempt = list(lines)
-                attempt[j - 1], attempt[j] = attempt[j], attempt[j - 1]
-                if not await problem.is_interesting(b"\n".join(attempt)):
+            attempt = list(lines)
+            attempt[j - 1], attempt[j] = attempt[j], attempt[j - 1]
+            new_test_case = b"\n".join(attempt)
+            if problem.sort_key(new_test_case) < problem.sort_key(
+                problem.current_test_case
+            ):
+                if not await problem.is_interesting(new_test_case):
                     break
                 else:
                     j -= 1

shrinkray/passes/definitions.py

@@ -4,20 +4,20 @@ This module defines the core type aliases and abstractions for reduction:
 
 - ReductionPass[T]: A function that attempts to reduce a test case
 - ReductionPump[T]: A function that may temporarily increase test case size
-- Format[S, T]: A bidirectional transformation between types
 - compose(): Combines a Format with a pass to work on a different type
 
 These abstractions enable format-agnostic reduction: the same pass
 (e.g., "delete duplicate elements") can work on bytes, lines, tokens,
 JSON arrays, or any other sequence-like type.
+
+Note: Format, ParseError, and DumpError are defined in shrinkray.problem.
 """
 
-from abc import ABC, abstractmethod
 from collections.abc import Awaitable, Callable
 from functools import wraps
 from typing import TypeVar
 
-from shrinkray.problem import ReductionProblem
+from shrinkray.problem import Format, ParseError, ReductionProblem
 
 
 S = TypeVar("S")
@@ -36,70 +36,6 @@ ReductionPass = Callable[[ReductionProblem[T]], Awaitable[None]]
 ReductionPump = Callable[[ReductionProblem[T]], Awaitable[T]]
 
 
-class ParseError(Exception):
-    """Raised when a Format cannot parse its input."""
-
-    pass
-
-
-class DumpError(Exception):
-    """Raised when a Format cannot serialize its output.
-
-    This occurs because not all internal representations map to valid
-    output in the target format. For example, a reduction might create
-    an invalid AST structure that cannot be converted back to source code.
-    """
-
-    pass
-
-
-class Format[S, T](ABC):
-    """A bidirectional transformation between two types.
-
-    Formats enable format-agnostic passes by abstracting the
-    parse/serialize cycle. For example:
-
-    - Split(b"\\n"): bytes <-> list[bytes] (lines)
-    - Tokenize(): bytes <-> list[bytes] (tokens)
-    - JSON: bytes <-> Any (Python objects)
-    - DimacsCNF: bytes <-> list[list[int]] (SAT clauses)
-
-    A Format must satisfy the round-trip property:
-        dumps(parse(x)) should be equivalent to x
-        (possibly with normalization)
-
-    Example usage:
-        # Delete duplicate lines
-        compose(Split(b"\\n"), delete_duplicates)
-
-        # Reduce integer literals in source code
-        compose(IntegerFormat(), reduce_integer)
-    """
-
-    @property
-    def name(self) -> str:
-        """Human-readable name for this format, used in pass names."""
-        return repr(self)
-
-    @abstractmethod
-    def parse(self, input: S) -> T:
-        """Parse input into the target type. Raises ParseError on failure."""
-        ...
-
-    def is_valid(self, input: S) -> bool:
-        """Check if input can be parsed by this format."""
-        try:
-            self.parse(input)
-            return True
-        except ParseError:
-            return False
-
-    @abstractmethod
-    def dumps(self, input: T) -> S:
-        """Serialize the target type back to the source type."""
-        ...
-
-
 def compose(format: Format[S, T], reduction_pass: ReductionPass[T]) -> ReductionPass[S]:
     """Wrap a reduction pass to work through a Format transformation.
 

shrinkray/passes/genericlanguages.py

@@ -3,7 +3,7 @@ Module of reduction passes designed for "things that look like programming langu
 """
 
 import re
-from collections.abc import Callable, Sized
+from collections.abc import Callable
 from functools import wraps
 from string import ascii_lowercase, ascii_uppercase
 from typing import AnyStr
@@ -12,9 +12,15 @@ import trio
 from attr import define
 
 from shrinkray.passes.bytes import ByteReplacement, delete_intervals
-from shrinkray.passes.definitions import Format, ParseError, ReductionPass
+from shrinkray.passes.definitions import ReductionPass
 from shrinkray.passes.patching import PatchApplier, Patches, apply_patches
-from shrinkray.problem import BasicReductionProblem, ReductionProblem
+from shrinkray.problem import (
+    BasicReductionProblem,
+    Format,
+    ParseError,
+    ReductionProblem,
+    sort_key_for_initial,
+)
 from shrinkray.work import NotFound
 
 
@@ -240,10 +246,6 @@ async def simplify_brackets(problem: ReductionProblem[bytes]) -> None:
 IDENTIFIER = re.compile(rb"(\b[A-Za-z][A-Za-z0-9_]*\b)|([0-9]+)")
 
 
-def shortlex[T: Sized](s: T) -> tuple[int, T]:
-    return (len(s), s)
-
-
 async def normalize_identifiers(problem: ReductionProblem[bytes]) -> None:
     """Replace identifiers with shorter alternatives.
 
@@ -261,8 +263,10 @@ async def normalize_identifiers(problem: ReductionProblem[bytes]) -> None:
                 replacements.add(c)
                 break
 
-    replacements = sorted(replacements, key=shortlex)
-    targets = sorted(identifiers, key=shortlex, reverse=True)
+    sort_key = sort_key_for_initial(problem.current_test_case)
+
+    replacements = sorted(replacements, key=sort_key)
+    targets = sorted(identifiers, key=sort_key, reverse=True)
 
     # TODO: This could use better parallelisation.
     for t in targets:
@@ -272,7 +276,7 @@ async def normalize_identifiers(problem: ReductionProblem[bytes]) -> None:
             continue
 
         async def can_replace(r):
-            if shortlex(r) >= shortlex(t):
+            if sort_key(r) >= sort_key(t):
                 return False
             attempt = pattern.sub(r, source)
             assert attempt != source

shrinkray/passes/json.py

@@ -4,9 +4,9 @@ from typing import Any
 
 from attrs import define
 
-from shrinkray.passes.definitions import Format, ParseError, ReductionPass
+from shrinkray.passes.definitions import ReductionPass
 from shrinkray.passes.patching import Patches, apply_patches
-from shrinkray.problem import ReductionProblem
+from shrinkray.problem import Format, ParseError, ReductionProblem
 
 
 def is_json(s: bytes) -> bool:

shrinkray/passes/sat.py

@@ -1,15 +1,10 @@
 from collections import Counter, defaultdict
 from collections.abc import Callable, Iterable, Iterator
 
-from shrinkray.passes.definitions import (
-    DumpError,
-    Format,
-    ParseError,
-    ReductionPass,
-)
+from shrinkray.passes.definitions import ReductionPass
 from shrinkray.passes.patching import Conflict, SetPatches, apply_patches
 from shrinkray.passes.sequences import delete_elements
-from shrinkray.problem import ReductionProblem
+from shrinkray.problem import DumpError, Format, ParseError, ReductionProblem
 
 
 Clause = list[int]

shrinkray/problem.py

@@ -12,12 +12,13 @@ the details of caching, parallelism, and state management.
 """
 
 import hashlib
+import string
 import time
 from abc import ABC, abstractmethod
 from collections.abc import Awaitable, Callable, Sized
 from datetime import timedelta
+from functools import total_ordering
 from typing import (
-    TYPE_CHECKING,
     Any,
     Protocol,
     TypeVar,
@@ -29,12 +30,10 @@ import trio
 from attrs import define
 from humanize import naturalsize, precisedelta
 
+from shrinkray.formatting import try_decode
 from shrinkray.work import WorkContext
 
 
-if TYPE_CHECKING:
-    from shrinkray.passes.definitions import Format
-
 S = TypeVar("S")
 T = TypeVar("T")
 
@@ -71,9 +70,196 @@ def shortlex[SizedT: Sized](value: SizedT) -> tuple[int, SizedT]:
     return (len(value), value)
 
 
-def default_sort_key(value: Any):
-    if isinstance(value, str | bytes):
+@total_ordering
+class LazyChainedSortKey:
+    """A comparison key that lazily evaluates a chain of comparison functions.
+
+    This class provides an ordering that compares values by applying a sequence
+    of functions in order. The first function that produces different values
+    for two inputs determines the ordering. If all functions return equal
+    values, the inputs are considered equal.
+
+    This is used to implement the natural ordering for strings, which compares
+    by length, then average squared line length, then number of lines, etc.
+
+    The "lazy" aspect is that comparison functions are only evaluated until
+    one returns different values, avoiding unnecessary computation.
+    """
+
+    def __init__(self, functions: list[Callable[[T], Any]], value: T):
+        self.functions = functions
+        self.value = value
+
+    def __eq__(self, other):
+        if not isinstance(other, LazyChainedSortKey):
+            return NotImplemented
+        assert len(self.functions) == len(other.functions)
+        return self.value == other.value
+
+    def __lt__(self, other):
+        if self == other:
+            return False
+        if not isinstance(other, LazyChainedSortKey):
+            return NotImplemented
+        for f in self.functions:
+            self_key = f(self.value)
+            other_key = f(other.value)
+            if self_key < other_key:
+                return True
+            elif self_key > other_key:
+                return False
+        # All comparison functions returned equal values for different inputs.
+        # This shouldn't happen with the current functions (natural_string_lex
+        # compares character-by-character) but if it does, neither is less.
+        return False
+
+
+# Natural character ordering: whitespace < digits < lowercase < uppercase.
+# Characters not in this string are sorted by ord() after all known characters.
+NATURAL_CHARACTER_ORDER = (
+    string.whitespace + string.digits + string.ascii_lowercase + string.ascii_uppercase
+)
+NATURAL_CHARACTER_ORDER_INDEX = {s: i for i, s in enumerate(NATURAL_CHARACTER_ORDER)}
+
+
+def character_index(c: str) -> int:
+    """Return the sorting index for a character in natural ordering.
+
+    Characters in NATURAL_CHARACTER_ORDER get their position in that string.
+    Unknown characters (punctuation, unicode, etc.) sort after all known
+    characters, ordered by their Unicode code point.
+    """
+    return NATURAL_CHARACTER_ORDER_INDEX.get(c, len(NATURAL_CHARACTER_ORDER) + ord(c))
+
+
+def natural_string_lex(s: str) -> list[int]:
+    """Convert a string to a list of character indices for lexicographic comparison.
+
+    This transforms the string so that comparing the resulting lists gives
+    the natural character ordering (whitespace < digits < lowercase < uppercase).
+    """
+    return list(map(character_index, s))
+
+
+# The chain of comparison functions used for natural string ordering.
+# Each function is tried in sequence; the first that differs determines order.
+#
+# 1. Total length - shorter strings are always preferred
+# 2. Average squared line length - penalizes very long lines, preferring balanced code
+#    Formula: sum(len(line)²) / count(lines)²
+# 3. Number of lines - fewer lines is better (after accounting for balance)
+# 4. List of line lengths - lexicographically compare line length sequences
+# 5. Natural character order - whitespace < digits < lowercase < uppercase
+NATURAL_ORDERING_FUNCTIONS: list[Callable[[str], Any]] = [
+    len,
+    lambda s: sum(len(line) ** 2 for line in s.split("\n")) / len(s.split("\n")) ** 2,
+    lambda s: len(s.splitlines()),
+    lambda s: list(map(len, s.splitlines())),
+    natural_string_lex,
+]
+
+
+def natural_key(s: str) -> LazyChainedSortKey:
+    """Return a comparison key for natural string ordering.
+
+    Natural ordering uses a chain of heuristics to determine which string
+    is "smaller" (more reduced). This is designed to produce human-readable
+    minimal test cases with balanced line lengths and natural character choices.
+
+    See NATURAL_ORDERING_FUNCTIONS for the complete ordering criteria.
+    """
+    return LazyChainedSortKey(functions=NATURAL_ORDERING_FUNCTIONS, value=s)
+
+
+def sort_key_for_initial(initial: Any) -> Callable[[Any], Any]:
+    """Create a sort key function appropriate for the given initial value.
+
+    This examines the initial test case and returns a comparison function
+    that will be used to order all test cases during reduction.
+
+    For bytes:
+        - If decodable as text, uses natural ordering on the decoded string
+        - Falls back to shortlex for binary data that can't be decoded
+
+    For dicts:
+        - Orders by total size of values, then number of keys
+        - Then compares values for each key in order of largest-first
+
+    For other types:
+        - Falls back to natural ordering on repr()
+
+    The returned function can be used as a sort key or comparison key.
+    """
+    if isinstance(initial, bytes):
+        encoding, _ = try_decode(initial)
+        if encoding is None:
+            return shortlex
+        else:
+
+            def natural_for_encoding(b: bytes) -> Any:
+                try:
+                    s = b.decode(encoding)
+                    return (0, natural_key(s))
+                except UnicodeDecodeError:
+                    return (1, shortlex(b))
+
+            return natural_for_encoding
+    elif isinstance(initial, dict):
+        keys = sorted(initial, key=lambda k: shortlex(initial[k]), reverse=True)
+        natural_keys = {k: sort_key_for_initial(v) for k, v in initial.items()}
+
+        def dict_total_size(s):
+            return sum(len(v) for v in s.values())
+
+        def key_sort_key(k):
+            def f(x):
+                try:
+                    v = x[k]
+                except KeyError:
+                    return (0,)
+                else:
+                    return (1, natural_keys[k](v))
+
+            return f
+
+        functions = [
+            dict_total_size,
+            len,
+        ] + [key_sort_key(k) for k in keys]
+
+        def dict_sort_key(v):
+            return LazyChainedSortKey(
+                functions=functions,
+                value=v,
+            )
+
+        return dict_sort_key
+    else:
+        # We don't use this branch in the main app, but this
+        # function is also used in tests.
+        def fallback_sort_key(s):
+            return natural_key(repr(s))
+
+        return fallback_sort_key
+
+
+def default_sort_key(value: Any) -> Any:
+    """Return a comparison key for a value using type-appropriate ordering.
+
+    This is a simpler alternative to sort_key_for_initial that doesn't
+    examine the initial value to determine the best ordering.
+
+    - bytes: shortlex ordering (length, then lexicographic)
+    - str: natural ordering (length, line balance, character order)
+    - other: shortlex on repr()
+
+    Note: This really should return some sort of Comparable type, but Python
+    doesn't have a built-in protocol for that.
+    """
+    if isinstance(value, bytes):
         return shortlex(value)
+    elif isinstance(value, str):
+        return natural_key(value)
     else:
         return shortlex(repr(value))
 
@@ -85,6 +271,70 @@ def default_display(value: Any) -> str:
     return f"value of size {len(value)}"
 
 
+class ParseError(Exception):
+    """Raised when a Format cannot parse its input."""
+
+    pass
+
+
+class DumpError(Exception):
+    """Raised when a Format cannot serialize its output.
+
+    This occurs because not all internal representations map to valid
+    output in the target format. For example, a reduction might create
+    an invalid AST structure that cannot be converted back to source code.
+    """
+
+    pass
+
+
+class Format[S, T](ABC):
+    """A bidirectional transformation between two types.
+
+    Formats enable format-agnostic passes by abstracting the
+    parse/serialize cycle. For example:
+
+    - Split(b"\\n"): bytes <-> list[bytes] (lines)
+    - Tokenize(): bytes <-> list[bytes] (tokens)
+    - JSON: bytes <-> Any (Python objects)
+    - DimacsCNF: bytes <-> list[list[int]] (SAT clauses)
+
+    A Format must satisfy the round-trip property:
+        dumps(parse(x)) should be equivalent to x
+        (possibly with normalization)
+
+    Example usage:
+        # Delete duplicate lines
+        compose(Split(b"\\n"), delete_duplicates)
+
+        # Reduce integer literals in source code
+        compose(IntegerFormat(), reduce_integer)
+    """
+
+    @property
+    def name(self) -> str:
+        """Human-readable name for this format, used in pass names."""
+        return repr(self)
+
+    @abstractmethod
+    def parse(self, input: S) -> T:
+        """Parse input into the target type. Raises ParseError on failure."""
+        ...
+
+    def is_valid(self, input: S) -> bool:
+        """Check if input can be parsed by this format."""
+        try:
+            self.parse(input)
+            return True
+        except ParseError:
+            return False
+
+    @abstractmethod
+    def dumps(self, input: T) -> S:
+        """Serialize the target type back to the source type."""
+        ...
+
+
 def default_size(value: Any) -> int:
     try:
         return len(value)
@@ -182,9 +432,7 @@ class ReductionProblem[T](ABC):
         # Cache of View objects for each Format, to avoid re-parsing
         self.__view_cache: dict[Any, ReductionProblem[Any]] = {}
 
-    def view(
-        self, format: "Format[T, S] | type[Format[T, S]]"
-    ) -> "ReductionProblem[S]":
+    def view(self, format: Format[T, S] | type[Format[T, S]]) -> "ReductionProblem[S]":
         """Create a view of this problem through a Format.
 
         A View wraps this problem, parsing the current test case through
@@ -481,8 +729,6 @@ class View[S, T](ReductionProblem[T]):
         return self.__current
 
     async def is_interesting(self, test_case: T) -> bool:
-        from shrinkray.passes.definitions import DumpError
-
         try:
             return await self.__problem.is_interesting(self.__dump(test_case))
         except DumpError:

shrinkray/reducer.py

@@ -49,7 +49,12 @@ from shrinkray.passes.patching import PatchApplier, Patches
 from shrinkray.passes.python import PYTHON_PASSES, is_python
 from shrinkray.passes.sat import SAT_PASSES, DimacsCNF
 from shrinkray.passes.sequences import block_deletion, delete_duplicates
-from shrinkray.problem import ReductionProblem, ReductionStats, shortlex
+from shrinkray.problem import (
+    ReductionProblem,
+    ReductionStats,
+    shortlex,
+    sort_key_for_initial,
+)
 
 
 @define
@@ -531,6 +536,8 @@ class KeyProblem(ReductionProblem[bytes]):
         self.base_problem = base_problem
         self.applier = applier
         self.key = key
+        # Use the appropriate sort key for this value (natural for text, shortlex for binary)
+        self._sort_key_fn = sort_key_for_initial(self.current_test_case)
 
     @property
     def current_test_case(self) -> bytes:
@@ -547,7 +554,7 @@ class KeyProblem(ReductionProblem[bytes]):
         return len(test_case)
 
     def sort_key(self, test_case: bytes) -> Any:
-        return shortlex(test_case)
+        return self._sort_key_fn(test_case)
 
     def display(self, value: bytes) -> str:
         return repr(value)

shrinkray/state.py

@@ -10,7 +10,7 @@ import time
 from abc import ABC, abstractmethod
 from datetime import timedelta
 from tempfile import TemporaryDirectory
-from typing import Any
+from typing import Any, TypeVar
 
 import humanize
 import trio
@@ -21,12 +21,15 @@ from shrinkray.problem import (
     BasicReductionProblem,
     InvalidInitialExample,
     ReductionProblem,
-    shortlex,
+    sort_key_for_initial,
 )
 from shrinkray.reducer import DirectoryShrinkRay, Reducer, ShrinkRay
 from shrinkray.work import Volume, WorkContext
 
 
+T = TypeVar("T")
+
+
 class TimeoutExceededOnInitial(InvalidInitialExample):
     def __init__(self, runtime: float, timeout: float) -> None:
         self.runtime = runtime
@@ -36,13 +39,32 @@ class TimeoutExceededOnInitial(InvalidInitialExample):
         )
 
 
+# Constants for dynamic timeout
+DYNAMIC_TIMEOUT_CALIBRATION_TIMEOUT = 300.0  # 5 minutes for first call
+DYNAMIC_TIMEOUT_MULTIPLIER = 10
+DYNAMIC_TIMEOUT_MAX = 300.0  # 5 minutes maximum
+DYNAMIC_TIMEOUT_MIN = 1.0  # 1 second minimum to prevent edge cases
+
+
+def compute_dynamic_timeout(runtime: float) -> float:
+    """Compute dynamic timeout based on measured runtime.
+
+    The timeout is set to 10x the measured runtime, clamped between
+    DYNAMIC_TIMEOUT_MIN and DYNAMIC_TIMEOUT_MAX.
+    """
+    return max(
+        DYNAMIC_TIMEOUT_MIN,
+        min(runtime * DYNAMIC_TIMEOUT_MULTIPLIER, DYNAMIC_TIMEOUT_MAX),
+    )
+
+
 @define(slots=False)
 class ShrinkRayState[TestCase](ABC):
     input_type: Any  # InputType from __main__
     in_place: bool
     test: list[str]
     filename: str
-    timeout: float
+    timeout: float | None
     base: str
     parallelism: int
     initial: TestCase
@@ -127,7 +149,8 @@ class ShrinkRayState[TestCase](ABC):
             completed = await trio.run_process(command, **kwargs)
             runtime = time.time() - start_time
 
-            if runtime >= self.timeout and self.first_call:
+            # Check for timeout violation (only when timeout is explicitly set)
+            if self.timeout is not None and runtime >= self.timeout and self.first_call:
                 self.initial_exit_code = completed.returncode
                 self.first_call = False
                 raise TimeoutExceededOnInitial(
@@ -137,6 +160,9 @@ class ShrinkRayState[TestCase](ABC):
 
             if self.first_call:
                 self.initial_exit_code = completed.returncode
+                # Set dynamic timeout if not explicitly specified
+                if self.timeout is None:
+                    self.timeout = compute_dynamic_timeout(runtime)
             self.first_call = False
 
             # Store captured output
@@ -168,9 +194,19 @@ class ShrinkRayState[TestCase](ABC):
             sp = await nursery.start(call_with_kwargs)
 
             try:
-                with trio.move_on_after(
-                    self.timeout * 10 if self.first_call else self.timeout
-                ):
+                # Determine effective timeout for this call
+                if self.first_call:
+                    # For first call: use calibration timeout if dynamic, otherwise 10x explicit timeout
+                    if self.timeout is None:
+                        effective_timeout = DYNAMIC_TIMEOUT_CALIBRATION_TIMEOUT
+                    else:
+                        effective_timeout = self.timeout * 10
+                else:
+                    # For subsequent calls, timeout must be set (either explicit or computed)
+                    assert self.timeout is not None
+                    effective_timeout = self.timeout
+
+                with trio.move_on_after(effective_timeout):
                     await sp.wait()
 
                 runtime = time.time() - start_time
@@ -179,7 +215,12 @@ class ShrinkRayState[TestCase](ABC):
                     # Process didn't terminate before timeout - kill it
                     await self._interrupt_wait_and_kill(sp)
 
-                if runtime >= self.timeout and self.first_call:
+                # Check for timeout violation (only when timeout is explicitly set)
+                if (
+                    self.timeout is not None
+                    and runtime >= self.timeout
+                    and self.first_call
+                ):
                     raise TimeoutExceededOnInitial(
                         timeout=self.timeout,
                         runtime=runtime,
@@ -187,6 +228,10 @@ class ShrinkRayState[TestCase](ABC):
             finally:
                 if self.first_call:
                     self.initial_exit_code = sp.returncode
+                    # Set dynamic timeout if not explicitly specified
+                    if self.timeout is None:
+                        runtime = time.time() - start_time
+                        self.timeout = compute_dynamic_timeout(runtime)
                 self.first_call = False
 
             result: int | None = sp.returncode
@@ -268,6 +313,7 @@ class ShrinkRayState[TestCase](ABC):
             is_interesting=self.is_interesting,
             initial=self.initial,
             work=work,
+            sort_key=sort_key_for_initial(self.initial),
             **self.extra_problem_kwargs,
         )
 
@@ -551,20 +597,9 @@ class ShrinkRayDirectoryState(ShrinkRayState[dict[str, bytes]]):
     def setup_formatter(self): ...
 
     @property
-    def extra_problem_kwargs(self):
-        def dict_size(test_case: dict[str, bytes]) -> int:
-            return sum(len(v) for v in test_case.values())
-
-        def dict_sort_key(test_case: dict[str, bytes]) -> Any:
-            return (
-                len(test_case),
-                dict_size(test_case),
-                sorted((k, shortlex(v)) for k, v in test_case.items()),
-            )
-
+    def extra_problem_kwargs(self) -> dict[str, Any]:
         return {
-            "sort_key": dict_sort_key,
-            "size": dict_size,
+            "size": lambda tc: sum(len(v) for v in tc.values()),
         }
 
     def new_reducer(

shrinkray/subprocess/client.py

@@ -127,7 +127,7 @@ class SubprocessClient:
         file_path: str,
         test: list[str],
         parallelism: int | None = None,
-        timeout: float = 1.0,
+        timeout: float | None = None,
         seed: int = 0,
         input_type: str = "all",
         in_place: bool = False,
@@ -138,10 +138,9 @@ class SubprocessClient:
         trivial_is_error: bool = True,
     ) -> Response:
         """Start the reduction process."""
-        params = {
+        params: dict[str, Any] = {
             "file_path": file_path,
             "test": test,
-            "timeout": timeout,
             "seed": seed,
             "input_type": input_type,
             "in_place": in_place,
@@ -153,6 +152,8 @@ class SubprocessClient:
         }
         if parallelism is not None:
             params["parallelism"] = parallelism
+        if timeout is not None:
+            params["timeout"] = timeout
         return await self.send_command("start", params)
 
     async def get_status(self) -> Response:

shrinkray/subprocess/worker.py

@@ -169,7 +169,7 @@ class ReducerWorker:
         filename = params["file_path"]
         test = params["test"]
         parallelism = params.get("parallelism", os.cpu_count() or 1)
-        timeout = params.get("timeout", 1.0)
+        timeout = params.get("timeout")  # None means dynamic timeout
         seed = params.get("seed", 0)
         input_type = InputType[params.get("input_type", "all")]
         in_place = params.get("in_place", False)

shrinkray/tui.py

@@ -98,7 +98,7 @@ class ReductionClientProtocol(Protocol):
         file_path: str,
         test: list[str],
         parallelism: int | None = None,
-        timeout: float = 1.0,
+        timeout: float | None = None,
         seed: int = 0,
         input_type: str = "all",
         in_place: bool = False,
@@ -616,7 +616,7 @@ class ShrinkRayApp(App[None]):
         file_path: str,
         test: list[str],
         parallelism: int | None = None,
-        timeout: float = 1.0,
+        timeout: float | None = None,
         seed: int = 0,
         input_type: str = "all",
         in_place: bool = False,
@@ -657,6 +657,7 @@ class ShrinkRayApp(App[None]):
             yield Label(
                 "Shrink Ray - [h] help, [p] passes, [c] skip pass, [q] quit",
                 id="status-label",
+                markup=False,
             )
             with Vertical(id="stats-container"):
                 yield StatsDisplay(id="stats-display")
@@ -807,7 +808,7 @@ async def _validate_initial_example(
     file_path: str,
     test: list[str],
     parallelism: int | None,
-    timeout: float,
+    timeout: float | None,
     seed: int,
     input_type: str,
     in_place: bool,
@@ -855,7 +856,7 @@ def run_textual_ui(
     file_path: str,
     test: list[str],
     parallelism: int | None = None,
-    timeout: float = 1.0,
+    timeout: float | None = None,
     seed: int = 0,
     input_type: str = "all",
     in_place: bool = False,

{shrinkray-25.12.27.0.dist-info → shrinkray-25.12.27.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: shrinkray
-Version: 25.12.27.0
+Version: 25.12.27.2
 Summary: Shrink Ray
 Author-email: "David R. MacIver" <david@drmaciver.com>
 License: MIT
@@ -28,7 +28,8 @@ Requires-Dist: hypothesmith>=0.3.1; extra == "dev"
 Requires-Dist: pytest>=8.0.0; extra == "dev"
 Requires-Dist: pytest-trio>=0.8.0; extra == "dev"
 Requires-Dist: pytest-asyncio>=0.21.0; extra == "dev"
-Requires-Dist: pytest-textual-snapshot>=1.0.0; extra == "dev"
+Requires-Dist: syrupy>=5.0.0; extra == "dev"
+Requires-Dist: jinja2>=3.0.0; extra == "dev"
 Requires-Dist: coverage[toml]>=7.4.0; extra == "dev"
 Requires-Dist: pygments>=2.17.0; extra == "dev"
 Requires-Dist: basedpyright>=1.1.0; extra == "dev"
@@ -158,30 +159,3 @@ not to handle well, but it's easy to extend with additional transformations.
 It is also fairly easy to add support for new formats as needed.
 
 If you run into a test case and interestingness test that you care about that shrink ray handles badly please let me know and I'll likely see about improving its handling of that format.
-
-## Parallelism
-
-You can control the number of parallel tasks shrinkray will run with the `--parallelism` flag. By default this will be the number of CPU cores you have available
-
-Shrink Ray is designed to be able to run heavily in parallel, with a basic heuristic of aiming to be embarrassingly parallel when making no progress, mostly sequential when making progress, and smoothly scaling in between the two. It mostly succeeds at this.
-
-Currently the bottleneck on scaling to a very large number of cores is how fast the controlling Python program can generate variant test cases to try and pass them to the interestingness test. This isn't well optimised at present and I don't currently have good benchmarks for it, but I'd expect you to be able to get linear speedups on most workflows while running 10-20 test cases in parallel, and to start to struggle past that.
-
-This also depends on the performance of the interestingness test - the slower your test is to run, the more you'll be able to scale linearly with the number of cores available.
-
-I'm quite interested in getting this part to scale well, so please let me know if you find examples where it doesn't seem to work.
-
-## Bug Reports
-
-Shrink Ray is still pretty new and under-tested software, so it definitely has bugs. If you run into any, [please file an issue](https://github.com/DRMacIver/shrinkray/issues).
-
-As well as obvious bugs (crashes, etc) I'm also very interested in hearing about usability issues and cases where the reduced test case isn't very good.
-
-Requests for new features, new supported formats, etc. also welcome although I'm less likely to jump right on them.
-
-## Sponsorship
-
-Shrink Ray is something of a labour of love - I wanted to have a tool that actually put into practice many of my ideas about test-case reduction, as I think the previous state of the art was well behind where I'd like it to be.
-
-That being said, it is first and foremost designed to be a useful tool for practical engineering problems.
-If you find it useful as such, please [consider sponsoring my development of it](https://github.com/sponsors/DRMacIver).

shrinkray-25.12.27.2.dist-info/RECORD

@@ -0,0 +1,33 @@
+shrinkray/__init__.py,sha256=b5MvcvhsEGYya3GRXNbCJAlAL5IZHSsETLK_vtfmXRY,18
+shrinkray/__main__.py,sha256=K3_s96Tyoi7SxNOyoZXkfiEoSxVBL__TJ3o2Cefadmg,11093
+shrinkray/cli.py,sha256=1-qjaIchyCDd-YCdGWtK7q9j9qr6uX6AqtwW8m5QCQg,1697
+shrinkray/display.py,sha256=WYN05uqmUVpZhwi2pxr1U-wLHWZ9KiL0RUlTCBJ1N3E,2430
+shrinkray/formatting.py,sha256=tXCGnhJn-WJGpHMaLHRCAXK8aKJBbrOdiW9QGERrQEk,3121
+shrinkray/problem.py,sha256=_edENYk8OC5o_2ng0WZrhIfilhlY5IuOrqt0qWBZAuM,25979
+shrinkray/process.py,sha256=-eP8h5X0ESbkcTic8FFEzkd4-vwaZ0YI5tLxUR25L8U,1599
+shrinkray/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+shrinkray/reducer.py,sha256=t2BBjq-EZ0qBbsez4ep0tFYCkDT8RwKYnjJ2fRsXubg,20096
+shrinkray/state.py,sha256=owyDNiSeHKnCsgOQczSz796KEANem-4YWTmbmZBRE0w,24040
+shrinkray/tui.py,sha256=3RskLo6JvKdUQIHi40R5ka-F_1GkBXyA_d_SkYbLlCw,31601
+shrinkray/ui.py,sha256=xuDUwU-MM3AetvwUB7bfzav0P_drUsBrKFPhON_Nr-k,2251
+shrinkray/work.py,sha256=GEZ14Kk3bvwUxAnACvY-wom2lVWaGrELMNxrDjv03dk,8110
+shrinkray/passes/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+shrinkray/passes/bytes.py,sha256=U1sNAuqUlsaLpRSJuS9X4H7S2jOiilTUIIE9-WjSd4k,23967
+shrinkray/passes/clangdelta.py,sha256=t9EQ_kc159HRs48JwB5JvlJCsiCscrZgf2nhHCZRZX0,8419
+shrinkray/passes/definitions.py,sha256=TDDPimp7DE60xgidAE11npt2KicQAEi0UydIzjS9VHw,2469
+shrinkray/passes/genericlanguages.py,sha256=ZqTfEHUTRbkm6PiTkBc_y3Q5Q0MJAnibNEiuKDOhiS0,10432
+shrinkray/passes/json.py,sha256=FydtjpVD3lMjp9OEwv5mu8CAz96ofGxIZOzt9o6eLUA,2586
+shrinkray/passes/patching.py,sha256=1uOTir3IbywKmsg6IIhSnxHFovZTdUCS-8PSwzgza00,8936
+shrinkray/passes/python.py,sha256=3WN1lZTf5oVL8FCTGomhrCuE04wIX9ocKcmFV86NMZA,6875
+shrinkray/passes/sat.py,sha256=OboY6jsKf6lph3pAFh535plvhNOVzEF8HJ66WEqsNm4,19483
+shrinkray/passes/sequences.py,sha256=jCK1fWBxCz79u7JWSps9wf7Yru7W_FAsJwdgg--CLxU,3040
+shrinkray/subprocess/__init__.py,sha256=FyV2y05uwQ1RTZGwREI0aAVaLX1jiwRcWsdsksFmdbM,451
+shrinkray/subprocess/client.py,sha256=erqnPglPO0YNdwEKlmhB3yDo6Mfc00Lxh4T85lZhsDo,9341
+shrinkray/subprocess/protocol.py,sha256=LuHl0IkKpDzYhAGZz_EiTHNqDNq_v1ozg5aUSl7UzE4,6203
+shrinkray/subprocess/worker.py,sha256=ke-9DYFH117EpJEntkucTrn7ep7pygzmV-VXkRe1o-E,19294
+shrinkray-25.12.27.2.dist-info/licenses/LICENSE,sha256=iMKX79AuokJfIZUnGUARdUp30vVAoIPOJ7ek8TY63kk,1072
+shrinkray-25.12.27.2.dist-info/METADATA,sha256=-rrEVZeXrERITn0adTTOOi6Ui7fK6Dz-GvyFEQcHvtw,7600
+shrinkray-25.12.27.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+shrinkray-25.12.27.2.dist-info/entry_points.txt,sha256=wIZvnGyOdVeaLTiv2klnSyTe-EKkkwn4SwHh9bmJ7qk,104
+shrinkray-25.12.27.2.dist-info/top_level.txt,sha256=fLif8-rFoFOnf5h8-vs3ECkKNWQopTQh3xvl1s7pchQ,10
+shrinkray-25.12.27.2.dist-info/RECORD,,

shrinkray-25.12.27.0.dist-info/RECORD

@@ -1,33 +0,0 @@
-shrinkray/__init__.py,sha256=b5MvcvhsEGYya3GRXNbCJAlAL5IZHSsETLK_vtfmXRY,18
-shrinkray/__main__.py,sha256=AL1WNfohkA5Uu7uiCQRhNjzfUc4qKC-UjMTDraPlT8I,10919
-shrinkray/cli.py,sha256=1-qjaIchyCDd-YCdGWtK7q9j9qr6uX6AqtwW8m5QCQg,1697
-shrinkray/display.py,sha256=WYN05uqmUVpZhwi2pxr1U-wLHWZ9KiL0RUlTCBJ1N3E,2430
-shrinkray/formatting.py,sha256=tXCGnhJn-WJGpHMaLHRCAXK8aKJBbrOdiW9QGERrQEk,3121
-shrinkray/problem.py,sha256=Kp7QN10E4tzjdpoqJve8_RT26VpywzQwY0gX2VkBGCo,17277
-shrinkray/process.py,sha256=-eP8h5X0ESbkcTic8FFEzkd4-vwaZ0YI5tLxUR25L8U,1599
-shrinkray/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-shrinkray/reducer.py,sha256=xhLo_GF7qrIVoiLHed6Wt4nxjdE-9jj_7K9F76un89o,19877
-shrinkray/state.py,sha256=jqAKlirXjKK6smA2G1BnGjjoH7NgVokQhER2Ax18q64,22381
-shrinkray/tui.py,sha256=YiiCDVxGl5g5TlcZN62NMYiD_IqRV68qoRDkNcasvc4,31540
-shrinkray/ui.py,sha256=xuDUwU-MM3AetvwUB7bfzav0P_drUsBrKFPhON_Nr-k,2251
-shrinkray/work.py,sha256=GEZ14Kk3bvwUxAnACvY-wom2lVWaGrELMNxrDjv03dk,8110
-shrinkray/passes/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-shrinkray/passes/bytes.py,sha256=pX2kBeH38SM4j55f8bNF_2PVBCDo6XdUj73EvOFH9f8,23904
-shrinkray/passes/clangdelta.py,sha256=t9EQ_kc159HRs48JwB5JvlJCsiCscrZgf2nhHCZRZX0,8419
-shrinkray/passes/definitions.py,sha256=WEYtcWOBv7ghZRUtJWBxHRU3SnO000EFohfAD4le7KM,4267
-shrinkray/passes/genericlanguages.py,sha256=qbTuJgUieHqWtf7cly2tm0qdbrVXzVoWcAnFo7fny94,10400
-shrinkray/passes/json.py,sha256=AcmroHgb38Aa3aApwcuYNzmyya_vnCi2RFSVqomiDg8,2586
-shrinkray/passes/patching.py,sha256=1uOTir3IbywKmsg6IIhSnxHFovZTdUCS-8PSwzgza00,8936
-shrinkray/passes/python.py,sha256=3WN1lZTf5oVL8FCTGomhrCuE04wIX9ocKcmFV86NMZA,6875
-shrinkray/passes/sat.py,sha256=5Zv4IgGfg3SYplMAAaPLkbBNh4fuVpci4F9GdVacayA,19504
-shrinkray/passes/sequences.py,sha256=jCK1fWBxCz79u7JWSps9wf7Yru7W_FAsJwdgg--CLxU,3040
-shrinkray/subprocess/__init__.py,sha256=FyV2y05uwQ1RTZGwREI0aAVaLX1jiwRcWsdsksFmdbM,451
-shrinkray/subprocess/client.py,sha256=i2CJ7xU8_jklqDl-yphpIv-mwD0gG93jVzEDEtsdWvE,9277
-shrinkray/subprocess/protocol.py,sha256=LuHl0IkKpDzYhAGZz_EiTHNqDNq_v1ozg5aUSl7UzE4,6203
-shrinkray/subprocess/worker.py,sha256=IyO9FNQ11vWwrc2ikH-Ca3iGoc03MJOIRNUqK3R06aE,19269
-shrinkray-25.12.27.0.dist-info/licenses/LICENSE,sha256=iMKX79AuokJfIZUnGUARdUp30vVAoIPOJ7ek8TY63kk,1072
-shrinkray-25.12.27.0.dist-info/METADATA,sha256=jdpwOHRfJE7bJnBFBbYlpEw7bINLJaU_v2Pz_aBGsmw,9693
-shrinkray-25.12.27.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-shrinkray-25.12.27.0.dist-info/entry_points.txt,sha256=wIZvnGyOdVeaLTiv2klnSyTe-EKkkwn4SwHh9bmJ7qk,104
-shrinkray-25.12.27.0.dist-info/top_level.txt,sha256=fLif8-rFoFOnf5h8-vs3ECkKNWQopTQh3xvl1s7pchQ,10
-shrinkray-25.12.27.0.dist-info/RECORD,,