glitchlings 0.4.4__cp313-cp313-manylinux_2_28_x86_64.whl

glitchlings/zoo/jargoyle.py

@@ -0,0 +1,335 @@
+import random
+import re
+from collections.abc import Iterable
+from dataclasses import dataclass
+from types import ModuleType
+from typing import Any, Literal, cast
+
+from glitchlings.lexicon import Lexicon, get_default_lexicon
+
+from ._rate import resolve_rate
+from .core import AttackWave, Glitchling
+
+_wordnet_module: ModuleType | None
+
+try:  # pragma: no cover - optional WordNet dependency
+    import glitchlings.lexicon.wordnet as _wordnet_module
+except (
+    ImportError,
+    ModuleNotFoundError,
+    AttributeError,
+):  # pragma: no cover - triggered when nltk unavailable
+    _wordnet_module = None
+
+_wordnet_runtime: ModuleType | None = _wordnet_module
+
+WordNetLexicon: type[Lexicon] | None
+if _wordnet_runtime is None:
+
+    def _lexicon_dependencies_available() -> bool:
+        return False
+
+    def _lexicon_ensure_wordnet() -> None:
+        raise RuntimeError(
+            "The WordNet backend is no longer bundled by default. Install NLTK "
+            "and download its WordNet corpus manually if you need legacy synonyms."
+        )
+
+    WordNetLexicon = None
+else:
+    WordNetLexicon = cast(type[Lexicon], _wordnet_runtime.WordNetLexicon)
+    _lexicon_dependencies_available = _wordnet_runtime.dependencies_available
+    _lexicon_ensure_wordnet = _wordnet_runtime.ensure_wordnet
+
+
+ensure_wordnet = _lexicon_ensure_wordnet
+
+
+def dependencies_available() -> bool:
+    """Return ``True`` when a synonym backend is accessible."""
+    if _lexicon_dependencies_available():
+        return True
+
+    try:
+        # Fall back to the configured default lexicon (typically the bundled vector cache).
+        get_default_lexicon(seed=None)
+    except (RuntimeError, ImportError, ModuleNotFoundError, AttributeError):
+        return False
+    return True
+
+
+# Backwards compatibility for callers relying on the previous private helper name.
+_ensure_wordnet = ensure_wordnet
+
+
+PartOfSpeech = Literal["n", "v", "a", "r"]
+PartOfSpeechInput = PartOfSpeech | Iterable[PartOfSpeech] | Literal["any"]
+NormalizedPartsOfSpeech = tuple[PartOfSpeech, ...]
+
+_VALID_POS: tuple[PartOfSpeech, ...] = ("n", "v", "a", "r")
+
+
+def _split_token(token: str) -> tuple[str, str, str]:
+    """Split a token into leading punctuation, core word, and trailing punctuation."""
+    match = re.match(r"^(\W*)(.*?)(\W*)$", token)
+    if not match:
+        return "", token, ""
+    prefix, core, suffix = match.groups()
+    return prefix, core, suffix
+
+
+def _normalize_parts_of_speech(
+    part_of_speech: PartOfSpeechInput,
+) -> NormalizedPartsOfSpeech:
+    """Coerce user input into a tuple of valid WordNet POS tags."""
+    if isinstance(part_of_speech, str):
+        lowered = part_of_speech.lower()
+        if lowered == "any":
+            return _VALID_POS
+        if lowered not in _VALID_POS:
+            raise ValueError("part_of_speech must be one of 'n', 'v', 'a', 'r', or 'any'")
+        return (cast(PartOfSpeech, lowered),)
+
+    normalized: list[PartOfSpeech] = []
+    for pos in part_of_speech:
+        if pos not in _VALID_POS:
+            raise ValueError("part_of_speech entries must be one of 'n', 'v', 'a', or 'r'")
+        if pos not in normalized:
+            normalized.append(pos)
+    if not normalized:
+        raise ValueError("part_of_speech iterable may not be empty")
+    return tuple(normalized)
+
+
+@dataclass(frozen=True)
+class CandidateInfo:
+    """Metadata for a candidate token that may be replaced."""
+
+    prefix: str
+    core_word: str
+    suffix: str
+    part_of_speech: str | None
+    synonyms: list[str]
+
+
+def substitute_random_synonyms(
+    text: str,
+    rate: float | None = None,
+    part_of_speech: PartOfSpeechInput = "n",
+    seed: int | None = None,
+    rng: random.Random | None = None,
+    *,
+    replacement_rate: float | None = None,
+    lexicon: Lexicon | None = None,
+) -> str:
+    """Replace words with random lexicon-driven synonyms.
+
+    Parameters
+    ----------
+    - text: Input text.
+    - rate: Max proportion of candidate words to replace (default 0.01).
+    - part_of_speech: WordNet POS tag(s) to target. Accepts "n", "v", "a", "r",
+      any iterable of those tags, or "any" to include all four. Backends that do
+      not differentiate parts of speech simply ignore the setting.
+    - rng: Optional RNG instance used for deterministic sampling.
+    - seed: Optional seed if `rng` not provided.
+    - lexicon: Optional :class:`~glitchlings.lexicon.Lexicon` implementation to
+      supply synonyms. Defaults to the configured lexicon priority, typically the
+      packaged vector cache.
+
+    Determinism
+    - Candidates collected in left-to-right order; no set() reordering.
+    - Replacement positions chosen via rng.sample.
+    - Synonyms sourced through the lexicon; the default backend derives
+      deterministic subsets per word and part-of-speech using the active seed.
+
+    """
+    effective_rate = resolve_rate(
+        rate=rate,
+        legacy_value=replacement_rate,
+        default=0.1,
+        legacy_name="replacement_rate",
+    )
+
+    active_rng: random.Random
+    if rng is not None:
+        active_rng = rng
+    else:
+        active_rng = random.Random(seed)
+
+    active_lexicon: Lexicon
+    restore_lexicon_seed = False
+    original_lexicon_seed: int | None = None
+
+    if lexicon is None:
+        active_lexicon = get_default_lexicon(seed=seed)
+    else:
+        active_lexicon = lexicon
+        if seed is not None:
+            original_lexicon_seed = active_lexicon.seed
+            if original_lexicon_seed != seed:
+                active_lexicon.reseed(seed)
+                restore_lexicon_seed = True
+
+    try:
+        target_pos = _normalize_parts_of_speech(part_of_speech)
+
+        # Split but keep whitespace separators so we can rebuild easily
+        tokens = re.split(r"(\s+)", text)
+
+        # Collect candidate word indices (even positions are words because separators are kept)
+        candidate_indices: list[int] = []
+        candidate_metadata: dict[int, CandidateInfo] = {}
+        for idx, tok in enumerate(tokens):
+            if idx % 2 != 0 or not tok or tok.isspace():
+                continue
+
+            prefix, core_word, suffix = _split_token(tok)
+            if not core_word:
+                continue
+
+            chosen_pos: str | None = None
+            synonyms: list[str] = []
+
+            for tag in target_pos:
+                if not active_lexicon.supports_pos(tag):
+                    continue
+                synonyms = active_lexicon.get_synonyms(core_word, pos=tag)
+                if synonyms:
+                    chosen_pos = tag
+                    break
+
+            if not synonyms and active_lexicon.supports_pos(None):
+                synonyms = active_lexicon.get_synonyms(core_word, pos=None)
+
+            if synonyms:
+                candidate_indices.append(idx)
+                candidate_metadata[idx] = CandidateInfo(
+                    prefix=prefix,
+                    core_word=core_word,
+                    suffix=suffix,
+                    part_of_speech=chosen_pos,
+                    synonyms=synonyms,
+                )
+
+        if not candidate_indices:
+            return text
+
+        clamped_rate = max(0.0, effective_rate)
+        if clamped_rate == 0.0:
+            return text
+
+        population = len(candidate_indices)
+        effective_fraction = min(clamped_rate, 1.0)
+        expected_replacements = population * effective_fraction
+        max_replacements = int(expected_replacements)
+        remainder = expected_replacements - max_replacements
+        if remainder > 0.0 and active_rng.random() < remainder:
+            max_replacements += 1
+        if clamped_rate >= 1.0:
+            max_replacements = population
+        max_replacements = min(population, max_replacements)
+        if max_replacements <= 0:
+            return text
+
+        # Choose which positions to replace deterministically via rng.sample
+        replace_positions = active_rng.sample(candidate_indices, k=max_replacements)
+        # Process in ascending order to avoid affecting later indices
+        replace_positions.sort()
+
+        for pos in replace_positions:
+            metadata = candidate_metadata[pos]
+            if not metadata.synonyms:
+                continue
+
+            replacement = active_rng.choice(metadata.synonyms)
+            tokens[pos] = f"{metadata.prefix}{replacement}{metadata.suffix}"
+
+        return "".join(tokens)
+    finally:
+        if restore_lexicon_seed:
+            active_lexicon.reseed(original_lexicon_seed)
+
+
+class Jargoyle(Glitchling):
+    """Glitchling that swaps words with lexicon-driven synonyms."""
+
+    def __init__(
+        self,
+        *,
+        rate: float | None = None,
+        replacement_rate: float | None = None,
+        part_of_speech: PartOfSpeechInput = "n",
+        seed: int | None = None,
+        lexicon: Lexicon | None = None,
+    ) -> None:
+        self._param_aliases = {"replacement_rate": "rate"}
+        self._owns_lexicon = lexicon is None
+        self._external_lexicon_original_seed = (
+            lexicon.seed if isinstance(lexicon, Lexicon) else None
+        )
+        self._initializing = True
+        effective_rate = resolve_rate(
+            rate=rate,
+            legacy_value=replacement_rate,
+            default=0.01,
+            legacy_name="replacement_rate",
+        )
+        prepared_lexicon = lexicon or get_default_lexicon(seed=seed)
+        if lexicon is not None and seed is not None:
+            prepared_lexicon.reseed(seed)
+        try:
+            super().__init__(
+                name="Jargoyle",
+                corruption_function=substitute_random_synonyms,
+                scope=AttackWave.WORD,
+                seed=seed,
+                rate=effective_rate,
+                part_of_speech=part_of_speech,
+                lexicon=prepared_lexicon,
+            )
+        finally:
+            self._initializing = False
+
+    def set_param(self, key: str, value: Any) -> None:
+        super().set_param(key, value)
+
+        aliases = getattr(self, "_param_aliases", {})
+        canonical = aliases.get(key, key)
+
+        if canonical == "seed":
+            current_lexicon = getattr(self, "lexicon", None)
+            if isinstance(current_lexicon, Lexicon):
+                if getattr(self, "_owns_lexicon", False):
+                    current_lexicon.reseed(self.seed)
+                else:
+                    if self.seed is not None:
+                        current_lexicon.reseed(self.seed)
+                    else:
+                        if hasattr(self, "_external_lexicon_original_seed"):
+                            original_seed = getattr(self, "_external_lexicon_original_seed", None)
+                            current_lexicon.reseed(original_seed)
+        elif canonical == "lexicon" and isinstance(value, Lexicon):
+            if getattr(self, "_initializing", False):
+                if getattr(self, "_owns_lexicon", False):
+                    if self.seed is not None:
+                        value.reseed(self.seed)
+                else:
+                    if getattr(self, "_external_lexicon_original_seed", None) is None:
+                        self._external_lexicon_original_seed = value.seed
+                    if self.seed is not None:
+                        value.reseed(self.seed)
+                return
+
+            self._owns_lexicon = False
+            self._external_lexicon_original_seed = value.seed
+            if self.seed is not None:
+                value.reseed(self.seed)
+            elif value.seed != self._external_lexicon_original_seed:
+                value.reseed(self._external_lexicon_original_seed)
+
+
+jargoyle = Jargoyle()
+
+
+__all__ = ["Jargoyle", "dependencies_available", "ensure_wordnet", "jargoyle"]

glitchlings/zoo/mim1c.py

@@ -0,0 +1,109 @@
+import random
+from collections.abc import Collection
+from typing import Literal
+
+from confusable_homoglyphs import confusables
+
+from ._rate import resolve_rate
+from .core import AttackOrder, AttackWave, Glitchling
+
+
+def swap_homoglyphs(
+    text: str,
+    rate: float | None = None,
+    classes: list[str] | Literal["all"] | None = None,
+    banned_characters: Collection[str] | None = None,
+    seed: int | None = None,
+    rng: random.Random | None = None,
+    *,
+    replacement_rate: float | None = None,
+) -> str:
+    """Replace characters with visually confusable homoglyphs.
+
+    Parameters
+    ----------
+    - text: Input text.
+    - rate: Max proportion of eligible characters to replace (default 0.02).
+    - classes: Restrict replacements to these Unicode script classes (default
+      ["LATIN", "GREEK", "CYRILLIC"]). Use "all" to allow any.
+    - banned_characters: Characters that must never appear as replacements.
+    - seed: Optional seed if `rng` not provided.
+    - rng: Optional RNG; overrides seed.
+
+    Notes
+    -----
+    - Only replaces characters present in ``confusables.confusables_data`` with
+      single-codepoint alternatives.
+    - Maintains determinism by shuffling candidates and sampling via the provided RNG.
+
+    """
+    effective_rate = resolve_rate(
+        rate=rate,
+        legacy_value=replacement_rate,
+        default=0.02,
+        legacy_name="replacement_rate",
+    )
+
+    if rng is None:
+        rng = random.Random(seed)
+
+    if classes is None:
+        classes = ["LATIN", "GREEK", "CYRILLIC"]
+
+    target_chars = [char for char in text if char.isalnum()]
+    confusable_chars = [char for char in target_chars if char in confusables.confusables_data]
+    clamped_rate = max(0.0, effective_rate)
+    num_replacements = int(len(confusable_chars) * clamped_rate)
+    done = 0
+    rng.shuffle(confusable_chars)
+    banned_set = set(banned_characters or ())
+    for char in confusable_chars:
+        if done >= num_replacements:
+            break
+        options = [o["c"] for o in confusables.confusables_data[char] if len(o["c"]) == 1]
+        if classes != "all":
+            options = [opt for opt in options if confusables.alias(opt) in classes]
+        if banned_set:
+            options = [opt for opt in options if opt not in banned_set]
+        if not options:
+            continue
+        text = text.replace(char, rng.choice(options), 1)
+        done += 1
+    return text
+
+
+class Mim1c(Glitchling):
+    """Glitchling that swaps characters for visually similar homoglyphs."""
+
+    def __init__(
+        self,
+        *,
+        rate: float | None = None,
+        replacement_rate: float | None = None,
+        classes: list[str] | Literal["all"] | None = None,
+        banned_characters: Collection[str] | None = None,
+        seed: int | None = None,
+    ) -> None:
+        self._param_aliases = {"replacement_rate": "rate"}
+        effective_rate = resolve_rate(
+            rate=rate,
+            legacy_value=replacement_rate,
+            default=0.02,
+            legacy_name="replacement_rate",
+        )
+        super().__init__(
+            name="Mim1c",
+            corruption_function=swap_homoglyphs,
+            scope=AttackWave.CHARACTER,
+            order=AttackOrder.LAST,
+            seed=seed,
+            rate=effective_rate,
+            classes=classes,
+            banned_characters=banned_characters,
+        )
+
+
+mim1c = Mim1c()
+
+
+__all__ = ["Mim1c", "mim1c"]

glitchlings/zoo/ocr_confusions.tsv

@@ -0,0 +1,30 @@
+# Source  Replacements (space-separated)
+li h
+h li
+rn m
+m rn
+cl d
+d cl
+I l
+l I 1
+1 l I
+0 O
+O 0
+B 8
+8 B
+S 5
+5 S
+Z 2
+2 Z
+G 6
+6 G
+“ "
+” "
+‘ '
+’ '
+— -
+– -
+vv w
+w vv
+ri n
+n ri

glitchlings/zoo/redactyl.py

@@ -0,0 +1,193 @@
+import random
+import re
+from typing import Any, cast
+
+from ._rate import resolve_rate
+from ._rust_extensions import get_rust_operation
+from ._sampling import weighted_sample_without_replacement
+from ._text_utils import (
+    WordToken,
+    collect_word_tokens,
+    split_preserving_whitespace,
+)
+from .core import AttackWave, Glitchling
+
+FULL_BLOCK = "█"
+
+# Load Rust-accelerated operation if available
+_redact_words_rust = get_rust_operation("redact_words")
+
+
+def _python_redact_words(
+    text: str,
+    *,
+    replacement_char: str,
+    rate: float,
+    merge_adjacent: bool,
+    rng: random.Random,
+    unweighted: bool = False,
+) -> str:
+    """Redact random words by replacing their characters.
+
+    Parameters
+    ----------
+    - text: Input text.
+    - replacement_char: The character to use for redaction (default FULL_BLOCK).
+    - rate: Max proportion of words to redact (default 0.05).
+    - merge_adjacent: If True, merges adjacent redactions across intervening non-word chars.
+    - rng: RNG used for sampling decisions.
+    - unweighted: When True, sample words uniformly instead of by length.
+
+    """
+    tokens = split_preserving_whitespace(text)
+    word_tokens = collect_word_tokens(tokens)
+    if not word_tokens:
+        raise ValueError("Cannot redact words because the input text contains no redactable words.")
+
+    population = [token.index for token in word_tokens]
+    weights = [1.0 if unweighted else float(token.core_length) for token in word_tokens]
+
+    clamped_rate = max(0.0, min(rate, 1.0))
+    raw_quota = len(population) * clamped_rate
+    num_to_redact = int(raw_quota)
+    if clamped_rate > 0.0:
+        num_to_redact = max(1, num_to_redact)
+    num_to_redact = min(num_to_redact, len(population))
+    if num_to_redact <= 0:
+        return "".join(tokens)
+
+    indices_to_redact = weighted_sample_without_replacement(
+        population,
+        weights,
+        k=num_to_redact,
+        rng=rng,
+    )
+    indices_to_redact.sort()
+
+    token_by_index: dict[int, WordToken] = {token.index: token for token in word_tokens}
+
+    for i in indices_to_redact:
+        if i >= len(tokens):
+            break
+
+        token = token_by_index.get(i)
+        if token is None:
+            continue
+
+        prefix, core, suffix = token.prefix, token.core, token.suffix
+        tokens[i] = f"{prefix}{replacement_char * len(core)}{suffix}"
+
+    text = "".join(tokens)
+
+    if merge_adjacent:
+        text = re.sub(
+            rf"{replacement_char}\W+{replacement_char}",
+            lambda m: replacement_char * (len(m.group(0)) - 1),
+            text,
+        )
+
+    return text
+
+
+def redact_words(
+    text: str,
+    replacement_char: str = FULL_BLOCK,
+    rate: float | None = None,
+    merge_adjacent: bool = False,
+    seed: int = 151,
+    rng: random.Random | None = None,
+    *,
+    redaction_rate: float | None = None,
+    unweighted: bool = False,
+) -> str:
+    """Redact random words by replacing their characters."""
+    effective_rate = resolve_rate(
+        rate=rate,
+        legacy_value=redaction_rate,
+        default=0.025,
+        legacy_name="redaction_rate",
+    )
+
+    if rng is None:
+        rng = random.Random(seed)
+
+    clamped_rate = max(0.0, min(effective_rate, 1.0))
+    unweighted_flag = bool(unweighted)
+
+    use_rust = _redact_words_rust is not None and isinstance(merge_adjacent, bool)
+
+    if use_rust:
+        assert _redact_words_rust is not None  # Type narrowing for mypy
+        return cast(
+            str,
+            _redact_words_rust(
+                text,
+                replacement_char,
+                clamped_rate,
+                merge_adjacent,
+                unweighted_flag,
+                rng,
+            ),
+        )
+
+    return _python_redact_words(
+        text,
+        replacement_char=replacement_char,
+        rate=clamped_rate,
+        merge_adjacent=merge_adjacent,
+        rng=rng,
+        unweighted=unweighted_flag,
+    )
+
+
+class Redactyl(Glitchling):
+    """Glitchling that redacts words with block characters."""
+
+    def __init__(
+        self,
+        *,
+        replacement_char: str = FULL_BLOCK,
+        rate: float | None = None,
+        redaction_rate: float | None = None,
+        merge_adjacent: bool = False,
+        seed: int = 151,
+        unweighted: bool = False,
+    ) -> None:
+        self._param_aliases = {"redaction_rate": "rate"}
+        effective_rate = resolve_rate(
+            rate=rate,
+            legacy_value=redaction_rate,
+            default=0.025,
+            legacy_name="redaction_rate",
+        )
+        super().__init__(
+            name="Redactyl",
+            corruption_function=redact_words,
+            scope=AttackWave.WORD,
+            seed=seed,
+            replacement_char=replacement_char,
+            rate=effective_rate,
+            merge_adjacent=merge_adjacent,
+            unweighted=unweighted,
+        )
+
+    def pipeline_operation(self) -> dict[str, Any] | None:
+        replacement_char = self.kwargs.get("replacement_char")
+        rate = self.kwargs.get("rate")
+        merge_adjacent = self.kwargs.get("merge_adjacent")
+        if replacement_char is None or rate is None or merge_adjacent is None:
+            return None
+        unweighted = bool(self.kwargs.get("unweighted", False))
+        return {
+            "type": "redact",
+            "replacement_char": str(replacement_char),
+            "redaction_rate": float(rate),
+            "merge_adjacent": bool(merge_adjacent),
+            "unweighted": unweighted,
+        }
+
+
+redactyl = Redactyl()
+
+
+__all__ = ["Redactyl", "redactyl"]