ummaya 0.2.4 → 0.2.5

package/src/ummaya/tools/documents/formats/archive.py

@@ -0,0 +1,528 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Promoted archive/container document engines."""
+
+from __future__ import annotations
+
+import gzip
+import html
+import io
+import re
+import shutil
+import subprocess
+import tarfile
+import tempfile
+import zipfile
+from decimal import Decimal
+from pathlib import Path, PurePosixPath
+from typing import NoReturn
+
+from ummaya.tools.documents.engines import DocumentMutationBlockedError
+from ummaya.tools.documents.models import (
+    BlockedReason,
+    DocumentExtraction,
+    DocumentFormat,
+    DocumentPatch,
+    DocumentPatchOperation,
+    FormField,
+    MetadataValue,
+    OperationType,
+    ParagraphBlock,
+)
+
+_ARCHIVE_FORMATS = (
+    DocumentFormat.epub,
+    DocumentFormat.zip,
+    DocumentFormat.seven_z,
+    DocumentFormat.tar,
+    DocumentFormat.gz,
+)
+_ARCHIVE_BODY_RE = re.compile(r"^/archive/(?P<member>.+)/body$")
+_GZIP_BODY_TARGET = "/gzip/payload"
+_MAX_INLINE_TEXT_BYTES = 128 * 1024
+_BSDTAR_TIMEOUT_SECONDS = 15
+
+
+class ArchiveContainerDocumentEngine:
+    """Bounded writer for archive containers with child-payload replacement."""
+
+    render_engine_id = "archive-container-structural-svg"
+    render_artifact_extension = "svg"
+    render_mime_type = "image/svg+xml"
+
+    def __init__(self, document_format: DocumentFormat) -> None:
+        if document_format not in _ARCHIVE_FORMATS:
+            raise ValueError(f"unsupported archive document format: {document_format.value}")
+        self.document_format = document_format
+        self.engine_id = f"archive-container-{document_format.value}"
+
+    def inspect(self, path: Path, *, artifact_id: str) -> DocumentExtraction:
+        """Inspect archive members without extracting to the workspace."""
+        entries = _read_entries(path, document_format=self.document_format)
+        paragraphs = [
+            ParagraphBlock(
+                block_id=f"archive-member-{index}",
+                text=entry.preview,
+                source_path=entry.target_path,
+            )
+            for index, entry in enumerate(entries, start=1)
+        ]
+        fields = [
+            FormField(
+                field_id=f"archive-member-{index}",
+                label=entry.name,
+                path=entry.target_path,
+                field_type="text",
+                required=False,
+                current_value=entry.text,
+                source_confidence=Decimal("0.85"),
+            )
+            for index, entry in enumerate(entries, start=1)
+            if entry.editable
+        ]
+        return DocumentExtraction(
+            artifact_id=artifact_id,
+            paragraphs=paragraphs,
+            fields=fields,
+            metadata=_metadata(
+                path,
+                document_format=self.document_format,
+                entry_count=len(entries),
+            ),
+        )
+
+    def apply_patch(self, path: Path, patch: DocumentPatch) -> bytes:
+        """Return a repacked archive derivative after replacing child payloads."""
+        replacements: dict[str, bytes] = {}
+        gzip_replacement: bytes | None = None
+        for operation in patch.operations:
+            target_name = _operation_target_name(operation)
+            replacement = _operation_replacement(operation)
+            if self.document_format is DocumentFormat.gz:
+                if target_name != _GZIP_BODY_TARGET:
+                    _raise_unsupported_operation(operation)
+                gzip_replacement = replacement
+                continue
+            member_name = _member_name_from_target(target_name)
+            replacements[member_name] = replacement
+
+        if self.document_format is DocumentFormat.gz:
+            if gzip_replacement is None:
+                raise DocumentMutationBlockedError(
+                    BlockedReason.validation_failed,
+                    "Gzip archive patch did not include /gzip/payload.",
+                )
+            return _write_gzip_payload(gzip_replacement)
+        if not replacements:
+            raise DocumentMutationBlockedError(
+                BlockedReason.validation_failed,
+                "Archive patch did not include a child member replacement.",
+            )
+        if self.document_format in {DocumentFormat.zip, DocumentFormat.epub}:
+            return _replace_zip_members(
+                path,
+                replacements,
+                preserve_epub_mimetype=self.document_format is DocumentFormat.epub,
+            )
+        if self.document_format is DocumentFormat.seven_z:
+            return _replace_7z_members(path, replacements)
+        if self.document_format is DocumentFormat.tar:
+            return _replace_tar_members(path, replacements)
+        raise DocumentMutationBlockedError(
+            BlockedReason.unsupported_format,
+            f"Unsupported archive format: {self.document_format.value}",
+        )
+
+    def render(self, path: Path, *, artifact_id: str, output_dir: Path) -> tuple[bytes, ...]:
+        """Render archive member previews as structural SVG evidence."""
+        _ = output_dir
+        extraction = self.inspect(path, artifact_id=artifact_id)
+        lines = [paragraph.text for paragraph in extraction.paragraphs]
+        return (_render_lines_svg(lines, title=f"{self.document_format.value.upper()} archive"),)
+
+
+class _ArchiveEntry:
+    def __init__(self, *, name: str, text: str, editable: bool) -> None:
+        self.name = name
+        self.text = text
+        self.editable = editable
+
+    @property
+    def target_path(self) -> str:
+        if self.name == "gzip-payload":
+            return _GZIP_BODY_TARGET
+        return f"/archive/{self.name}/body"
+
+    @property
+    def preview(self) -> str:
+        if self.text:
+            return f"{self.name}: {self.text}"
+        return self.name
+
+
+def _read_entries(path: Path, *, document_format: DocumentFormat) -> list[_ArchiveEntry]:
+    if document_format in {DocumentFormat.zip, DocumentFormat.epub}:
+        return _read_zip_entries(path)
+    if document_format is DocumentFormat.seven_z:
+        return _read_7z_entries(path)
+    if document_format is DocumentFormat.tar:
+        return _read_tar_entries(path)
+    if document_format is DocumentFormat.gz:
+        return [_ArchiveEntry(name="gzip-payload", text=_read_gzip_text(path), editable=True)]
+    raise DocumentMutationBlockedError(
+        BlockedReason.unsupported_format,
+        f"Unsupported archive format: {document_format.value}",
+    )
+
+
+def _read_zip_entries(path: Path) -> list[_ArchiveEntry]:
+    entries: list[_ArchiveEntry] = []
+    try:
+        with zipfile.ZipFile(path) as archive:
+            for info in archive.infolist():
+                if info.is_dir() or not _safe_member_name(info.filename):
+                    continue
+                text = _read_zip_member_text(archive, info)
+                entries.append(_ArchiveEntry(name=info.filename, text=text, editable=bool(text)))
+    except zipfile.BadZipFile as exc:
+        raise DocumentMutationBlockedError(BlockedReason.corrupt, "Invalid ZIP container.") from exc
+    return entries
+
+
+def _read_tar_entries(path: Path) -> list[_ArchiveEntry]:
+    entries: list[_ArchiveEntry] = []
+    try:
+        with tarfile.open(path, mode="r:*") as archive:
+            for member in archive.getmembers():
+                if not member.isfile() or not _safe_member_name(member.name):
+                    continue
+                payload = archive.extractfile(member)
+                if payload is None:
+                    continue
+                text = _decode_inline_text(payload.read(_MAX_INLINE_TEXT_BYTES))
+                entries.append(_ArchiveEntry(name=member.name, text=text, editable=bool(text)))
+    except tarfile.TarError as exc:
+        raise DocumentMutationBlockedError(BlockedReason.corrupt, "Invalid TAR container.") from exc
+    return entries
+
+
+def _read_7z_entries(path: Path) -> list[_ArchiveEntry]:
+    with tempfile.TemporaryDirectory(prefix="ummaya-7z-read-") as temp_root:
+        root = Path(temp_root)
+        member_names = _extract_7z_archive(path, root)
+        entries: list[_ArchiveEntry] = []
+        for member_name in member_names:
+            member_path = _safe_extracted_member_path(root, member_name)
+            if member_path is None or not member_path.is_file() or member_path.is_symlink():
+                continue
+            payload = member_path.read_bytes()[:_MAX_INLINE_TEXT_BYTES]
+            text = _decode_inline_text(payload)
+            entries.append(_ArchiveEntry(name=member_name, text=text, editable=bool(text)))
+        return entries
+
+
+def _read_zip_member_text(archive: zipfile.ZipFile, info: zipfile.ZipInfo) -> str:
+    if info.file_size > _MAX_INLINE_TEXT_BYTES:
+        return ""
+    try:
+        return _decode_inline_text(archive.read(info.filename))
+    except (KeyError, RuntimeError, NotImplementedError, zipfile.BadZipFile):
+        return ""
+
+
+def _read_gzip_text(path: Path) -> str:
+    try:
+        with gzip.open(path, "rb") as archive:
+            return _decode_inline_text(archive.read(_MAX_INLINE_TEXT_BYTES))
+    except OSError as exc:
+        raise DocumentMutationBlockedError(BlockedReason.corrupt, "Invalid GZIP payload.") from exc
+
+
+def _decode_inline_text(payload: bytes) -> str:
+    if b"\x00" in payload:
+        return ""
+    try:
+        return payload.decode("utf-8").strip()
+    except UnicodeDecodeError:
+        return ""
+
+
+def _operation_target_name(operation: DocumentPatchOperation) -> str:
+    if operation.operation_type not in {OperationType.replace_text, OperationType.set_field_value}:
+        _raise_unsupported_operation(operation)
+    return operation.target_path
+
+
+def _operation_replacement(operation: DocumentPatchOperation) -> bytes:
+    if operation.value is None:
+        _raise_unsupported_operation(operation)
+    return str(operation.value).encode("utf-8")
+
+
+def _member_name_from_target(target_path: str) -> str:
+    match = _ARCHIVE_BODY_RE.fullmatch(target_path)
+    if match is None:
+        raise DocumentMutationBlockedError(
+            BlockedReason.unsupported_operation,
+            f"Archive target must match /archive/<member>/body: {target_path}",
+        )
+    member_name = match.group("member")
+    if not _safe_member_name(member_name):
+        raise DocumentMutationBlockedError(
+            BlockedReason.path_traversal_detected,
+            f"Archive member target is unsafe: {member_name}",
+        )
+    return member_name
+
+
+def _replace_zip_members(
+    path: Path,
+    replacements: dict[str, bytes],
+    *,
+    preserve_epub_mimetype: bool,
+) -> bytes:
+    buffer = io.BytesIO()
+    found: set[str] = set()
+    with (
+        zipfile.ZipFile(path) as source,
+        zipfile.ZipFile(buffer, "w", compression=zipfile.ZIP_DEFLATED) as target,
+    ):
+        if preserve_epub_mimetype and "mimetype" in source.namelist():
+            target.writestr(
+                "mimetype",
+                source.read("mimetype"),
+                compress_type=zipfile.ZIP_STORED,
+            )
+        for info in source.infolist():
+            if info.is_dir():
+                continue
+            if preserve_epub_mimetype and info.filename == "mimetype":
+                continue
+            if not _safe_member_name(info.filename):
+                raise DocumentMutationBlockedError(
+                    BlockedReason.path_traversal_detected,
+                    f"Archive member path is unsafe: {info.filename}",
+                )
+            payload = replacements.get(info.filename)
+            if payload is None:
+                payload = source.read(info.filename)
+            else:
+                found.add(info.filename)
+            target.writestr(info.filename, payload)
+    _ensure_all_replacements_found(replacements, found)
+    return buffer.getvalue()
+
+
+def _replace_tar_members(path: Path, replacements: dict[str, bytes]) -> bytes:
+    buffer = io.BytesIO()
+    found: set[str] = set()
+    with tarfile.open(path, mode="r:*") as source, tarfile.open(fileobj=buffer, mode="w") as target:
+        for member in source.getmembers():
+            if not _safe_member_name(member.name):
+                raise DocumentMutationBlockedError(
+                    BlockedReason.path_traversal_detected,
+                    f"Archive member path is unsafe: {member.name}",
+                )
+            if not member.isfile():
+                target.addfile(member)
+                continue
+            payload = replacements.get(member.name)
+            if payload is None:
+                source_payload = source.extractfile(member)
+                payload = b"" if source_payload is None else source_payload.read()
+            else:
+                found.add(member.name)
+            info = tarfile.TarInfo(member.name)
+            info.size = len(payload)
+            info.mode = member.mode
+            info.mtime = member.mtime
+            target.addfile(info, io.BytesIO(payload))
+    _ensure_all_replacements_found(replacements, found)
+    return buffer.getvalue()
+
+
+def _replace_7z_members(path: Path, replacements: dict[str, bytes]) -> bytes:
+    with tempfile.TemporaryDirectory(prefix="ummaya-7z-write-") as temp_root:
+        root = Path(temp_root) / "src"
+        root.mkdir()
+        member_names = _extract_7z_archive(path, root)
+        found: set[str] = set()
+        for member_name, payload in replacements.items():
+            member_path = _safe_extracted_member_path(root, member_name)
+            if member_path is None or not member_path.is_file() or member_path.is_symlink():
+                continue
+            member_path.write_bytes(payload)
+            found.add(member_name)
+        _ensure_all_replacements_found(replacements, found)
+
+        output_path = Path(temp_root) / "updated.7z"
+        _run_bsdtar(
+            [
+                "-cf",
+                str(output_path),
+                "--format=7zip",
+                "-C",
+                str(root),
+                *member_names,
+            ]
+        )
+        return output_path.read_bytes()
+
+
+def _write_gzip_payload(payload: bytes) -> bytes:
+    buffer = io.BytesIO()
+    with gzip.GzipFile(fileobj=buffer, mode="wb") as archive:
+        archive.write(payload)
+    return buffer.getvalue()
+
+
+def _ensure_all_replacements_found(replacements: dict[str, bytes], found: set[str]) -> None:
+    missing = sorted(set(replacements) - found)
+    if missing:
+        raise DocumentMutationBlockedError(
+            BlockedReason.validation_failed,
+            f"Archive member does not exist: {missing[0]}",
+        )
+
+
+def _safe_member_name(name: str) -> bool:
+    if not name or "\x00" in name or "\\" in name:
+        return False
+    path = PurePosixPath(name)
+    return not path.is_absolute() and not any(part in {"", ".", ".."} for part in path.parts)
+
+
+def _extract_7z_archive(path: Path, destination: Path) -> list[str]:
+    member_names = _list_7z_member_names(path)
+    _run_bsdtar(
+        [
+            "-xf",
+            str(path),
+            "-C",
+            str(destination),
+            "--no-same-owner",
+            "--no-same-permissions",
+        ]
+    )
+    _validate_extracted_7z_tree(destination, member_names)
+    return member_names
+
+
+def _list_7z_member_names(path: Path) -> list[str]:
+    result = _run_bsdtar(["-tf", str(path)], capture_output=True)
+    names = [line.strip() for line in result.stdout.decode("utf-8").splitlines() if line.strip()]
+    unsafe = [name for name in names if not _safe_member_name(name)]
+    if unsafe:
+        raise DocumentMutationBlockedError(
+            BlockedReason.path_traversal_detected,
+            f"7z archive member path is unsafe: {unsafe[0]}",
+        )
+    return names
+
+
+def _validate_extracted_7z_tree(root: Path, member_names: list[str]) -> None:
+    expected = {_normalize_member_name(name) for name in member_names}
+    actual: set[str] = set()
+    for item in root.rglob("*"):
+        relative = item.relative_to(root).as_posix()
+        if item.is_symlink():
+            raise DocumentMutationBlockedError(
+                BlockedReason.unsupported_operation,
+                f"7z archive symlink entries are not promoted: {relative}",
+            )
+        if item.is_file():
+            actual.add(relative)
+    unexpected = sorted(actual - expected)
+    if unexpected:
+        raise DocumentMutationBlockedError(
+            BlockedReason.path_traversal_detected,
+            f"7z archive extracted an unexpected member: {unexpected[0]}",
+        )
+
+
+def _safe_extracted_member_path(root: Path, member_name: str) -> Path | None:
+    normalized = _normalize_member_name(member_name)
+    if not _safe_member_name(normalized):
+        return None
+    candidate = root.joinpath(*PurePosixPath(normalized).parts)
+    try:
+        candidate.resolve().relative_to(root.resolve())
+    except ValueError:
+        return None
+    return candidate
+
+
+def _normalize_member_name(member_name: str) -> str:
+    return member_name.rstrip("/")
+
+
+def _run_bsdtar(
+    args: list[str],
+    *,
+    capture_output: bool = False,
+) -> subprocess.CompletedProcess[bytes]:
+    executable = shutil.which("bsdtar")
+    if executable is None:
+        raise DocumentMutationBlockedError(
+            BlockedReason.unsupported_operation,
+            "7z archive handling requires a local bsdtar/libarchive runtime.",
+        )
+    try:
+        return subprocess.run(  # noqa: S603 - static executable lookup plus static argv.
+            [executable, *args],
+            check=True,
+            capture_output=capture_output,
+            timeout=_BSDTAR_TIMEOUT_SECONDS,
+        )
+    except subprocess.CalledProcessError as exc:
+        raise DocumentMutationBlockedError(
+            BlockedReason.corrupt,
+            f"7z archive operation failed through bsdtar: {exc.stderr.decode('utf-8', 'replace')}",
+        ) from exc
+    except subprocess.TimeoutExpired as exc:
+        raise DocumentMutationBlockedError(
+            BlockedReason.oversized_expanded_bytes,
+            "7z archive operation exceeded the local timeout budget.",
+        ) from exc
+
+
+def _metadata(
+    path: Path,
+    *,
+    document_format: DocumentFormat,
+    entry_count: int,
+) -> dict[str, MetadataValue]:
+    return {
+        "adapter_id": f"archive-container-{document_format.value}-adapter",
+        "engine_id": f"archive-container-{document_format.value}",
+        "format": document_format.value,
+        "source_name": path.name,
+        "entry_count": entry_count,
+        "mutation_policy": "archive_child_derivative_write_render_save",
+        "render_oracle": "archive-container-structural-svg",
+    }
+
+
+def _render_lines_svg(lines: list[str], *, title: str) -> bytes:
+    escaped_title = html.escape(title)
+    safe_lines = [html.escape(line) for line in lines if line]
+    height = max(160, 72 + len(safe_lines) * 28)
+    text_nodes = [
+        f'<text x="32" y="{84 + index * 28}" font-size="15" font-family="monospace">{line}</text>'
+        for index, line in enumerate(safe_lines[:80])
+    ]
+    svg = (
+        f'<svg xmlns="http://www.w3.org/2000/svg" width="1040" height="{height}" '
+        f'viewBox="0 0 1040 {height}">'
+        '<rect width="100%" height="100%" fill="#ffffff"/>'
+        f'<text x="32" y="40" font-size="22" font-family="sans-serif" '
+        f'font-weight="700">{escaped_title}</text>' + "".join(text_nodes) + "</svg>"
+    )
+    return svg.encode("utf-8")
+
+
+def _raise_unsupported_operation(operation: DocumentPatchOperation) -> NoReturn:
+    raise DocumentMutationBlockedError(
+        BlockedReason.unsupported_operation,
+        f"Archive operation is not supported: "
+        f"{operation.operation_type.value} {operation.target_path}",
+    )

package/src/ummaya/tools/documents/formats/base.py

@@ -0,0 +1,41 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Shared engine-backed adapter protocol for document formats."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import TYPE_CHECKING, Protocol
+
+from ummaya.tools.documents.models import DocumentExtraction, DocumentFormat, KnownDocumentFormat
+
+if TYPE_CHECKING:
+    from ummaya.tools.documents.tool_defs import DocumentFieldPatch
+
+
+class DocumentFormatAdapter(Protocol):
+    """Format-scoped adapter boundary below the single document primitive."""
+
+    adapter_id: str
+    known_formats: tuple[KnownDocumentFormat, ...]
+    promoted_formats: tuple[DocumentFormat, ...]
+
+    def inspect(self, path: Path, *, artifact_id: str) -> DocumentExtraction:
+        """Return LLM-readable document IR for a local artifact."""
+
+    def normalize_fill_patches(
+        self,
+        patches: tuple[DocumentFieldPatch, ...],
+        *,
+        extraction: DocumentExtraction | None,
+    ) -> tuple[DocumentFieldPatch, ...]:
+        """Normalize adapter-specific model-facing fill targets before mutation."""
+
+
+class EngineBackedFormatAdapter(Protocol):
+    """Thin wrapper around a promoted external or optional document engine."""
+
+    document_format: DocumentFormat
+    engine_id: str
+
+    def inspect(self, path: Path, *, artifact_id: str) -> DocumentExtraction:
+        """Delegate inspection to the selected engine."""