nodes/ComfyAssets-ComfyUI_PromptManager
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat(annotations): image tag storage, vocabulary participation and capture store (#276)

Browse Source 
* feat(annotations): add image annotation schema with restore validation

Install the image annotation group beside image_favorites and saved_views:
media_annotation_metadata (one row, schema version 1), media_annotation_files
(one private locator per physical file), media_annotation_bytes (one record
per observed SHA-256 with the verified size and mtime_ns) and media_image_tags
(the ordered tag set of a byte record, joined into the global tags registry).

The group migrates in a SAVEPOINT that validates before and after creating
it, with no backfill from prompt or pair tags and no filesystem probe.
annotation_schema_error reads only: it compares the catalog objects after
whitespace compaction, the columns, foreign keys and indexes, and the
persisted values (UUIDs, hex digests, bounds, the 4096-character locator cap
measured past embedded NULs, exactly one metadata row). verify_database_file
checks it with allow_missing=not require_library_state, so old backups
restore and a migrated restore stage must carry it.

Tests pin the exact DDL, a fresh install and idempotent reopen, a current
database without the group gaining it with legacy rows untouched, every
partial or malformed shape rejected without repair, install failure rolling
back and retrying, a WAL backup/restore round trip, staged-failure
immutability of source, live and safety files with every handle closed, and
the require_library_state gate.

* feat(annotations): persist per-image tag sets in the shared vocabulary

Add database/annotations.py (AnnotationOperations, mixed into
PromptDatabase): read_image_tags, replace_image_tags and add_image_tags
run in one _pair_transaction fenced by the storage epoch, own the byte
record by exact locator and digest, always supply UUID ids, rewrite the
ordered junction and bump tag_revision plus image_index_revision.
Revision mismatches raise AnnotationCommandError("tag-conflict") with
the current set; the identical set at the current or previous revision
is replayed without writes; a union over 100 names is "tag-limit".
normalize_image_tags validates each name with validate_vocabulary_tag_name
and folds case-insensitive repeats keeping the first spelling.
refresh_image_fingerprint, current_image_tags (chunked, fingerprint
matched, newest verification wins) and iter_image_tag_records serve the
S1 cheap reads.

Vocabulary participation: _move_tag_links also carries media_image_tags,
legacy merge_tags moves image-only sources, tag rows gain image_count
(files with any byte record carrying the tag), impact gains images and
rename/delete receipts gain images_changed. vocabulary-state.mjs accepts
the optional counts and names images in the delete text.

Browser fixtures clear the three annotation data tables on reset (never
the metadata row); TESTING.md documents the schema beside favorites.

* feat(annotations): add bounded private media capture store

Add utils/media_capture.py: CaptureStore copies one rooted media file
into an owner-only prompt-manager-captures-* temp directory, streaming
it in 1 MiB chunks while computing the whole-file SHA-256, and compares
the opened descriptor with the path (dev, ino where meaningful, size,
mtime_ns, ctime_ns) before and after the copy; any disagreement raises
CaptureChanged and removes the copy. Symlinked files or directories,
root escapes, non-media names and directories raise CaptureDenied; the
locator is normcase(join(realpath(root), relative)), the listing's
identity, so a symlinked root matches it.

Limits are the plan's: 256 MiB per file on stat and stream
(CaptureTooLarge), 8 completed captures, 512 MiB reserved and 2
capture/verify workers with nonblocking admission (CaptureCapacity),
900 s idle and 1800 s hard lifetimes on an injected clock. acquire and
release_use keep an in-use capture past expiry, release and
invalidate_all until its last use ends; close removes only owned files
and the owned directory. A header-only probe of the copy marks Generate
unsupported above 4096 px per side or 16,777,216 px, for unreadable
headers and for video; verify re-hashes the current file without a
copy. Every handle closes before unlink.

tests/test_media_capture.py covers exact bytes and owner-only modes,
alias and escape denial, the symlinked-root locator, change detection
during the copy (size, same-size rewrite with restored mtime through
the POSIX change time, inode replacement), both size caps, count, byte
and worker admission, idle and hard expiry with in-use retention,
cleanup scope, handle release at unlink, dimension and video
readiness, and the allowlist equality with the inspection service.
TESTING.md documents the store and the Windows change-time limit.

* test(annotations): wait on the POSIX change time only where it exists

The same-size rewrite case waits for the filesystem clock to pass the
source's ctime so the detection is deterministic on Linux's coarse
timestamps; Windows reports the creation time as st_ctime, where the
probe could never advance, so the wait runs only on POSIX.

* fix(annotations): compare the change time only between fstat reads

Python 3.12+ on Windows reports the change time from stat but the
creation time from fstat, so a freshly written file whose creation and
write straddle a clock tick failed the pre-read identity check with
CaptureChanged. The path-versus-descriptor checks now compare size,
mtime_ns and dev/ino; ctime_ns is compared between the fstat before
and after the copy, where it still detects a same-size rewrite with a
restored mtime on POSIX.

The symlinked-root test asserted the locator against the unresolved
temp path, which macOS (/var -> /private/var) and Windows (short names)
resolve differently; it now compares against the real path.

* test(browser): give the inspector evidence matrix the concept comparisons' budget

The touch run of 'full inspector attaches matched theme and mobile
evidence with reachable content' captures 32 screenshots across four
viewports and took 23.5 s, then 31.6 s, in the sharded coverage job,
hitting the 30 s default. Like the concept comparison specs it now sets
an explicit 120 s timeout; the CI-like pinned run (4 CPUs, 3 workers,
repeat-each 3) passes 198/198 with the matrix at 18-20 s on touch.

* fix(annotations): compare capture identity like with like and keep paths out of errors

Native Windows CI (Python 3.12/3.13) rejected untouched files with
CaptureChanged because the path stat and the descriptor fstat disagree
there (change time versus creation time, possibly dev/ino encodings).
The descriptor is now checked against the path only on the API-stable
facts (size, mtime_ns); the full identity (size, mtime_ns, ctime_ns,
dev/ino where meaningful) is compared between the fstat before and
after the copy and between the path stats before and after. A test
injects shifted fstat results to simulate the Windows discrepancy, and
a fresh file is captured twice with no platform guard.

Error messages carry strerror only, never the path (the path goes to
the debug log); a test asserts no CaptureError names the root. Files
are unlinked outside the store lock; close() during an in-flight
capture lets that worker remove its temp file and then the directory.
get() documents that readers of capture.path must acquire/release_use.
The test fixture resolves its temp root (macOS /private/var, Windows
8.3 names). TESTING.md states the real identity contract.

* fix(annotations): move image revisions when a fingerprint or the vocabulary changes

A replayed save that observes a new verified (size, mtime_ns) now bumps
image_index_revision, like refresh_image_fingerprint, so caches keyed
on it see the record become current again; an identical replay still
moves nothing.

Vocabulary rename, merge and delete (native and legacy paths) bump the
tag_revision of every image set carrying the source tag, in the same
transaction and before the links move or cascade, plus one
image_index_revision bump when any set was affected. A Tag dialog
opened before the change therefore gets tag-conflict with the renamed
set as current instead of recreating the old name.

* test(annotations): accept change detection wherever fstat exposes a change time

Native Windows CI on Python 3.13 detected the same-size, mtime-restored
rewrite through the descriptor's change time (3.10-3.12 report the
creation time there and complete the torn copy). The test asserted the
platform, not the signal: it now accepts CaptureChanged plus temp
removal wherever it happens, requires detection on POSIX and on
Windows 3.13+, and otherwise requires the digest to describe the
copied bytes. A new test injects a drifting fstat change time so the
detection path runs on every platform; TESTING.md records the limit as
"Windows before Python 3.13".

* test(annotations): assert the rewrite outcome, not the platform

Native Windows CI detected the same-size, mtime-restored rewrite on
3.12 and missed it on 3.13 in the same run: the path stat's change time
(Python 3.12+ fast path) only moves when the clock ticks between the
pre-stat and the rewrite, and the tick wait was skipped on Windows. The
wait now runs everywhere with fresh probe names (Windows tunnels the
creation time of a reused name), and the test asserts the outcome: a
detected change leaves no temp file; an undetected one is accepted on
Windows only and must produce a digest of the copied bytes; POSIX must
detect it. The injected-fstat test keeps the detection path covered on
every platform.

* fix(images): keep the Filters label node across a listing reload

Every listing reload reports the active refinement count, and
setActiveFilters rewrote the button's textContent even when the label
was unchanged, replacing its text node. WebKit drops a click whose
mousedown hit a node that was detached before mouseup, so a Safari
user who types a search and clicks Filters inside the 250 ms debounce
lost the click, and the webkit page workflows timed out in the
source-change recovery tests (1 in ~30 runs of the pinned image; an
instrumented probe showed pointerdown/mousedown, a DOM mutation,
pointerup/mouseup and no click event).

The label is rewritten only when it changes. The toolbar unit test
counts textContent writes (1 for an unchanged count before, 0 after),
and a workspace-images test pins the text node identity across a
search reload: 0/6 before the fix on chromium and webkit, 6/6 after,
with the recovery and Refine tests 36/36 in the pinned CI-like run.
This commit is contained in:
Vito
2026-10-22 07:19:53 -07:00
committed by GitHub
parent bc14f29db7
commit b7cf210f0f
27 changed files with 4300 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
database/annotation_schema.py
+290
View File
@@ -0,0 +1,290 @@
"""Image annotation schema group (#249): file locators, byte records, image tags.
The group installs like ``image_favorites`` and ``saved_views``: optional in old
backups, required once a database is current. It owns
- ``media_annotation_metadata``: one row, the group's schema version and the
image index revision;
- ``media_annotation_files``: one private canonical locator per physical file
(``normcase(join(realpath(root), relative))``, server-side only);
- ``media_annotation_bytes``: one record per observed whole-file SHA-256 of a
file, with the size and ``mtime_ns`` verified when those bytes were last seen
and the tag set's revision;
- ``media_image_tags``: the ordered image tag set of one byte record, joined
into the global ``tags`` registry so vocabulary maintenance carries image
links like prompt and pair links.
Installation never backfills anything from prompt or pair tags and never
touches the filesystem. Validation reads only: the catalog objects are compared
after whitespace compaction, columns, foreign keys and indexes through the
PRAGMAs, and persisted values in single bounded SQL passes.
"""
import re
import sqlite3
from typing import Optional
from .image_favorites import MAX_KEY_LENGTH
SCHEMA_VERSION = 1
METADATA_TABLE = "media_annotation_metadata"
FILES_TABLE = "media_annotation_files"
BYTES_TABLE = "media_annotation_bytes"
IMAGE_TAGS_TABLE = "media_image_tags"
TABLES = (METADATA_TABLE, FILES_TABLE, BYTES_TABLE, IMAGE_TAGS_TABLE)
MAX_LOCATOR_LENGTH = MAX_KEY_LENGTH
_TIMESTAMP_BYTES = 64
_TABLE_DDL = {
METADATA_TABLE: (
f"CREATE TABLE {METADATA_TABLE} (singleton_id INTEGER PRIMARY KEY "
"CHECK(singleton_id=1), schema_version INTEGER NOT NULL "
f"CHECK(schema_version={SCHEMA_VERSION}), "
"image_index_revision INTEGER NOT NULL DEFAULT 0 "
"CHECK(image_index_revision>=0))"
),
FILES_TABLE: (
f"CREATE TABLE {FILES_TABLE} (id TEXT PRIMARY KEY CHECK(length(id)=36), "
"locator TEXT NOT NULL UNIQUE, created_at TEXT NOT NULL)"
),
BYTES_TABLE: (
f"CREATE TABLE {BYTES_TABLE} (id TEXT PRIMARY KEY CHECK(length(id)=36), "
f"file_id TEXT NOT NULL REFERENCES {FILES_TABLE}(id) ON DELETE CASCADE, "
"sha256 TEXT NOT NULL CHECK(length(sha256)=64 AND sha256 NOT GLOB "
"'*[^0-9a-f]*'), size INTEGER NOT NULL CHECK(size>=0), "
"mtime_ns INTEGER NOT NULL, verified_at TEXT NOT NULL, "
"first_seen_at TEXT NOT NULL, tag_revision INTEGER NOT NULL DEFAULT 0 "
"CHECK(tag_revision>=0), UNIQUE(file_id, sha256))"
),
IMAGE_TAGS_TABLE: (
f"CREATE TABLE {IMAGE_TAGS_TABLE} (byte_id TEXT NOT NULL REFERENCES "
f"{BYTES_TABLE}(id) ON DELETE CASCADE, tag_id INTEGER NOT NULL "
"REFERENCES tags(id) ON DELETE CASCADE, position INTEGER NOT NULL "
"CHECK(position>=0), PRIMARY KEY(byte_id, tag_id))"
),
}
_INDEX_DDL = {
"idx_media_annotation_bytes_fingerprint": (
"CREATE INDEX idx_media_annotation_bytes_fingerprint ON "
f"{BYTES_TABLE}(file_id, size, mtime_ns)"
),
"idx_media_image_tags_tag": (
f"CREATE INDEX idx_media_image_tags_tag ON {IMAGE_TAGS_TABLE}(tag_id)"
),
"idx_media_image_tags_position": (
"CREATE UNIQUE INDEX idx_media_image_tags_position ON "
f"{IMAGE_TAGS_TABLE}(byte_id, position)"
),
}
# SQLite names the indexes behind TEXT primary keys and UNIQUE constraints
# ``sqlite_autoindex_<table>_<n>``; their catalog rows carry no SQL.
_AUTOINDEXES = {FILES_TABLE: 2, BYTES_TABLE: 2, IMAGE_TAGS_TABLE: 1}
# (name, declared type, not null, default, primary key position)
_COLUMNS = {
METADATA_TABLE: (
("singleton_id", "INTEGER", 0, None, 1),
("schema_version", "INTEGER", 1, None, 0),
("image_index_revision", "INTEGER", 1, "0", 0),
),
FILES_TABLE: (
("id", "TEXT", 0, None, 1),
("locator", "TEXT", 1, None, 0),
("created_at", "TEXT", 1, None, 0),
),
BYTES_TABLE: (
("id", "TEXT", 0, None, 1),
("file_id", "TEXT", 1, None, 0),
("sha256", "TEXT", 1, None, 0),
("size", "INTEGER", 1, None, 0),
("mtime_ns", "INTEGER", 1, None, 0),
("verified_at", "TEXT", 1, None, 0),
("first_seen_at", "TEXT", 1, None, 0),
("tag_revision", "INTEGER", 1, "0", 0),
),
IMAGE_TAGS_TABLE: (
("byte_id", "TEXT", 1, None, 1),
("tag_id", "INTEGER", 1, None, 2),
("position", "INTEGER", 1, None, 0),
),
}
# (column, parent table, parent column, on delete)
_FOREIGN_KEYS = {
METADATA_TABLE: set(),
FILES_TABLE: set(),
BYTES_TABLE: {("file_id", FILES_TABLE, "id", "CASCADE")},
IMAGE_TAGS_TABLE: {
("byte_id", BYTES_TABLE, "id", "CASCADE"),
("tag_id", "tags", "id", "CASCADE"),
},
}
def _compact(sql: str) -> str:
return re.sub(r"\s+", "", sql).lower()
def _uuid(column: str) -> str:
"""SQL: *column* is a lowercase hyphenated 36-character UUID string."""
return (
f"typeof({column}) = 'text' AND length({column}) = 36 "
f"AND substr({column}, 9, 1) = '-' AND substr({column}, 14, 1) = '-' "
f"AND substr({column}, 19, 1) = '-' AND substr({column}, 24, 1) = '-' "
f"AND length(replace({column}, '-', '')) = 32 "
f"AND replace({column}, '-', '') NOT GLOB '*[^0-9a-f]*'"
)
def _timestamp(column: str) -> str:
return (
f"typeof({column}) = 'text' "
f"AND length(CAST({column} AS BLOB)) <= {_TIMESTAMP_BYTES}"
)
def _non_negative(column: str) -> str:
return f"typeof({column}) = 'integer' AND {column} >= 0"
# Every row of a table must satisfy its predicate; one pass each, LIMIT 1.
_ROW_PREDICATES = {
FILES_TABLE: " AND ".join(
(
_uuid("id"),
"typeof(locator) = 'text'",
# UTF-8 is at most 4 bytes per character; the decoded length of
# the few candidates above the byte bound is checked in Python.
f"length(CAST(locator AS BLOB)) <= {MAX_LOCATOR_LENGTH * 4}",
_timestamp("created_at"),
)
),
BYTES_TABLE: " AND ".join(
(
_uuid("id"),
_uuid("file_id"),
"typeof(sha256) = 'text' AND length(sha256) = 64 "
"AND sha256 NOT GLOB '*[^0-9a-f]*'",
_non_negative("size"),
"typeof(mtime_ns) = 'integer'",
_timestamp("verified_at"),
_timestamp("first_seen_at"),
_non_negative("tag_revision"),
)
),
IMAGE_TAGS_TABLE: " AND ".join(
(_uuid("byte_id"), "typeof(tag_id) = 'integer'", _non_negative("position"))
),
}
def _expected_catalog() -> set:
objects = {("table", name, _compact(sql)) for name, sql in _TABLE_DDL.items()}
objects |= {("index", name, _compact(sql)) for name, sql in _INDEX_DDL.items()}
objects |= {
("index", f"sqlite_autoindex_{table}_{n}", "")
for table, count in _AUTOINDEXES.items()
for n in range(1, count + 1)
}
return objects
def _catalog(conn: sqlite3.Connection) -> list:
names = TABLES + tuple(_INDEX_DDL)
marks = ",".join("?" * len(names))
table_marks = ",".join("?" * len(TABLES))
return conn.execute( # only "?" placeholders are interpolated
"SELECT type,name,sql FROM sqlite_master "
f"WHERE name IN ({marks}) OR tbl_name IN ({table_marks})", # nosec B608
names + TABLES,
).fetchall()
def _structure_error(conn: sqlite3.Connection) -> Optional[str]:
for table, expected in _COLUMNS.items():
columns = tuple(
(row[1], row[2].upper(), row[3], row[4], row[5])
for row in conn.execute(f"PRAGMA table_info({table})")
)
if columns != expected:
return f"Image annotations table {table} has invalid columns"
foreign_keys = {
(row[3], row[2], row[4], row[6])
for row in conn.execute(f"PRAGMA foreign_key_list({table})")
}
if foreign_keys != _FOREIGN_KEYS[table]:
return f"Image annotations table {table} has invalid foreign keys"
if conn.execute(f"PRAGMA foreign_key_check({table})").fetchone():
return f"Image annotations table {table} contains foreign key violations"
return None
def _value_error(conn: sqlite3.Connection) -> Optional[str]:
# Only this module's constants are interpolated into the SQL below.
rows = conn.execute(
"SELECT COUNT(*), MIN(typeof(singleton_id) = 'integer' " # nosec B608
"AND singleton_id = 1 AND typeof(schema_version) = 'integer' "
f"AND schema_version = {SCHEMA_VERSION} "
f"AND {_non_negative('image_index_revision')}) FROM {METADATA_TABLE}"
).fetchone()
if rows[0] != 1 or rows[1] != 1:
return "Image annotations metadata is invalid"
for table, predicate in _ROW_PREDICATES.items():
if conn.execute(
f"SELECT 1 FROM {table} WHERE NOT ({predicate}) LIMIT 1" # nosec B608
).fetchone():
return f"Image annotations table {table} contains invalid values"
# length() stops at an embedded NUL, so the SQL pass bounds bytes and the
# few candidates over the character bound are measured decoded here.
for (raw,) in conn.execute(
f"SELECT CAST(locator AS BLOB) FROM {FILES_TABLE} " # nosec B608
f"WHERE length(CAST(locator AS BLOB)) > {MAX_LOCATOR_LENGTH}"
):
if len(bytes(raw).decode("utf-8", "replace")) > MAX_LOCATOR_LENGTH:
return f"Image annotations table {FILES_TABLE} contains invalid values"
return None
def annotation_schema_error(
conn: sqlite3.Connection, *, allow_missing: bool = True
) -> Optional[str]:
"""Reject partial or unsupported annotation schemas and invalid rows.
Reads only: nothing is ever repaired. A database without any of the
group's objects is acceptable for backups that predate it unless
``allow_missing`` is false.
"""
objects = _catalog(conn)
if not objects:
return None if allow_missing else "Missing image annotation schema"
present = {name for kind, name, _sql in objects if kind == "table"}
if present != set(TABLES):
return "Incomplete image annotation schema"
actual = {(kind, name, _compact(sql or "")) for kind, name, sql in objects}
if actual != _expected_catalog():
return "Image annotations has an invalid schema definition"
return _structure_error(conn) or _value_error(conn)
def migrate_annotation_state(conn: sqlite3.Connection) -> None:
"""Install the group atomically; validate before and after creating it."""
conn.execute("SAVEPOINT media_annotation_schema")
try:
error = annotation_schema_error(conn)
if error:
raise sqlite3.DatabaseError(error)
for sql in _TABLE_DDL.values():
conn.execute(sql.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
for sql in _INDEX_DDL.values():
conn.execute(sql.replace("INDEX", "INDEX IF NOT EXISTS", 1))
conn.execute(
f"INSERT OR IGNORE INTO {METADATA_TABLE} "
"(singleton_id, schema_version, image_index_revision) VALUES (1, ?, 0)",
(SCHEMA_VERSION,),
)
error = annotation_schema_error(conn, allow_missing=False)
if error:
raise sqlite3.DatabaseError(error)
conn.execute("RELEASE media_annotation_schema")
except BaseException:
conn.execute("ROLLBACK TO media_annotation_schema")
conn.execute("RELEASE media_annotation_schema")
raise
database/annotations.py
+457
View File
@@ -0,0 +1,457 @@
"""Per-image tag sets (#249): byte-exact storage in the shared vocabulary.
A tag set belongs to one byte record (``media_annotation_bytes``): the
whole-file SHA-256 of one physical file, owned by the file's private locator
(``media_annotation_files``). Callers pass verified facts (digest, size,
``mtime_ns``); nothing here reads the filesystem or resolves an owner through
prompt links, so tagging one image never changes another image, a prompt or a
pair.
Commands run in one ``_pair_transaction`` fenced by the storage epoch, like the
native Library and vocabulary commands. A save is a full replacement guarded
by the set's ``tag_revision``; the identical set sent again (a lost-response
retry, or the current set at its own revision) answers ``replayed: true`` and
writes nothing. ``add_image_tags`` unions into the existing set, keeping the
stored spelling. Names enter the global ``tags`` registry exactly as spelled,
so vocabulary rename, merge and delete carry image links with prompt and pair
links (``PromptDatabase._move_tag_links``).
Cheap reads (spec amendment S1) treat a byte record as *current* only while
the file's present ``(size, mtime_ns)`` equals the record's verified
fingerprint; a file without a current record shows no image tags.
"""
import re
import sqlite3
from contextlib import contextmanager
from typing import Iterator
from uuid import uuid4
from .annotation_schema import (
BYTES_TABLE,
FILES_TABLE,
IMAGE_TAGS_TABLE,
MAX_LOCATOR_LENGTH,
METADATA_TABLE,
)
from .library_commands import _UUID, LibraryCommandError, _validate_epoch
from .models import utc_now_iso
from .vocabulary import validate_vocabulary_tag_name
MAX_IMAGE_TAGS = 100
_LOOKUP_CHUNK = 500
_DIGEST = re.compile(r"[0-9a-f]{64}")
_MAX_INT = 2**63 - 1
_MIN_INT = -(2**63)
# Only this module's table constants are interpolated into the SQL below.
_OWNER = f"SELECT id FROM {FILES_TABLE} WHERE locator=?" # nosec B608
_INSERT_OWNER = (
f"INSERT INTO {FILES_TABLE}(id,locator,created_at) VALUES(?,?,?)" # nosec B608
)
_BYTE = (
f"SELECT id,tag_revision,size,mtime_ns FROM {BYTES_TABLE} " # nosec B608
"WHERE file_id=? AND sha256=?"
)
_INSERT_BYTE = (
f"INSERT INTO {BYTES_TABLE}(id,file_id,sha256,size,mtime_ns," # nosec B608
"verified_at,first_seen_at,tag_revision) VALUES(?,?,?,?,?,?,?,0)"
)
_REFRESH_BYTE = (
f"UPDATE {BYTES_TABLE} SET size=?,mtime_ns=?,verified_at=? WHERE id=?" # nosec B608
)
_REFRESH_BY_LOCATOR = (
f"UPDATE {BYTES_TABLE} SET size=?,mtime_ns=?,verified_at=? " # nosec B608
f"WHERE sha256=? AND file_id=({_OWNER})"
)
_BYTE_TAGS = (
f"SELECT t.name FROM {IMAGE_TAGS_TABLE} mit " # nosec B608
"JOIN tags t ON t.id=mit.tag_id WHERE mit.byte_id=? ORDER BY mit.position"
)
_CLEAR_TAGS = f"DELETE FROM {IMAGE_TAGS_TABLE} WHERE byte_id=?" # nosec B608
_INSERT_TAG = (
f"INSERT INTO {IMAGE_TAGS_TABLE}(byte_id,tag_id,position) " # nosec B608
"VALUES(?,?,?)"
)
_BUMP_REVISION = (
f"UPDATE {BYTES_TABLE} SET tag_revision=tag_revision+1 WHERE id=?" # nosec B608
)
_REVISION = f"SELECT tag_revision FROM {BYTES_TABLE} WHERE id=?" # nosec B608
_BUMP_INDEX = (
f"UPDATE {METADATA_TABLE} SET " # nosec B608
"image_index_revision=image_index_revision+1 WHERE singleton_id=1"
)
_READ = (
"SELECT b.id,b.tag_revision," # nosec B608
f"(SELECT COUNT(*) FROM {BYTES_TABLE} o WHERE o.file_id=f.id AND o.sha256!=?) "
f"FROM {FILES_TABLE} f LEFT JOIN {BYTES_TABLE} b "
"ON b.file_id=f.id AND b.sha256=? WHERE f.locator=?"
)
_FINGERPRINTS = (
"SELECT f.locator,b.id,b.size,b.mtime_ns,b.verified_at " # nosec B608
f"FROM {FILES_TABLE} f JOIN {BYTES_TABLE} b ON b.file_id=f.id "
"WHERE f.locator IN ({marks})"
)
_TAGS_OF = (
f"SELECT mit.byte_id,t.name FROM {IMAGE_TAGS_TABLE} mit " # nosec B608
"JOIN tags t ON t.id=mit.tag_id WHERE mit.byte_id IN ({marks}) "
"ORDER BY mit.byte_id,mit.position"
)
_RECORDS = (
"SELECT f.locator,b.id,b.size,b.mtime_ns,t.name " # nosec B608
f"FROM {FILES_TABLE} f JOIN {BYTES_TABLE} b ON b.file_id=f.id "
f"LEFT JOIN {IMAGE_TAGS_TABLE} mit ON mit.byte_id=b.id "
"LEFT JOIN tags t ON t.id=mit.tag_id "
"ORDER BY f.locator,b.first_seen_at,b.id,mit.position"
)
class AnnotationCommandError(LibraryCommandError):
"""An image tag command lost a race or hit the set's bound.
``code`` is ``tag-conflict`` (the set changed since it was read) or
``tag-limit`` (the union would exceed ``MAX_IMAGE_TAGS``); ``current`` is
the stored set ``{sha256, tags, tag_revision}`` read inside the
transaction, so the caller can show it.
"""
def __init__(self, code: str, epoch: str, current: dict | None = None):
super().__init__(code, epoch)
self.current = current
def normalize_image_tags(names: object) -> list[str]:
"""Trim and validate each name, fold case-insensitive repeats, cap at 100.
The first occurrence keeps its spelling and position. Raises ``ValueError``
for anything but a list of valid vocabulary names or for more than
``MAX_IMAGE_TAGS`` distinct names.
"""
if not isinstance(names, list):
raise ValueError("tags must be a list of strings")
result: list[str] = []
seen: set[str] = set()
for value in names:
name = validate_vocabulary_tag_name(value)
key = name.casefold()
if key in seen:
continue
seen.add(key)
result.append(name)
if len(result) > MAX_IMAGE_TAGS:
raise ValueError(f"An image can carry at most {MAX_IMAGE_TAGS} tags")
return result
def _validate_locator(locator: object) -> None:
if not isinstance(locator, str) or not 0 < len(locator) <= MAX_LOCATOR_LENGTH:
raise ValueError(
f"locator must be a nonblank path of at most {MAX_LOCATOR_LENGTH} chars"
)
def _validate_digest(sha256: object) -> None:
if not isinstance(sha256, str) or not _DIGEST.fullmatch(sha256):
raise ValueError("sha256 must be 64 lowercase hexadecimal characters")
def _validate_int(value: object, name: str, minimum: int) -> None:
if type(value) is not int or not minimum <= value <= _MAX_INT:
raise ValueError(f"{name} must be an integer not below {minimum}")
def _validate_target(locator: object, sha256: object, size: object, mtime_ns: object):
_validate_locator(locator)
_validate_digest(sha256)
_validate_int(size, "size", 0)
_validate_int(mtime_ns, "mtime_ns", _MIN_INT)
def _validate_fingerprint(locator: object, fingerprint: object) -> None:
_validate_locator(locator)
if (
not isinstance(fingerprint, tuple)
or len(fingerprint) != 2
or any(type(value) is not int for value in fingerprint)
):
raise ValueError("stats must map locators to (size, mtime_ns) integer pairs")
def _new_id() -> str:
"""A canonical UUID; the schema's CHECK lets NULL through, this does not."""
value = str(uuid4())
if not _UUID.fullmatch(value):
raise ValueError("Annotation ids must be canonical UUIDs")
return value
def _receipt(epoch: str, sha256: str, tags: list, revision: int, replayed: bool):
return {
"library_epoch": epoch,
"sha256": sha256,
"tags": list(tags),
"tag_revision": revision,
"replayed": replayed,
}
class AnnotationOperations:
"""PromptDatabase mixin: image tag sets, fingerprints and cheap reads.
``_`` helpers take the open ``_pair_transaction`` connection.
"""
@contextmanager
def _annotation_read(self) -> Iterator[sqlite3.Connection]:
"""One deferred read transaction, so paired SELECTs see one snapshot."""
conn = self.model.get_connection()
if conn.in_transaction:
raise sqlite3.ProgrammingError(
"Annotation reads require their own snapshot"
)
with conn:
conn.execute("BEGIN")
yield conn
@staticmethod
def _annotation_owner(conn: sqlite3.Connection, locator: str, now: str) -> str:
row = conn.execute(_OWNER, (locator,)).fetchone()
if row is not None:
return row[0]
file_id = _new_id()
conn.execute(_INSERT_OWNER, (file_id, locator, now))
return file_id
@staticmethod
def _annotation_byte(
conn: sqlite3.Connection,
file_id: str,
sha256: str,
size: int,
mtime_ns: int,
now: str,
) -> tuple[str, int, bool, bool]:
"""The byte record for these bytes, refreshed or created.
Returns ``(id, tag_revision, created, refingerprinted)``; the last is
True when an existing record's verified ``(size, mtime_ns)`` changed,
which moves its currency for cheap reads (the caller bumps the index
revision when nothing else does).
"""
row = conn.execute(_BYTE, (file_id, sha256)).fetchone()
if row is None:
byte_id = _new_id()
conn.execute(
_INSERT_BYTE, (byte_id, file_id, sha256, size, mtime_ns, now, now)
)
return byte_id, 0, True, False
conn.execute(_REFRESH_BYTE, (size, mtime_ns, now, row[0]))
return row[0], row[1], False, (row[2], row[3]) != (size, mtime_ns)
@staticmethod
def _byte_tags(conn: sqlite3.Connection, byte_id: str) -> list[str]:
return [row[0] for row in conn.execute(_BYTE_TAGS, (byte_id,))]
def _write_image_tags(
self, conn: sqlite3.Connection, byte_id: str, tags: list[str]
) -> int:
"""Rewrite the junction in order, bump both revisions; return the new one."""
tag_ids = self._ensure_tags(conn, tags)
conn.execute(_CLEAR_TAGS, (byte_id,))
conn.executemany(
_INSERT_TAG,
[(byte_id, tag_ids[name], position) for position, name in enumerate(tags)],
)
conn.execute(_BUMP_REVISION, (byte_id,))
conn.execute(_BUMP_INDEX)
return conn.execute(_REVISION, (byte_id,)).fetchone()[0]
def _current_set(self, conn, sha256, byte_id, revision) -> dict:
return {
"sha256": sha256,
"tags": self._byte_tags(conn, byte_id),
"tag_revision": revision,
}
def read_image_tags(self, locator: str, sha256: str) -> dict:
"""The stored set of these exact bytes, or an empty set at revision 0.
``prior_count`` is the number of other byte records at the locator
(earlier or later bytes of the same file).
"""
_validate_locator(locator)
_validate_digest(sha256)
with self._annotation_read() as conn:
row = conn.execute(_READ, (sha256, sha256, locator)).fetchone()
if row is None:
return {"tags": [], "tag_revision": 0, "prior_count": 0}
tags = self._byte_tags(conn, row[0]) if row[0] is not None else []
return {"tags": tags, "tag_revision": row[1] or 0, "prior_count": row[2]}
def replace_image_tags(
self,
*,
locator: str,
sha256: str,
size: int,
mtime_ns: int,
expected_epoch: str,
expected_tag_revision: int,
tags: list,
) -> dict:
"""Replace the set of these exact bytes, guarded by epoch and revision.
Creates the owner and byte record when the bytes are new (an explicit
empty set is a real save). The identical set at the current revision,
or one behind it (a lost-response retry), is ``replayed`` and writes
nothing; any other revision mismatch is ``tag-conflict`` with the
current set and no writes.
"""
_validate_target(locator, sha256, size, mtime_ns)
_validate_epoch(expected_epoch)
_validate_int(expected_tag_revision, "expected_tag_revision", 0)
tags = normalize_image_tags(tags)
with self._pair_transaction() as conn:
epoch = self._library_epoch(conn, expected_epoch)
now = utc_now_iso()
file_id = self._annotation_owner(conn, locator, now)
byte_id, revision, created, refingerprinted = self._annotation_byte(
conn, file_id, sha256, size, mtime_ns, now
)
current = self._byte_tags(conn, byte_id)
if (
not created
and current == tags
and expected_tag_revision in (revision, revision - 1)
):
if refingerprinted:
conn.execute(_BUMP_INDEX)
return _receipt(epoch, sha256, current, revision, True)
if expected_tag_revision != revision:
raise AnnotationCommandError(
"tag-conflict",
epoch,
{"sha256": sha256, "tags": current, "tag_revision": revision},
)
revision = self._write_image_tags(conn, byte_id, tags)
return _receipt(epoch, sha256, tags, revision, False)
def add_image_tags(
self,
*,
locator: str,
sha256: str,
size: int,
mtime_ns: int,
expected_epoch: str,
tags: list,
) -> dict:
"""Union ``tags`` into the set of these exact bytes, keeping stored spelling.
``added`` counts the names that were new (case-insensitively). A union
over ``MAX_IMAGE_TAGS`` is ``tag-limit`` with the current set and no
writes; a union that adds nothing to an existing record is ``replayed``.
"""
_validate_target(locator, sha256, size, mtime_ns)
_validate_epoch(expected_epoch)
tags = normalize_image_tags(tags)
with self._pair_transaction() as conn:
epoch = self._library_epoch(conn, expected_epoch)
now = utc_now_iso()
file_id = self._annotation_owner(conn, locator, now)
byte_id, revision, created, refingerprinted = self._annotation_byte(
conn, file_id, sha256, size, mtime_ns, now
)
current = self._byte_tags(conn, byte_id)
present = {name.casefold() for name in current}
added = [name for name in tags if name.casefold() not in present]
merged = current + added
if len(merged) > MAX_IMAGE_TAGS:
raise AnnotationCommandError(
"tag-limit",
epoch,
{"sha256": sha256, "tags": current, "tag_revision": revision},
)
if not created and not added:
if refingerprinted:
conn.execute(_BUMP_INDEX)
return {**_receipt(epoch, sha256, current, revision, True), "added": 0}
revision = self._write_image_tags(conn, byte_id, merged)
return {
**_receipt(epoch, sha256, merged, revision, False),
"added": len(added),
}
def refresh_image_fingerprint(
self, locator: str, sha256: str, size: int, mtime_ns: int
) -> bool:
"""Record that these exact bytes were just seen with this stat.
True when a byte record was refreshed; nothing is created and no tag
or tag revision changes. The image index revision moves because the
record's currency for cheap reads may have changed.
"""
_validate_target(locator, sha256, size, mtime_ns)
with self.model.get_connection() as conn:
cursor = conn.execute(
_REFRESH_BY_LOCATOR, (size, mtime_ns, utc_now_iso(), sha256, locator)
)
if cursor.rowcount > 0:
conn.execute(_BUMP_INDEX)
return cursor.rowcount > 0
def current_image_tags(self, stats: dict) -> dict:
"""Image tags of the files whose current ``(size, mtime_ns)`` is verified.
``stats`` maps locators to their present fingerprint. A locator is
answered (possibly with an empty set) only when a byte record carries
exactly that verified fingerprint; the most recently verified record
wins when several do. Lookups go in chunks of 500 locators.
"""
if not isinstance(stats, dict):
raise ValueError(
"stats must map locators to (size, mtime_ns) integer pairs"
)
for locator, fingerprint in stats.items():
_validate_fingerprint(locator, fingerprint)
locators = list(stats)
chosen: dict[str, tuple[tuple[str, str], str]] = {}
result: dict[str, list[str]] = {}
with self._annotation_read() as conn:
for start in range(0, len(locators), _LOOKUP_CHUNK):
chunk = locators[start : start + _LOOKUP_CHUNK]
marks = ",".join("?" * len(chunk))
for locator, byte_id, size, mtime_ns, verified_at in conn.execute(
_FINGERPRINTS.format(marks=marks), chunk
):
if (size, mtime_ns) != stats[locator]:
continue
key = (verified_at, byte_id)
if locator not in chosen or key > chosen[locator][0]:
chosen[locator] = (key, byte_id)
owners = {byte_id: locator for locator, (_key, byte_id) in chosen.items()}
result = {locator: [] for locator in chosen}
byte_ids = list(owners)
for start in range(0, len(byte_ids), _LOOKUP_CHUNK):
chunk = byte_ids[start : start + _LOOKUP_CHUNK]
marks = ",".join("?" * len(chunk))
for byte_id, name in conn.execute(_TAGS_OF.format(marks=marks), chunk):
result[owners[byte_id]].append(name)
return result
def iter_image_tag_records(self) -> Iterator[tuple[str, int, int, list[str]]]:
"""Stream every byte record as ``(locator, size, mtime_ns, tags)``.
One SELECT ordered by locator, then by the bytes' first sighting, with
each set's names in position order.
"""
conn = self.model.get_connection()
current: tuple | None = None
for locator, byte_id, size, mtime_ns, name in conn.execute(_RECORDS):
if current is None or current[0] != byte_id:
if current is not None:
yield current[1]
current = (byte_id, (locator, size, mtime_ns, []))
if name is not None:
current[1][3].append(name)
if current is not None:
yield current[1]
database/models.py
+10 -3
View File
@@ -17,6 +17,7 @@ from typing import Optional, Set, Tuple
from .library_schema import library_schema_error, migrate_library_state
from .image_favorites import image_favorites_schema_error, migrate_image_favorites
from .annotation_schema import annotation_schema_error, migrate_annotation_state
from .saved_views import migrate_saved_views, saved_views_schema_error
from .execution_history_schema import (
history_schema_error,
@@ -302,6 +303,7 @@ class PromptModel:
migrate_library_state(conn)
migrate_image_favorites(conn)
migrate_saved_views(conn)
migrate_annotation_state(conn)
def _migrate_prompt_pairs(self, conn: sqlite3.Connection) -> None:
conn.execute("SAVEPOINT prompt_pair_schema")
@@ -1021,9 +1023,9 @@ class PromptModel:
Args:
path: File to inspect
require_library_state: Also require the tables every current
database has (Library state, image favorites and saved
views), as a migrated restore stage must; old backups may
lack them.
database has (Library state, image favorites, saved views
and image annotations), as a migrated restore stage must;
old backups may lack them.
Returns:
(ok, reason): reason is "ok" when the file is acceptable, otherwise
@@ -1088,6 +1090,11 @@ class PromptModel:
)
if views_error:
return False, views_error
annotation_error = annotation_schema_error(
conn, allow_missing=not require_library_state
)
if annotation_error:
return False, annotation_error
return True, "ok"
except sqlite3.Error as e:
return False, f"Invalid SQLite database: {e}"
database/operations.py
+36 -6
View File
@@ -12,6 +12,7 @@ from .models import PromptModel, normalize_image_path, utc_now_iso
from .pairs import PairDependencyError, PairOperations
from .library import LibraryOperations, titled
from .image_favorites import ImageFavoriteOperations
from .annotations import AnnotationOperations
from .saved_views import SavedViewOperations
from .library_commands import LibraryCommandOperations
from .vocabulary import VocabularyOperations
@@ -322,6 +323,7 @@ class PromptDatabase(
ExecutionOutputOperations,
HistoryReadOperations,
ImageFavoriteOperations,
AnnotationOperations,
SavedViewOperations,
):
"""Database operations class for managing prompts."""
@@ -1084,16 +1086,39 @@ class PromptDatabase(
}
@staticmethod
def _bump_image_tag_bytes(conn: sqlite3.Connection, tag_id: int) -> int:
"""Move the tag revision of every image set that carries *tag_id*.
Vocabulary rename, merge and delete change what an image's set reads
as, so a Tag dialog opened before the change must see ``tag-conflict``
on Save instead of recreating the old name. Bumps the image index
revision once when any set was affected; returns that count.
"""
cursor = conn.execute(
"UPDATE media_annotation_bytes SET tag_revision = tag_revision + 1 "
"WHERE id IN (SELECT byte_id FROM media_image_tags WHERE tag_id = ?)",
(tag_id,),
)
if cursor.rowcount > 0:
conn.execute(
"UPDATE media_annotation_metadata SET "
"image_index_revision = image_index_revision + 1 WHERE singleton_id = 1"
)
return cursor.rowcount
@classmethod
def _move_tag_links(
conn: sqlite3.Connection, source_id: int, target_id: int
cls, conn: sqlite3.Connection, source_id: int, target_id: int
) -> None:
"""Repoint component and pair links from source to target.
"""Repoint component, pair and image links from source to target.
Owners that already carry the target keep that one link; the source's
leftover duplicates are removed. The source registry row is left for the
caller to delete.
leftover duplicates are removed (an image set keeps the target's
position). The source registry row is left for the caller to delete.
Image sets that carried the source get a new tag revision first.
"""
for junction in ("prompt_tags", "prompt_pair_tags"):
cls._bump_image_tag_bytes(conn, source_id)
for junction in ("prompt_tags", "prompt_pair_tags", "media_image_tags"):
conn.execute(
f"UPDATE OR IGNORE {junction} SET tag_id = ? WHERE tag_id = ?",
(target_id, source_id),
@@ -1146,6 +1171,7 @@ class PromptDatabase(
conn.execute("DELETE FROM tags WHERE id = ?", (old_tag_id,))
else:
# Simple rename
self._bump_image_tag_bytes(conn, old_tag_id)
conn.execute(
"UPDATE tags SET name = ? WHERE id = ?", (new_name, old_tag_id)
)
@@ -1187,6 +1213,7 @@ class PromptDatabase(
affected = cursor.fetchone()["c"]
# CASCADE delete handles prompt_tags entries
self._bump_image_tag_bytes(conn, tag_id)
conn.execute("DELETE FROM tags WHERE id = ?", (tag_id,))
conn.commit()
@@ -1239,7 +1266,10 @@ class PromptDatabase(
pair_count = conn.execute(
"SELECT COUNT(*) FROM prompt_pair_tags WHERE tag_id = ?", (src_id,)
).fetchone()[0]
if src_count or pair_count:
image_count = conn.execute(
"SELECT COUNT(*) FROM media_image_tags WHERE tag_id = ?", (src_id,)
).fetchone()[0]
if src_count or pair_count or image_count:
self._move_tag_links(conn, src_id, target_id)
tags_merged += 1
affected += src_count
database/vocabulary.py
+30 -4
View File
@@ -3,6 +3,9 @@
A count is the number of eligible Library parents that carry the value
directly or through one active saved pair, which is exactly what
``list_library(tags=[name])`` / ``list_library(category=value)`` return.
A tag row's ``image_count`` is the number of files (private locators) with any
byte record carrying the tag (#249), so a file counts once however many of its
byte versions carry it, and equal bytes at two paths count twice.
Tag commands run in one ``_pair_transaction``: the storage epoch is checked
first, then the exact ``{id, name}`` identity, and the receipt is read on the
@@ -26,6 +29,11 @@ except ImportError:
from utils.validators import validate_category, validate_tags
_CATEGORY_DOMAIN = "prompt-manager-vocabulary-category:v1\n"
# Files carrying a tag through any of their byte records (#249 image tags).
_IMAGE_FILES = (
"SELECT mit.tag_id,COUNT(DISTINCT b.file_id) FROM media_image_tags mit "
"JOIN media_annotation_bytes b ON b.id=mit.byte_id"
)
def _tag_rows(page: str) -> str:
@@ -40,9 +48,13 @@ def _tag_rows(page: str) -> str:
"WHERE pt.tag_id IN (SELECT id FROM page) AND " + _ELIGIBLE + " "
"UNION SELECT ppt.tag_id,pp.positive_prompt_id FROM prompt_pairs pp "
"JOIN prompt_pair_tags ppt ON ppt.pair_id=pp.id "
"WHERE pp.archived=0 AND ppt.tag_id IN (SELECT id FROM page)) "
"SELECT page.id,page.name,COUNT(DISTINCT owners.pid) AS prompt_count "
"WHERE pp.archived=0 AND ppt.tag_id IN (SELECT id FROM page)),"
"images(tag_id,n) AS (" + _IMAGE_FILES + " "
"WHERE mit.tag_id IN (SELECT id FROM page) GROUP BY mit.tag_id) "
"SELECT page.id,page.name,COUNT(DISTINCT owners.pid) AS prompt_count,"
"COALESCE(MAX(images.n),0) AS image_count "
"FROM page LEFT JOIN owners ON owners.tag_id=page.id "
"LEFT JOIN images ON images.tag_id=page.id "
"GROUP BY page.id ORDER BY page.id"
)
@@ -61,7 +73,10 @@ _TAG_IMPACT = (
"AS negative_variants,"
"(SELECT COUNT(*) FROM prompt_pairs pp JOIN prompt_pair_tags ppt "
"ON ppt.pair_id=pp.id WHERE ppt.tag_id=:tag AND pp.archived!=0) "
"AS archived_negative_variants"
"AS archived_negative_variants,"
"(SELECT COUNT(DISTINCT b.file_id) FROM media_image_tags mit "
"JOIN media_annotation_bytes b ON b.id=mit.byte_id WHERE mit.tag_id=:tag) "
"AS images"
)
_MAX_ID = 2**63 - 1
@@ -205,7 +220,16 @@ class VocabularyOperations:
"SELECT COUNT(DISTINCT pair_id) FROM prompt_pair_tags WHERE tag_id=?",
(tag_id,),
).fetchone()[0]
return {"components_changed": components, "pairs_changed": pairs}
images = conn.execute(
"SELECT COUNT(DISTINCT b.file_id) FROM media_image_tags mit "
"JOIN media_annotation_bytes b ON b.id=mit.byte_id WHERE mit.tag_id=?",
(tag_id,),
).fetchone()[0]
return {
"components_changed": components,
"pairs_changed": pairs,
"images_changed": images,
}
def create_vocabulary_tag(self, name: str, *, expected_epoch: str) -> dict:
"""Insert an unused registry row, or reuse the exact existing name."""
@@ -250,6 +274,7 @@ class VocabularyOperations:
"SELECT id FROM tags WHERE name=?", (new_name,)
).fetchone()
if target is None:
self._bump_image_tag_bytes(conn, tag_id)
conn.execute("UPDATE tags SET name=? WHERE id=?", (new_name, tag_id))
result_id = tag_id
else:
@@ -275,6 +300,7 @@ class VocabularyOperations:
epoch = self._library_epoch(conn, expected_epoch)
self._vocabulary_identity(conn, epoch, tag_id, name)
counts = self._tag_owner_counts(conn, tag_id)
self._bump_image_tag_bytes(conn, tag_id)
conn.execute("DELETE FROM tags WHERE id=?", (tag_id,))
return {"library_epoch": epoch, "deleted": True, **counts}