olivv-cs-ComfyUI-FunPack/core/projects.py

"""Projects: an ordered list of scenes, one JSON file each.

A scene owns its prompt text and the result it last produced. Nothing else about
a scene lives here. v4's Scene accumulated three dozen fields over its life --
transitions, effects, audio gain, per-scene frames, guide sources -- and each of
those belongs to the feature that needs it, added when that feature lands, not to
the model every part of the app has to load.

Per-scene length and fps are deliberately absent: regeneration uses the PROJECT's
values, never a scene's own, so a scene cropped on the timeline and regenerated
comes back whole. A crop is a timeline decision; a regenerate is a new scene.

Synchronous, one file per project, stdlib json. Projects are small.
"""

from __future__ import annotations

import json
import re
import time
import uuid
from dataclasses import asdict, dataclass, field

from . import config, media

#: A generated id, never a user-supplied name, is what reaches the filesystem.
_ID = re.compile(r"\A[0-9a-f]{12}\Z")

MAX_NAME = 120


def _new_id() -> str:
    return uuid.uuid4().hex[:12]


def is_id(value) -> bool:
    """True for something this module generated. Everything from a request is
    checked against this before it is used to build a path."""
    return isinstance(value, str) and bool(_ID.match(value))


def _clean_name(raw, fallback="Untitled") -> str:
    name = (raw if isinstance(raw, str) else "").strip()
    return name[:MAX_NAME] or fallback


MAX_SETTING = 16384


def _whole(value) -> int | None:
    """A whole positive number, or nothing. Everything read back out of a
    project file goes through here: the file outlives the code that wrote it."""
    if isinstance(value, bool) or value is None:
        return None
    try:
        number = int(value)
    except (TypeError, ValueError):
        return None
    return number if 1 <= number <= MAX_SETTING else None


#: What a person can say about a result: it was good, or it was not.
#:
#: Binary because that is all anything downstream reads. dev proved it on H3 --
#: the steering only ever takes the SIGN of a rating, never its size, so a
#: ten-point scale was a UI claiming a precision nothing used, and at the sample
#: sizes this runs at a mediocre middle rating only diluted the direction.
#:
#: Core keeps the word and nothing else. What either of them MEANS belongs to
#: whatever learns from them, which is not here.
RATINGS = ("liked", "disliked")


@dataclass
class Scene:
    id: str = field(default_factory=_new_id)
    text: str = ""
    #: The asset this scene last produced, so a reload shows the timeline the
    #: user left rather than an empty one they have to regenerate.
    result: str | None = None
    #: How long this clip RUNS on the timeline -- a crop, made here. It is not
    #: what a regenerate uses: that reads the project, because the crop was a
    #: timeline decision and a regenerate is a new scene.
    length: int | None = None
    rating: str | None = None
    #: A media library id, picked to size this scene's generation -- not
    #: included in it. v4's "drop a picture on the timeline" only ever fed a
    #: resolution/aspect-ratio node; the picture's own pixels went nowhere. Per
    #: scene, not per project: different shots can want a different canvas.
    source_image: str | None = None
    #: Media library ids, in the order they were added -- reference images for
    #: whatever the pipeline's own reference input wants them for. Per scene
    #: for the same reason source_image is: a reference is about THIS shot.
    references: list[str] = field(default_factory=list)

    @staticmethod
    def from_dict(d) -> "Scene":
        d = d if isinstance(d, dict) else {}
        sid = d.get("id")
        result = d.get("result")
        rating = d.get("rating")
        source_image = d.get("source_image")
        raw_refs = d.get("references")
        return Scene(
            id=sid if is_id(sid) else _new_id(),
            text=d.get("text") if isinstance(d.get("text"), str) else "",
            result=result if isinstance(result, str) else None,
            length=_whole(d.get("length")),
            rating=rating if rating in RATINGS else None,
            source_image=source_image if media.is_id(source_image) else None,
            references=[r for r in raw_refs if media.is_id(r)] if isinstance(raw_refs, list) else [],
        )


def _clean_video(raw) -> dict:
    """What the app is holding for the pipeline's `project.video` inputs.

    Core does not know what a video setting IS. Which of them exist is the
    pipeline's business -- it says so with a role -- and naming width and height
    here would be core naming an implementation. What it knows is that a project
    file is the one thing in this app that outlives the code that wrote it, so
    everything read back out of one is checked: a whole positive number, or gone.
    """
    if not isinstance(raw, dict):
        return {}
    clean = {}
    for key, value in raw.items():
        # `True` is an int in Python and would land in a width.
        if not isinstance(key, str) or isinstance(value, bool):
            continue
        try:
            number = int(value)
        except (TypeError, ValueError):
            continue
        if 1 <= number <= MAX_SETTING:
            clean[key] = number
    return clean


def _clean_variables(raw) -> list:
    """[{"name": str, "value": str}, ...] -- anything else in a slot is
    dropped rather than guessed at, the same rule every other field reads
    back out of a project file follows. A name that is empty after stripping
    the leading `$` is not a variable anyone could ever reference, so it is
    dropped rather than kept as a row nothing can ever match."""
    if not isinstance(raw, list):
        return []
    out = []
    for item in raw:
        if not isinstance(item, dict):
            continue
        name = str(item.get("name") or "").lstrip("$").strip()
        if not name:
            continue
        value = item.get("value")
        out.append({"name": name[:MAX_NAME], "value": value if isinstance(value, str) else ""})
    return out


@dataclass
class Project:
    id: str = field(default_factory=_new_id)
    name: str = "Untitled"
    scenes: list[Scene] = field(default_factory=list)
    #: Settings the whole project is generated at -- size, length -- rather than
    #: any one scene. A scene cropped on the timeline and regenerated comes back
    #: at the project's length: the crop was a timeline decision and a regenerate
    #: is a new scene.
    video: dict = field(default_factory=dict)
    #: A negative prompt, sent at whatever "project.negative" input the
    #: pipeline declares. Project-level for the same reason `video` is:
    #: something set rarely, once, not re-typed per scene.
    negative: str = ""
    #: Text prepended to every scene's prompt. v4's "anchor" -- the shared
    #: subject/setting line a whole project holds once instead of retyping
    #: per scene.
    anchor: str = ""
    #: Text appended to every scene's prompt. v4's "postfix" -- symmetric to
    #: anchor, but with its own on/off switch: v4 proved that a postfix
    #: someone spent time writing needs to be turned off for a quick test
    #: without losing it, which clearing the field would do. Anchor never
    #: grew the same need in v4 and does not get one here either.
    postfix: str = ""
    postfix_enabled: bool = True
    #: ponytail: three flat strings (plus one bool), not a dict keyed by role
    #: name -- an earlier version of this comment said to generalise once a
    #: second project-level text role showed up, and now three have. Tried
    #: it: postfix's own enabled flag does not fit a plain role->string map
    #: any better than a fourth field would, so the "simpler" shape turns out
    #: not to be. Three named fields stay more readable than an abstraction
    #: that has to special-case one of its own four entries anyway.
    #:
    #: $name -> text, substituted into anchor/scene/postfix at generation
    #: (core/prompt_build.py). A LIST, not a dict: order is what a person set,
    #: and preserving it is what makes "the row I just added" stay at the
    #: bottom instead of jumping around alphabetically on the next save.
    variables: list = field(default_factory=list)
    updated_at: float = 0.0

    @staticmethod
    def from_dict(d) -> "Project":
        d = d if isinstance(d, dict) else {}
        pid = d.get("id")
        raw = d.get("scenes")
        negative = d.get("negative")
        anchor = d.get("anchor")
        postfix = d.get("postfix")
        postfix_enabled = d.get("postfix_enabled")
        return Project(
            id=pid if is_id(pid) else _new_id(),
            name=_clean_name(d.get("name")),
            scenes=[Scene.from_dict(s) for s in (raw if isinstance(raw, list) else [])],
            video=_clean_video(d.get("video")),
            negative=negative if isinstance(negative, str) else "",
            anchor=anchor if isinstance(anchor, str) else "",
            postfix=postfix if isinstance(postfix, str) else "",
            # True unless the file explicitly says False -- absent (a project
            # from before this field existed) must read as "on", the same as
            # v4's own default, not as "off" because nothing was there to say.
            postfix_enabled=postfix_enabled if isinstance(postfix_enabled, bool) else True,
            variables=_clean_variables(d.get("variables")),
            updated_at=float(d.get("updated_at") or 0.0),
        )

    def to_dict(self) -> dict:
        return asdict(self)


def _dir():
    config.PROJECTS_DIR.mkdir(parents=True, exist_ok=True)
    return config.PROJECTS_DIR


def _path(project_id: str):
    # is_id, not a sanitiser: a name that merely survives cleaning can still be
    # "..", and a store that repairs a bad id quietly is one that writes
    # somewhere nobody asked for.
    if not is_id(project_id):
        raise ValueError(f"not a project id: {project_id!r}")
    return _dir() / f"{project_id}.json"


def listing() -> list[dict]:
    """Every project, newest first, as the little the picker needs."""
    out = []
    for p in _dir().glob("*.json"):
        if not is_id(p.stem):
            continue
        try:
            d = json.loads(p.read_text(encoding="utf-8"))
        except (json.JSONDecodeError, OSError, UnicodeDecodeError):
            continue  # an unreadable file is not a reason to have no list
        if not isinstance(d, dict):
            continue
        out.append({
            "id": p.stem,
            "name": _clean_name(d.get("name"), p.stem),
            "scenes": len(d.get("scenes") or []),
            "updated_at": float(d.get("updated_at") or 0.0),
        })
    out.sort(key=lambda x: x["updated_at"], reverse=True)
    return out


def get(project_id: str) -> Project | None:
    try:
        path = _path(project_id)
    except ValueError:
        return None
    if not path.exists():
        return None
    try:
        return Project.from_dict(json.loads(path.read_text(encoding="utf-8")))
    except (json.JSONDecodeError, OSError, UnicodeDecodeError):
        return None


def save(project: Project) -> Project:
    project.updated_at = time.time()
    path = _path(project.id)
    tmp = path.with_suffix(".tmp")
    tmp.write_text(json.dumps(project.to_dict(), indent=2), encoding="utf-8")
    tmp.replace(path)  # a concurrent read never sees a half-written project
    return project


def create(name=None) -> Project:
    """A new project with one empty scene: a timeline with nothing on it cannot
    be typed into, so a brand-new project would need an Add before it could be
    used at all."""
    return save(Project(name=_clean_name(name), scenes=[Scene()]))


def delete(project_id: str) -> bool:
    try:
        path = _path(project_id)
    except ValueError:
        return False
    if not path.exists():
        return False
    path.unlink()
    return True