setiastrosuitepro 1.6.0__py3-none-any.whl

setiastro/saspro/doc_manager.py

@@ -0,0 +1,2634 @@
+# pro/doc_manager.py
+from __future__ import annotations
+from PyQt6.QtCore import QObject, pyqtSignal, Qt, QTimer
+from PyQt6.QtWidgets import QApplication, QMessageBox
+import os
+import numpy as np
+from setiastro.saspro.xisf import XISF as XISFReader
+from astropy.io import fits  # local import; optional dep
+from setiastro.saspro.legacy.image_manager import load_image as legacy_load_image, save_image as legacy_save_image
+from setiastro.saspro.legacy.image_manager import list_fits_extensions, load_fits_extension
+import uuid
+from setiastro.saspro.legacy.image_manager import attach_wcs_to_metadata  # or wherever you put it
+from astropy.wcs import WCS  # only if not already imported in this module
+from setiastro.saspro.debug_utils import debug_dump_metadata
+
+# Memory utilities for lazy loading and caching
+try:
+    from setiastro.saspro.memory_utils import get_thumbnail_cache, LazyImage
+except ImportError:
+    get_thumbnail_cache = None
+    LazyImage = None
+
+from setiastro.saspro.swap_manager import get_swap_manager
+from setiastro.saspro.widgets.image_utils import ensure_contiguous
+from typing import Any
+
+# --- WCS DEBUGGING ------------------------------------------------------
+_DEBUG_WCS = False  # flip to False when you’re done debugging
+
+def _debug_log_wcs_context(context: str, meta_or_hdr):
+    """
+    Tiny helper to print key WCS bits:
+      - NAXIS1/2
+      - CRPIX1/2
+      - CRVAL1/2
+      - CDELT / CD if present
+    Works if you pass either a metadata dict or a FITS-like header dict.
+    """
+    if not _DEBUG_WCS:
+        return
+
+    # Try to resolve a header from a metadata dict
+    hdr = None
+    if isinstance(meta_or_hdr, dict):
+        # metadata dict with possible header keys
+        hdr = (meta_or_hdr.get("original_header")
+               or meta_or_hdr.get("fits_header")
+               or meta_or_hdr.get("header"))
+        if hdr is None:
+            # maybe you passed the header dict directly
+            hdr = meta_or_hdr
+    else:
+        hdr = meta_or_hdr
+
+    if hdr is None:
+        print(f"[WCS DEBUG] {context}: no header found")
+        return
+
+    # Normalize dict-like header
+    if hasattr(hdr, "keys"):  # astropy Header or dict
+        try:
+            keys = list(hdr.keys())
+        except Exception:
+            keys = []
+    else:
+        print(f"[WCS DEBUG] {context}: header is non-mapping type {type(hdr)}")
+        return
+
+    def _get(k, default=None):
+        try:
+            return hdr.get(k, default)
+        except Exception:
+            try:
+                return hdr[k]
+            except Exception:
+                return default
+
+    naxis1 = _get("NAXIS1")
+    naxis2 = _get("NAXIS2")
+    crpix1 = _get("CRPIX1")
+    crpix2 = _get("CRPIX2")
+    crval1 = _get("CRVAL1")
+    crval2 = _get("CRVAL2")
+
+    cd11   = _get("CD1_1")
+    cd12   = _get("CD1_2")
+    cd21   = _get("CD2_1")
+    cd22   = _get("CD2_2")
+    cdelt1 = _get("CDELT1")
+    cdelt2 = _get("CDELT2")
+
+    print(f"[WCS DEBUG] {context}:")
+    print(f"  NAXIS1={naxis1}  NAXIS2={naxis2}")
+    print(f"  CRPIX1={crpix1}  CRPIX2={crpix2}")
+    print(f"  CRVAL1={crval1}  CRVAL2={crval2}")
+    if any(v is not None for v in (cd11, cd12, cd21, cd22)):
+        print(f"  CD = [[{cd11}, {cd12}], [{cd21}, {cd22}]]")
+    if cdelt1 is not None or cdelt2 is not None:
+        print(f"  CDELT1={cdelt1}  CDELT2={cdelt2}")
+    print("")
+
+_DEBUG_UNDO = False  # set True while chasing the GraXpert crash
+
+
+def _debug_log_undo(context: str, **info):
+    """
+    Lightweight logger for undo/redo/update activity.
+    Safe: never raises, even if repr() is weird.
+    """
+    if not _DEBUG_UNDO:
+        return
+    try:
+        bits = []
+        for k, v in info.items():
+            try:
+                s = str(v)
+            except Exception:
+                try:
+                    s = repr(v)
+                except Exception:
+                    s = f"<unrepr {type(v)}>"
+            bits.append(f"{k}={s}")
+        print(f"[UNDO DEBUG] {context}: " + ", ".join(bits))
+    except Exception as e:
+        # Last-resort safety – don't let logging itself kill us
+        try:
+            print(f"[UNDO DEBUG] {context}: <logging failed: {e}>")
+        except Exception:
+            pass
+
+from setiastro.saspro.file_utils import _normalize_ext
+
+def _normalize_image_01(arr: np.ndarray) -> np.ndarray:
+    """
+    Normalize an image to [0,1] in-place-ish:
+
+      1. If min < 0  → shift so min becomes 0.
+      2. Then if max > 1 → divide by max.
+
+    NaNs/Infs are ignored when computing min/max.
+    Returns float32 array.
+    """
+    if arr is None:
+        return arr
+
+    a = np.asarray(arr, dtype=np.float32)
+    finite = np.isfinite(a)
+    if not finite.any():
+        # completely bogus; give back zeros
+        return np.zeros_like(a, dtype=np.float32)
+
+    # Step 1: shift up if we have negatives
+    min_val = a[finite].min()
+    if min_val < 0.0:
+        a = a - min_val
+        finite = np.isfinite(a)
+
+    # Step 2: scale down if we exceed 1
+    max_val = a[finite].max()
+    if max_val > 1.0 and max_val > 0.0:
+        a = a / max_val
+
+    return a
+
+_ALLOWED_DEPTHS = {
+    "png":  {"8-bit"},
+    "jpg":  {"8-bit"},
+    "fits": ["8-bit", "16-bit", "32-bit unsigned", "32-bit floating point"],
+    "fit":  ["8-bit", "16-bit", "32-bit unsigned", "32-bit floating point"],
+    "tif":  {"8-bit", "16-bit", "32-bit unsigned", "32-bit floating point"},
+    "xisf": {"16-bit", "32-bit unsigned", "32-bit floating point"},
+}
+
+class TableDocument(QObject):
+    changed = pyqtSignal()
+
+    def __init__(self, rows: list[list], headers: list[str], metadata: dict | None = None, parent=None):
+        super().__init__(parent)
+        self.rows = rows              # list of list (2D) for QAbstractTableModel
+        self.headers = headers        # list of column names
+        self.metadata = dict(metadata or {})
+        self._undo = []
+        self._redo = []
+
+    def display_name(self) -> str:
+        dn = self.metadata.get("display_name")
+        if dn:
+            return dn
+        p = self.metadata.get("file_path")
+        return os.path.basename(p) if p else "Untitled Table"
+
+    def can_undo(self) -> bool: return False
+    def can_redo(self) -> bool: return False
+    def last_undo_name(self) -> str | None: return None
+    def last_redo_name(self) -> str | None: return None
+    def undo(self) -> str | None: return None
+    def redo(self) -> str | None: return None
+
+class ImageDocument(QObject):
+    changed = pyqtSignal()
+
+    def __init__(self, image: np.ndarray, metadata: dict | None = None, parent=None):
+        super().__init__(parent)
+        self.image = image
+        self.metadata = dict(metadata or {})
+        self.mask = None
+        # _undo / _redo now store tuples: (swap_id: str, metadata: dict, step_name: str)
+        self._undo: list[tuple[str, dict, str]] = []
+        self._redo: list[tuple[str, dict, str]] = []        
+        self.masks: dict[str, np.ndarray] = {}
+        self.active_mask_id: str | None = None
+        self.uid = uuid.uuid4().hex  # stable identity for DnD, layers, masks, etc.
+
+        # NEW: operation log — list of simple dicts
+        # Each entry: {
+        #   "id": str,
+        #   "step": str,
+        #   "params": dict,
+        #   "roi": (x,y,w,h) | None,
+        #   "source": "full" | "roi",
+        #   "ts": float
+        # }
+        self._op_log: list[dict] = []
+        
+        # Track unsaved changes explicitly
+        self.dirty: bool = False
+        
+        # Copy-on-write support: if this document shares image data with another,
+        # _cow_source holds reference to the source. On first write (apply_edit),
+        # we copy the image data and clear _cow_source.
+        self._cow_source: 'ImageDocument | None' = None
+    # --- history helpers (NEW) ---
+    # --- operation log helpers (NEW) -----------------------------------
+    def record_operation(
+        self,
+        step_name: str,
+        params: dict | None = None,
+        roi: tuple[int, int, int, int] | None = None,
+        source: str = "full",
+    ) -> str:
+        """
+        Append a param-record for this edit. This is *lightweight* metadata
+        used for replaying ROI recipes etc; it does NOT affect undo/redo.
+        """
+        import time as _time
+        op_id = uuid.uuid4().hex
+        entry = {
+            "id": op_id,
+            "step": step_name or "Edit",
+            "params": _dm_json_sanitize(params or {}),
+            "roi": tuple(roi) if roi else None,
+            "source": str(source or "full"),
+            "ts": float(_time.time()),
+        }
+        self._op_log.append(entry)
+        return op_id
+
+    def get_operation_log(self) -> list[dict]:
+        """Return a copy of the operation log (for UI / replay)."""
+        return list(self._op_log)
+
+    def clear_operation_log(self):
+        """Clear the operation log (does not touch pixel history)."""
+        self._op_log.clear()
+
+
+    def can_undo(self) -> bool:
+        return bool(self._undo)
+
+    def can_redo(self) -> bool:
+        return bool(self._redo)
+
+    def last_undo_name(self) -> str | None:
+        return self._undo[-1][2] if self._undo else None
+
+    def last_redo_name(self) -> str | None:
+        return self._redo[-1][2] if self._redo else None
+
+
+    def add_mask(self, mask: Any, mask_id: str | None = None, make_active: bool = True) -> str:
+        """
+        Store a mask on this document.
+
+        - `mask` can be a numpy array or any mask-like object.
+        - If `mask_id` is None, a random UUID is generated.
+        - Returns the mask_id used.
+        """
+        if mask_id is None:
+            mask_id = getattr(mask, "id", None) or uuid.uuid4().hex
+
+        # If it's an array, normalize to float32; otherwise just store as-is.
+        try:
+            arr = np.asarray(mask, dtype=np.float32)
+            self.masks[mask_id] = arr
+        except Exception:
+            self.masks[mask_id] = mask
+
+        if make_active:
+            self.active_mask_id = mask_id
+
+        return mask_id
+
+    def remove_mask(self, mask_id: str):
+        self.masks.pop(mask_id, None)
+        if self.active_mask_id == mask_id:
+            self.active_mask_id = None
+
+    def get_active_mask(self):
+        return self.masks.get(self.active_mask_id) if self.active_mask_id else None
+
+    def close(self):
+        """
+        Explicit cleanup of swap files.
+        """
+        sm = get_swap_manager()
+        # Clean up undo stack
+        for swap_id, _, _ in self._undo:
+            sm.delete_state(swap_id)
+        self._undo.clear()
+        
+        # Clean up redo stack
+        for swap_id, _, _ in self._redo:
+            sm.delete_state(swap_id)
+        self._redo.clear()
+
+    def __del__(self):
+        # Fallback cleanup if close() wasn't called (though explicit close is better)
+        try:
+            self.close()
+        except Exception:
+            pass
+
+
+    # in class ImageDocument
+    def apply_edit(self, new_image: np.ndarray, metadata: dict | None = None, step_name: str = "Edit"):
+        """
+        Smart edit:
+        - If this is an ROI view (has _roi_info), paste back into parent and emit region update.
+        - Else: push history on self and emit full-image update.
+        - IMPORTANT: merge metadata without nuking FITS/WCS headers.
+        """
+        import numpy as np
+
+        def _merge_meta(old_meta: dict | None, new_meta: dict | None, step_name: str):
+            """
+            Merge new_meta into old_meta but preserve critical header fields
+            unless they are explicitly overridden with non-None values.
+            """
+            old = dict(old_meta or {})
+            incoming = dict(new_meta or {})
+
+            critical_keys = (
+                "original_header",
+                "fits_header",
+                "wcs_header",
+                "file_meta",
+                "image_meta",
+            )
+
+            # Preserve critical keys unless caller *deliberately* overrides
+            for k in critical_keys:
+                if k in incoming:
+                    if incoming[k] is not None:
+                        old[k] = incoming[k]
+                # if not in incoming → leave old value alone
+
+            # Merge all remaining keys normally
+            for k, v in incoming.items():
+                if k in critical_keys:
+                    continue
+                old[k] = v
+
+            if step_name:
+                old.setdefault("step_name", step_name)
+            return old
+
+        # ------ ROI-aware branch (auto-pasteback) ------
+        roi_info = getattr(self, "_roi_info", None)
+        if roi_info:
+            parent = roi_info.get("parent_doc")
+            roi    = roi_info.get("roi")
+            if isinstance(parent, ImageDocument) and (getattr(parent, "image", None) is not None) and roi:
+                x, y, w, h = map(int, roi)
+
+                img = np.asarray(new_image)
+                if img.dtype != np.float32:
+                    img = img.astype(np.float32, copy=False)
+
+                base = np.asarray(parent.image)
+                if img.shape[:2] != (h, w):
+                    raise ValueError(f"Edited preview shape {img.shape[:2]} does not match ROI {(h, w)}")
+
+                # shape reconciliation
+                if base.ndim == 2 and img.ndim == 3 and img.shape[2] == 1:
+                    img = img[..., 0]
+                if base.ndim == 3 and img.ndim == 2:
+                    img = np.repeat(img[..., None], base.shape[2], axis=2)
+
+                new_full = base.copy()
+                new_full[y:y+h, x:x+w] = img
+
+                # push onto the PARENT’s history
+                if metadata:
+                    parent.metadata = _merge_meta(parent.metadata, metadata, step_name)
+                else:
+                    parent.metadata.setdefault("step_name", step_name)
+                
+                sm = get_swap_manager()
+                sid = sm.save_state(parent.image)
+                if sid:
+                    parent._undo.append((sid, parent.metadata.copy(), step_name))
+                
+                for old_sid, _, _ in parent._redo:
+                    sm.delete_state(old_sid)
+                parent._redo.clear()
+                
+                parent.image = new_full
+                parent.dirty = True
+                parent.changed.emit()
+
+                dm = getattr(self, "_doc_manager", None) or getattr(parent, "_doc_manager", None)
+                try:
+                    if dm is not None and hasattr(dm, "imageRegionUpdated"):
+                        dm.imageRegionUpdated.emit(parent, (x, y, w, h))
+                except Exception:
+                    print(f"[DocManager] Failed to emit imageRegionUpdated for ROI.")
+                return  # done
+
+        # ------ Normal (full-image) branch ------
+
+        # Copy-on-write
+        if self._cow_source is not None and self.image is not None:
+            self.image = self.image.copy()
+            self._cow_source = None
+        
+        if self.image is not None:
+            # snapshot current image + metadata for undo
+            try:
+                curr = np.asarray(self.image, dtype=np.float32)
+                curr = ensure_contiguous(curr)
+                
+                sm = get_swap_manager()
+                sid = sm.save_state(curr)
+                
+                _debug_log_undo(
+                    "ImageDocument.apply_edit.snapshot",
+                    doc_id=id(self),
+                    name=getattr(self, "display_name", lambda: "<no-name>")(),
+                    curr_shape=getattr(curr, "shape", None),
+                    undo_len_before=len(self._undo),
+                    redo_len_before=len(self._redo),
+                    step_name=step_name,
+                    swap_id=sid
+                )
+                if sid:
+                    self._undo.append((sid, self.metadata.copy(), step_name))
+            except Exception as e:
+                print(f"[ImageDocument] apply_edit: failed to snapshot current image for undo: {e}")
+            
+            # Clear redo stack and delete files
+            sm = get_swap_manager()
+            for old_sid, _, _ in self._redo:
+                sm.delete_state(old_sid)
+            self._redo.clear()
+
+        # --- header-safe metadata merge ---
+        if metadata:
+            self.metadata = _merge_meta(self.metadata, metadata, step_name)
+        else:
+            self.metadata.setdefault("step_name", step_name)
+
+        # normalize new image
+        img = np.asarray(new_image, dtype=np.float32)
+        if img.size == 0:
+            raise ValueError("apply_edit: new image is empty")
+
+        img = ensure_contiguous(img)
+
+        _debug_log_undo(
+            "ImageDocument.apply_edit.apply",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            new_shape=getattr(img, "shape", None),
+            undo_len_after=len(self._undo),
+            redo_len_after=len(self._redo),
+            step_name=step_name,
+        )
+
+        self.image = img
+        self.dirty = True
+        self.changed.emit()
+
+        dm = getattr(self, "_doc_manager", None)
+        try:
+            if dm is not None and hasattr(dm, "imageRegionUpdated"):
+                dm.imageRegionUpdated.emit(self, None)
+        except Exception:
+            pass
+
+
+
+    def undo(self) -> str | None:
+        # Extra-safe: if stack is empty, bail early.
+        if not self._undo:
+            _debug_log_undo(
+                "ImageDocument.undo.empty_stack",
+                doc_id=id(self),
+                name=getattr(self, "display_name", lambda: "<no-name>")(),
+            )
+            return None
+
+        _debug_log_undo(
+            "ImageDocument.undo.entry",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            undo_len=len(self._undo),
+            redo_len=len(self._redo),
+            top_step=self._undo[-1][2] if self._undo else None,
+        )
+
+        # Pop with an extra guard in case something cleared _undo between
+        # the check above and this call (re-entrancy / threading).
+        try:
+            prev_sid, prev_meta, name = self._undo.pop()
+        except IndexError:
+            _debug_log_undo(
+                "ImageDocument.undo.pop_index_error",
+                doc_id=id(self),
+                name=getattr(self, "display_name", lambda: "<no-name>")(),
+                undo_len=len(self._undo),
+                redo_len=len(self._redo),
+            )
+            return None
+
+        # Load previous image from swap
+        sm = get_swap_manager()
+        prev_img = sm.load_state(prev_sid)
+        
+        # We can delete the swap file now that we have it in RAM
+        # (unless we want to keep it for some reason, but standard undo consumes the state)
+        sm.delete_state(prev_sid)
+
+        if prev_img is None:
+             print(f"[ImageDocument] undo: failed to load swap state {prev_sid}")
+             return None
+
+        # Normalize previous image before using it
+        try:
+            prev_arr = np.asarray(prev_img, dtype=np.float32)
+            if prev_arr.size == 0:
+                raise ValueError("undo: previous image is empty")
+            prev_arr = np.ascontiguousarray(prev_arr)
+        except Exception as e:
+            print(f"[ImageDocument] undo: invalid prev_img in stack ({type(prev_img)}): {e}")
+            # Put it back so we don't corrupt history further? 
+            # Actually if load failed we are in trouble.
+            return None
+
+        _debug_log_undo(
+            "ImageDocument.undo.normalized_prev",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            prev_shape=getattr(prev_arr, "shape", None),
+            prev_dtype=getattr(prev_arr, "dtype", None),
+            step_name=name,
+            meta_step=prev_meta.get("step_name", None) if isinstance(prev_meta, dict) else None,
+        )
+
+        # Snapshot current state for redo (best-effort)
+        curr_img = self.image
+        curr_meta = self.metadata
+        try:
+            if curr_img is not None:
+                curr_arr = np.asarray(curr_img, dtype=np.float32)
+                curr_arr = np.ascontiguousarray(curr_arr)
+                
+                # Save to swap for Redo
+                sid = sm.save_state(curr_arr)
+                if sid:
+                    self._redo.append((sid, dict(curr_meta), name))
+            else:
+                # Handle None image? Should not happen usually
+                pass
+        except Exception as e:
+            print(f"[ImageDocument] undo: failed to snapshot current image for redo: {e}")
+
+        _debug_log_undo(
+            "ImageDocument.undo.before_apply",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            curr_shape=getattr(curr_img, "shape", None) if curr_img is not None else None,
+            curr_dtype=getattr(curr_img, "dtype", None) if curr_img is not None else None,
+        )
+
+        self.image = prev_arr
+        self.metadata = dict(prev_meta or {})
+        self.dirty = True
+        try:
+            self.changed.emit()
+        except Exception:
+            pass
+
+        _debug_log_undo(
+            "ImageDocument.undo.after_apply",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            new_shape=getattr(self.image, "shape", None),
+            new_dtype=getattr(self.image, "dtype", None),
+            undo_len=len(self._undo),
+            redo_len=len(self._redo),
+        )
+        return name
+
+
+    def redo(self) -> str | None:
+        if not self._redo:
+            return None
+
+        _debug_log_undo(
+            "ImageDocument.redo.entry",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            redo_len=len(self._redo),
+            undo_len=len(self._undo),
+            top_step=self._redo[-1][2] if self._redo else None,
+        )
+
+        nxt_sid, nxt_meta, name = self._redo.pop()
+
+        # Load next image from swap
+        sm = get_swap_manager()
+        nxt_img = sm.load_state(nxt_sid)
+        sm.delete_state(nxt_sid)
+
+        if nxt_img is None:
+            print(f"[ImageDocument] redo: failed to load swap state {nxt_sid}")
+            return None
+
+        # Normalize next image before using it
+        try:
+            nxt_arr = np.asarray(nxt_img, dtype=np.float32)
+            if nxt_arr.size == 0:
+                raise ValueError("redo: next image is empty")
+            nxt_arr = np.ascontiguousarray(nxt_arr)
+        except Exception as e:
+            print(f"[ImageDocument] redo: invalid nxt_img in stack ({type(nxt_img)}): {e}")
+            return None
+            
+        _debug_log_undo(
+            "ImageDocument.redo.normalized_next",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            nxt_shape=getattr(nxt_arr, "shape", None),
+            nxt_dtype=getattr(nxt_arr, "dtype", None),
+            step_name=name,
+            meta_step=nxt_meta.get("step_name", None) if isinstance(nxt_meta, dict) else None,
+        )
+        curr_img = self.image
+        curr_meta = self.metadata
+        try:
+            if curr_img is not None:
+                curr_arr = np.asarray(curr_img, dtype=np.float32)
+                curr_arr = np.ascontiguousarray(curr_arr)
+                
+                # Save current to swap for Undo
+                sid = sm.save_state(curr_arr)
+                if sid:
+                    self._undo.append((sid, dict(curr_meta), name))
+            else:
+                pass
+        except Exception as e:
+            print(f"[ImageDocument] redo: failed to snapshot current image for undo: {e}")
+        _debug_log_undo(
+            "ImageDocument.redo.before_apply",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            curr_shape=getattr(curr_img, "shape", None) if curr_img is not None else None,
+            curr_dtype=getattr(curr_img, "dtype", None) if curr_img is not None else None,
+        )
+        self.image = nxt_arr
+        self.metadata = dict(nxt_meta or {})
+        self.dirty = True
+        try:
+            self.changed.emit()
+        except Exception:
+            pass
+
+        _debug_log_undo(
+            "ImageDocument.redo.after_apply",
+            doc_id=id(self),
+            name=getattr(self, "display_name", lambda: "<no-name>")(),
+            new_shape=getattr(self.image, "shape", None),
+            new_dtype=getattr(self.image, "dtype", None),
+            undo_len=len(self._undo),
+            redo_len=len(self._redo),
+        )
+
+        return name
+
+
+    # existing methods unchanged below...
+    def set_image(self, img: np.ndarray, metadata: dict | None = None, step_name: str = "Edit"):
+        """
+        Treat set_image as an editing operation that records history.
+        (History previews and “Restore from History” call this.)
+        """
+        self.apply_edit(img, metadata or {}, step_name=step_name)
+
+    
+    # --- Add to ImageDocument (public history helpers) -------------------
+
+    def get_undo_stack(self):
+        """
+        Oldest → newest *before* current image.
+        Returns [(swap_id, meta, name), ...]
+        """
+        out = []
+        for sid, meta, name in self._undo:
+            out.append((sid, meta or {}, name or "Unnamed"))
+        return out
+
+    def display_name(self) -> str:
+        # Prefer an explicit display name if set
+        dn = self.metadata.get("display_name")
+        if dn:
+            return dn
+        p = self.metadata.get("file_path")
+        return os.path.basename(p) if p else "Untitled"
+
+
+def _dm_json_sanitize(obj):
+    """Tiny, local JSON sanitizer: keeps size small & avoids numpy/astropy weirdness."""
+    import numpy as _np
+    if isinstance(obj, (str, int, float, bool)) or obj is None:
+        return obj
+    if isinstance(obj, dict):
+        return {str(k): _dm_json_sanitize(v) for k, v in obj.items()}
+    if isinstance(obj, (list, tuple)):
+        return [_dm_json_sanitize(x) for x in obj]
+    # numpy array → small placeholder
+    try:
+        if isinstance(obj, _np.ndarray):
+            return {"__nd__": True, "shape": list(obj.shape), "dtype": str(obj.dtype)}
+        # numpy scalar
+        if hasattr(obj, "item"):
+            return obj.item()
+    except Exception:
+        pass
+    try:
+        return repr(obj)
+    except Exception:
+        return str(type(obj))
+
+def _compute_cropped_wcs(parent_hdr_like: dict | "fits.Header",
+                         x: int, y: int, w: int, h: int):
+    """
+    Returns a plain dict WCS header reflecting a pure pixel crop by (x,y,w,h).
+    Keeps CD/CDELT/PC/CRVAL as-is and shifts CRPIX by (+/-) the crop offset.
+    Also sets NAXIS1/2 to (w,h) and records custom CROPX/CROPY.
+    """
+    try:
+        from astropy.io.fits import Header  # type: ignore
+    except Exception:
+        Header = None  # type: ignore
+
+    # Normalize to a dict of key->value (no comments needed for the drag payload)
+    if Header is not None and isinstance(parent_hdr_like, Header):
+        base = {k: parent_hdr_like.get(k) for k in parent_hdr_like.keys()}
+    elif isinstance(parent_hdr_like, dict):
+        # If it’s an XISF-like dict, try to pull a FITSKeywords block first
+        fk = parent_hdr_like.get("FITSKeywords")
+        if isinstance(fk, dict) and fk:
+            base = {}
+            for k, arr in fk.items():
+                try:
+                    base[k] = (arr or [{}])[0].get("value", None)
+                except Exception:
+                    pass
+        else:
+            base = dict(parent_hdr_like)
+    else:
+        base = {}
+
+    # Shift CRPIX by the crop offset (ROI origin is (x,y) in full-image pixels)
+    crpix1 = base.get("CRPIX1")
+    crpix2 = base.get("CRPIX2")
+    if isinstance(crpix1, (int, float)) and isinstance(crpix2, (int, float)):
+        new_crpix1 = float(crpix1) - float(x)
+        new_crpix2 = float(crpix2) - float(y)
+        base["CRPIX1"] = new_crpix1
+        base["CRPIX2"] = new_crpix2
+    else:
+        new_crpix1 = crpix1
+        new_crpix2 = crpix2
+
+    # Update image size keys
+    base["NAXIS1"] = int(w)
+    base["NAXIS2"] = int(h)
+
+    # Optional helpful tags
+    base["CROPX"] = int(x)
+    base["CROPY"] = int(y)
+    base["SASKIND"] = "ROI-CROP"
+
+    # DEBUG: show how CRPIX changed for this crop
+    if _DEBUG_WCS:
+        print(f"[WCS DEBUG] _compute_cropped_wcs: roi=({x},{y},{w},{h})")
+        print(f"  CRPIX1: {crpix1} -> {new_crpix1}")
+        print(f"  CRPIX2: {crpix2} -> {new_crpix2}")
+        print("")
+
+    return base
+
+import logging
+
+log = logging.getLogger(__name__)
+
+def _pick_header_for_save(meta: dict) -> fits.Header | None:
+    """
+    Choose the best header to write to disk.
+
+    Priority:
+        1. 'wcs_header'      – if you stash a solved header here
+        2. 'fits_header'     – common name after ASTAP / plate solve
+        3. 'original_header' – whatever came from disk
+        4. 'header'          – older code paths
+    """
+    if not isinstance(meta, dict):
+        return None
+
+    for key in ("wcs_header", "fits_header", "original_header", "header"):
+        hdr = meta.get(key)
+        if isinstance(hdr, fits.Header):
+            log.debug("[_pick_header_for_save] using %s (%d cards)", key, len(hdr))
+            return hdr
+
+    log.debug("[_pick_header_for_save] no fits.Header found in metadata; "
+              "will let legacy_save_image fall back.")
+    return None
+
+def _snapshot_header_for_metadata(meta: dict):
+    """
+    If meta contains a header under common keys, add a JSON-safe snapshot at
+    meta["__header_snapshot__"] so viewers/project IO never choke.
+    """
+    if not isinstance(meta, dict):
+        return
+    if "__header_snapshot__" in meta:
+        return
+
+    hdr = (meta.get("original_header")
+           or meta.get("fits_header")
+           or meta.get("header"))
+
+    if hdr is None:
+        return
+
+    snap = None
+
+    # Try astropy Header (without hard dependency)
+    try:
+        from astropy.io.fits import Header  # type: ignore
+    except Exception:
+        Header = None  # type: ignore
+
+    try:
+        if Header is not None and isinstance(hdr, Header):
+            cards = []
+            for k in hdr.keys():
+                try:
+                    val = hdr[k]
+                except Exception:
+                    val = ""
+                try:
+                    cmt = hdr.comments[k] if hasattr(hdr, "comments") else ""
+                except Exception:
+                    cmt = ""
+                cards.append([str(k), _dm_json_sanitize(val), str(cmt)])
+            snap = {"format": "fits-cards", "cards": cards}
+        elif isinstance(hdr, dict):
+            # Already a dict-like header (e.g., XISF style)
+            snap = {"format": "dict",
+                    "items": {str(k): _dm_json_sanitize(v) for k, v in hdr.items()}}
+        else:
+            # Last resort string
+            snap = {"format": "repr", "text": repr(hdr)}
+    except Exception:
+        try:
+            snap = {"format": "repr", "text": str(hdr)}
+        except Exception:
+            snap = None
+
+    if snap:
+        meta["__header_snapshot__"] = snap
+
+def _safe_str(x) -> str:
+    try:
+        return str(x)
+    except Exception:
+        try:
+            return repr(x)
+        except Exception:
+            return "<unrepr>"
+
+def _fits_table_to_csv(hdu, out_csv_path: str, max_rows: int = 250000):
+    """
+    Convert a FITS (Bin)Table HDU to CSV. Returns the CSV path.
+    Limits to max_rows to avoid giant dumps.
+    """
+    try:
+        data = hdu.data
+        if data is None:
+            raise RuntimeError("No table data")
+
+        # Astropy table→numpy recarray is fine; iterate to strings
+        rec = np.asarray(data)
+        nrows = int(rec.shape[0]) if rec.ndim >= 1 else 0
+        if nrows == 0:
+            # write headers only
+            names = [str(n) for n in (getattr(data, "names", None) or [])]
+            with open(out_csv_path, "w", encoding="utf-8", newline="") as f:
+                if names:
+                    f.write(",".join(names) + "\n")
+            return out_csv_path
+
+        # Column names (fallback to numeric if missing)
+        names = list(getattr(data, "names", [])) or [f"C{i+1}" for i in range(rec.shape[1] if rec.ndim == 2 else len(rec.dtype.names or []))]
+
+        import csv
+        with open(out_csv_path, "w", encoding="utf-8", newline="") as f:
+            w = csv.writer(f)
+            w.writerow([_safe_str(n) for n in names])
+
+            # Decide how to iterate rows depending on structured vs 2D numeric
+            if rec.dtype.names:  # structured/record array
+                for ri in range(min(nrows, max_rows)):
+                    row = rec[ri]
+                    w.writerow([_safe_str(row[name]) for name in rec.dtype.names])
+            else:
+                # plain 2D numeric table
+                if rec.ndim == 1:
+                    for ri in range(min(nrows, max_rows)):
+                        w.writerow([_safe_str(rec[ri])])
+                else:
+                    for ri in range(min(nrows, max_rows)):
+                        w.writerow([_safe_str(x) for x in rec[ri]])
+
+        return out_csv_path
+    except Exception as e:
+        raise
+
+def _fits_table_to_rows_headers(hdu, max_rows: int = 500000) -> tuple[list[list], list[str]]:
+    """
+    Convert a FITS (Bin)Table/Table HDU to (rows, headers).
+    Truncates to max_rows for safety.
+    """
+    data = hdu.data
+    if data is None:
+        return [], []
+    rec = np.asarray(data)
+    # Column names
+    names = list(getattr(data, "names", [])) or (
+        list(rec.dtype.names) if rec.dtype.names else [f"C{i+1}" for i in range(rec.shape[1] if rec.ndim == 2 else 1)]
+    )
+    rows = []
+    nrows = int(rec.shape[0]) if rec.ndim >= 1 else 0
+    nrows = min(nrows, max_rows)
+    if rec.dtype.names:  # structured array
+        for ri in range(nrows):
+            row = rec[ri]
+            rows.append([_safe_str(row[name]) for name in rec.dtype.names])
+    else:
+        # numeric 2D/1D table
+        if rec.ndim == 1:
+            for ri in range(nrows):
+                rows.append([_safe_str(rec[ri])])
+        else:
+            for ri in range(nrows):
+                rows.append([_safe_str(x) for x in rec[ri]])
+    return rows, [str(n) for n in names]
+
+
+_shown_raw_preview_paths: set[str] = set()
+_raw_preview_boxes: list[QMessageBox] = []  # prevent GC while shown
+
+def _show_raw_preview_warning_nonmodal(path: str):
+    parent = QApplication.activeWindow()
+    box = QMessageBox(parent)
+    box.setIcon(QMessageBox.Icon.Warning)
+    box.setWindowTitle("RAW preview loaded")
+    box.setText(
+        "Linear RAW decoding failed for:\n"
+        f"{path}\n\n"
+        "Showing the camera’s embedded JPEG preview instead "
+        "(8-bit, non-linear). Some processing tools may be limited."
+    )
+    box.setStandardButtons(QMessageBox.StandardButton.Ok)
+    box.setWindowModality(Qt.WindowModality.NonModal)  # ← fix here
+    box.setAttribute(Qt.WidgetAttribute.WA_DeleteOnClose, True)
+
+    _raw_preview_boxes.append(box)
+    box.finished.connect(lambda _=None, b=box: _raw_preview_boxes.remove(b))
+    box.show()
+
+def maybe_warn_raw_preview(path: str, header):
+    if not header or not bool(header.get("RAW_PREV", False)):
+        return
+    if path in _shown_raw_preview_paths:
+        return
+    _shown_raw_preview_paths.add(path)
+    QTimer.singleShot(0, lambda p=path: _show_raw_preview_warning_nonmodal(p))
+
+_np = np
+
+class _RoiViewDocument(ImageDocument):
+    def __init__(self, parent_doc: ImageDocument, roi: tuple[int,int,int,int], name_suffix: str = " (Preview)"):
+        x, y, w, h = roi
+        meta = dict(parent_doc.metadata or {})
+        base = parent_doc.display_name()
+        meta["display_name"] = f"{base}{name_suffix}"
+        meta.setdefault("image_meta", {})
+        meta["image_meta"] = dict(meta["image_meta"], readonly=True, view_kind="roi-preview")
+
+        super().__init__(_np.zeros((max(1,h), max(1,w), 3), dtype=_np.float32), meta, parent=parent_doc.parent())
+
+        self._parent_doc = parent_doc
+        self._roi = ( x, y, w, h )
+        self._roi_info = {"parent_doc": parent_doc, "roi": tuple(self._roi)}
+        self.metadata["_roi_bounds"] = tuple(self._roi)
+        imi = dict(self.metadata.get("image_meta") or {})
+        imi.update({"roi": tuple(self._roi), "view_kind": "roi-preview"})
+        self.metadata["image_meta"] = imi
+
+        # build and store an ROI-shifted WCS header snapshot to use if detached
+        try:
+            phdr = (parent_doc.metadata.get("original_header")
+                    or parent_doc.metadata.get("fits_header")
+                    or parent_doc.metadata.get("header"))
+            rx, ry, rw, rh = self._roi
+            roi_wcs = _compute_cropped_wcs(phdr, rx, ry, rw, rh)
+            self.metadata["roi_wcs_header"] = roi_wcs  # plain dict, drop-in safe
+
+            # 🔴 KEY FIX: for a standalone ROI doc, treat this cropped WCS
+            # as the "original_header" so view-drops / duplicates inherit it.
+            if phdr is not None:
+                # optional: preserve the full parent header
+                self.metadata.setdefault("parent_full_header", phdr)
+            self.metadata["original_header"] = roi_wcs
+            try:
+                from .doc_manager import _snapshot_header_for_metadata  # if you move it, adjust import
+            except Exception:
+                _snapshot_header_for_metadata = None
+
+            try:
+                if _snapshot_header_for_metadata is not None:
+                    _snapshot_header_for_metadata(self.metadata)
+            except Exception:
+                pass
+
+            # DEBUG: log parent vs ROI WCS
+            if _DEBUG_WCS:
+                base_name = parent_doc.display_name() if hasattr(parent_doc, "display_name") else "<parent>"
+                print(f"[WCS DEBUG] _RoiViewDocument.__init__: parent='{base_name}' roi={self._roi}")
+                _debug_log_wcs_context("  parent_header", phdr)
+                _debug_log_wcs_context("  roi_header", self.metadata)
+        except Exception as e:
+            if _DEBUG_WCS:
+                print(f"[WCS DEBUG] _RoiViewDocument.__init__ exception: {e}")
+            pass
+        self.metadata["image_meta"] = imi
+        # NEW: transient preview overlay for this ROI (None means "show parent slice")
+        self._preview_override: _np.ndarray | None = None
+
+        self._pundo: list[tuple[_np.ndarray, dict, str]] = []  # (img, meta, name)
+        self._predo: list[tuple[_np.ndarray, dict, str]] = []  # (img, meta, name)
+
+    @property
+    def image(self):
+        p = self._parent_doc
+        if p is None or getattr(p, "image", None) is None:
+            return None
+        x, y, w, h = self._roi
+        # If a preview override exists, show it; else show the live parent slice
+        return self._preview_override if self._preview_override is not None else p.image[y:y+h, x:x+w]
+
+    @image.setter
+    def image(self, _val):
+        # ignore: writes should use DocManager(update/commit) paths
+        pass
+
+
+    def commit_to_parent(self, new_image: _np.ndarray | None = None,
+                        metadata: dict | None = None, step_name: str = "Edit"):
+        """
+        Paste current preview (or provided new_image) back into the parent image
+        with proper undo and region repaint.
+        """
+        parent = getattr(self, "_parent_doc", None)
+        if parent is None or parent.image is None:
+            return
+
+        x, y, w, h = self._roi
+        # choose source
+        src = new_image
+        if src is None:
+            src = self._preview_override if self._preview_override is not None else parent.image[y:y+h, x:x+w]
+
+        img = _np.asarray(src, dtype=_np.float32, copy=False)
+        base = parent.image
+
+        # channel reconciliation
+        if base.ndim == 2 and img.ndim == 3 and img.shape[2] == 1:
+            img = img[..., 0]
+        if base.ndim == 3 and img.ndim == 2:
+            img = _np.repeat(img[..., None], base.shape[2], axis=2)
+        if img.shape[:2] != (h, w):
+            raise ValueError(f"Commit shape {img.shape[:2]} does not match ROI {(h, w)}")
+
+        # push undo on parent and paste
+        parent._undo.append((base.copy(), parent.metadata.copy(), step_name))
+        parent._redo.clear()
+        if metadata: parent.metadata.update(metadata)
+        parent.metadata.setdefault("step_name", step_name)
+
+        new_full = base.copy()
+        new_full[y:y+h, x:x+w] = img
+        parent.image = new_full
+        try: parent.changed.emit()
+        except Exception as e:
+            import logging
+            logging.debug(f"Exception suppressed: {type(e).__name__}: {e}")
+
+        # notify region update + repaint
+        dm = getattr(self, "_doc_manager", None) or getattr(parent, "_doc_manager", None)
+        if dm is not None:
+            try: dm.imageRegionUpdated.emit(parent, (x, y, w, h))
+            except Exception as e:
+                import logging
+                logging.debug(f"Exception suppressed: {type(e).__name__}: {e}")
+
+
+    # --- helper to snapshot what's currently visible in the Preview
+    def _current_preview_copy(self) -> _np.ndarray:
+        img = self.image  # property: returns override or parent slice
+        if img is None:
+            return _np.zeros((1, 1), dtype=_np.float32)
+        arr = _np.asarray(img, dtype=_np.float32)
+        return _np.ascontiguousarray(arr)
+
+    # === KEEP YOUR WORKING BODY; only 3 added lines are marked "NEW" ===
+    def apply_edit(self, new_image, metadata=None, step_name="Edit"):
+        x, y, w, h = self._roi
+        img = np.asarray(new_image, dtype=np.float32, copy=False)
+        base = self._parent_doc.image
+
+        _debug_log_undo(
+            "_RoiViewDocument.apply_edit.entry",
+            roi=(x, y, w, h),
+            parent_id=id(self._parent_doc) if self._parent_doc is not None else None,
+            roi_doc_id=id(self),
+            new_shape=getattr(img, "shape", None),
+            step_name=step_name,
+            pundo_len=len(self._pundo),
+            predo_len=len(self._predo),
+        )
+        if base is not None:
+            if base.ndim == 2 and img.ndim == 3 and img.shape[2] == 1:
+                img = img[..., 0]
+            if base.ndim == 3 and img.ndim == 2:
+                img = np.repeat(img[..., None], base.shape[2], axis=2)
+        if img.shape[:2] != (h, w):
+            raise ValueError(f"Preview edit shape {img.shape[:2]} != ROI {(h, w)}")
+        
+        img = np.ascontiguousarray(img) 
+
+        # snapshot current visible preview for local undo
+        self._pundo.append((self._current_preview_copy(), dict(self.metadata), step_name))
+        self._predo.clear()
+
+        self._preview_override = img
+        _debug_log_undo(
+            "_RoiViewDocument.apply_edit.after",
+            roi=(x, y, w, h),
+            preview_shape=getattr(self._preview_override, "shape", None),
+            pundo_len=len(self._pundo),
+            predo_len=len(self._predo),
+            step_name=step_name,
+        )
+
+        if metadata:
+            self.metadata.update(metadata)
+        self.metadata.setdefault("step_name", step_name)
+
+        # 1) notify ROI listeners (e.g. the main window via _on_roi_changed)
+        try:
+            self.changed.emit()
+        except Exception:
+            pass
+
+        # 2) optionally: tell DocManager "ROI preview changed" using base doc + ROI
+        dm = getattr(self, "_doc_manager", None)
+        if dm is not None and hasattr(dm, "previewRepaintRequested"):
+            try:
+                dm.previewRepaintRequested.emit(self._parent_doc, self._roi)
+            except Exception:
+                pass
+
+
+
+    def _parent(self):
+        return getattr(self, "_parent_doc", None)
+
+    def can_undo(self) -> bool:
+        # Prefer local preview history if present
+        if self._pundo:
+            return True
+        # Otherwise mirror parent’s history
+        p = getattr(self, "_parent_doc", None)
+        if p is not None and hasattr(p, "can_undo"):
+            try:
+                return bool(p.can_undo())
+            except Exception:
+                return False
+        return False
+
+    def can_redo(self) -> bool:
+        if self._predo:
+            return True
+        p = getattr(self, "_parent_doc", None)
+        if p is not None and hasattr(p, "can_redo"):
+            try:
+                return bool(p.can_redo())
+            except Exception:
+                return False
+        return False
+
+    def last_undo_name(self) -> str | None:
+        if self._pundo:
+            return self._pundo[-1][2]
+        p = getattr(self, "_parent_doc", None)
+        if p is not None and hasattr(p, "last_undo_name"):
+            try:
+                return p.last_undo_name()
+            except Exception:
+                return None
+        return None
+
+    def last_redo_name(self) -> str | None:
+        if self._predo:
+            return self._predo[-1][2]
+        p = getattr(self, "_parent_doc", None)
+        if p is not None and hasattr(p, "last_redo_name"):
+            try:
+                return p.last_redo_name()
+            except Exception:
+                return None
+        return None
+
+    def undo(self) -> str | None:
+        # --- Case 1: ROI-local preview history ---
+        if self._pundo:
+            _debug_log_undo(
+                "_RoiViewDocument.undo.local.entry",
+                roi=self._roi,
+                roi_doc_id=id(self),
+                pundo_len=len(self._pundo),
+                predo_len=len(self._predo),
+            )
+            # move current → redo; pop undo → current
+            curr = self._current_preview_copy()
+            self._predo.append((curr, dict(self.metadata), self._pundo[-1][2]))
+
+            prev_img, prev_meta, name = self._pundo.pop()
+            self._preview_override = prev_img
+            self.metadata = dict(prev_meta)
+            _debug_log_undo(
+                "_RoiViewDocument.undo.local.apply",
+                roi=self._roi,
+                new_preview_shape=getattr(prev_img, "shape", None),
+                pundo_len=len(self._pundo),
+                predo_len=len(self._predo),
+                name=name,
+            )
+            try:
+                self.changed.emit()
+            except Exception:
+                pass
+
+            dm = getattr(self, "_doc_manager", None)
+            if dm is not None and hasattr(dm, "previewRepaintRequested"):
+                try:
+                    dm.previewRepaintRequested.emit(self._parent_doc, self._roi)
+                except Exception:
+                    pass
+            return name
+
+        # --- Case 2: no ROI-local history → delegate to parent ---
+        parent = getattr(self, "_parent_doc", None)
+        if parent is None or not hasattr(parent, "undo"):
+            return None
+        _debug_log_undo(
+            "_RoiViewDocument.undo.parent.entry",
+            roi=self._roi,
+            roi_doc_id=id(self),
+            parent_id=id(parent),
+            parent_undo_len=len(getattr(parent, "_undo", [])),
+            parent_redo_len=len(getattr(parent, "_redo", [])),
+        )
+
+        name = parent.undo()
+
+        # After parent changes, clear override so we show the new parent slice
+        self._preview_override = None
+
+        try:
+            self.changed.emit()
+        except Exception:
+            pass
+
+        dm = getattr(self, "_doc_manager", None) or getattr(parent, "_doc_manager", None)
+        if dm is not None and hasattr(dm, "previewRepaintRequested"):
+            try:
+                dm.previewRepaintRequested.emit(parent, self._roi)
+            except Exception:
+                pass
+        return name
+
+    def redo(self) -> str | None:
+        # --- Case 1: ROI-local preview history ---
+        if self._predo:
+            # move current → undo; pop redo → current
+            curr = self._current_preview_copy()
+            self._pundo.append((curr, dict(self.metadata), self._predo[-1][2]))
+
+            nxt_img, nxt_meta, name = self._predo.pop()
+            self._preview_override = nxt_img
+            self.metadata = dict(nxt_meta)
+
+            try:
+                self.changed.emit()
+            except Exception:
+                pass
+
+            dm = getattr(self, "_doc_manager", None)
+            if dm is not None and hasattr(dm, "previewRepaintRequested"):
+                try:
+                    dm.previewRepaintRequested.emit(self._parent_doc, self._roi)
+                except Exception:
+                    pass
+            return name
+
+        # --- Case 2: delegate to parent’s redo ---
+        parent = getattr(self, "_parent_doc", None)
+        if parent is None or not hasattr(parent, "redo"):
+            return None
+
+        name = parent.redo()
+
+        # Parent changed → reset override and repaint
+        self._preview_override = None
+
+        try:
+            self.changed.emit()
+        except Exception:
+            pass
+
+        dm = getattr(self, "_doc_manager", None) or getattr(parent, "_doc_manager", None)
+        if dm is not None and hasattr(dm, "previewRepaintRequested"):
+            try:
+                dm.previewRepaintRequested.emit(parent, self._roi)
+            except Exception:
+                pass
+        return name
+
+
+
+class LiveViewDocument(QObject):
+    """
+    Drop-in proxy that mirrors an ImageDocument API but always resolves
+    via DocManager + view to the ROI-aware document (if a Preview tab is active).
+    Reads: delegate to current resolved doc.
+    Writes: use DocManager.update_active_document(...) so ROI is pasted back.
+    """
+    changed = pyqtSignal()
+
+    def __init__(self, doc_manager: "DocManager", view, base_doc: "ImageDocument"):
+        super().__init__(parent=base_doc.parent())
+        self._dm = doc_manager
+        self._view = view              # ImageSubWindow widget
+        self._base = base_doc          # true ImageDocument
+
+        # Bridge base document change signals (ROI wrappers rarely emit)
+        try:
+            base_doc.changed.connect(self.changed.emit)
+        except Exception:
+            pass
+
+    # ---- core resolver ----
+    def _current(self):
+        try:
+            d = self._dm.get_document_for_view(self._view)
+            return d or self._base
+        except Exception:
+            return self._base
+
+    # ---- common API surface (reads) ----
+    @property
+    def image(self):
+        d = self._current()
+        return getattr(d, "image", None)
+
+    @property
+    def metadata(self):
+        d = self._current()
+        return getattr(d, "metadata", {}) or {}
+
+    def display_name(self):
+        d = self._current()
+        if hasattr(d, "display_name"):
+            try:
+                return d.display_name()
+            except Exception:
+                pass
+        return self._base.display_name() if hasattr(self._base, "display_name") else "Untitled"
+
+    # Mask access stays consistent with whichever doc is current
+    def get_active_mask(self):
+        d = self._current()
+        if hasattr(d, "get_active_mask"):
+            try:
+                return d.get_active_mask()
+            except Exception:
+                return None
+        return None
+
+    @property
+    def masks(self):
+        d = self._current()
+        return getattr(d, "masks", {})
+
+    @property
+    def active_mask_id(self):
+        d = self._current()
+        return getattr(d, "active_mask_id", None)
+
+    # ---- writes route through DocManager so ROI is honored ----
+    def apply_edit(self, new_image, metadata=None, step_name="Edit"):
+        #print("[LiveViewDocument] apply_edit called, routing via DocManager")
+        self._dm.update_active_document(new_image, dict(metadata or {}), step_name)
+
+    # ---- history helpers (optional pass-throughs) ----
+    def can_undo(self):
+        d = self._current()
+        return bool(getattr(d, "can_undo", lambda: False)())
+
+    def can_redo(self):
+        d = self._current()
+        return bool(getattr(d, "can_redo", lambda: False)())
+
+    def last_undo_name(self):
+        d = self._current()
+        return getattr(d, "last_undo_name", lambda: None)()
+
+    def last_redo_name(self):
+        d = self._current()
+        return getattr(d, "last_redo_name", lambda: None)()
+
+    def undo(self):
+        d = self._current()
+        _debug_log_undo(
+            "LiveViewDocument.undo.call",
+            live_id=id(self),
+            resolved_type=type(d).__name__ if d is not None else None,
+            resolved_id=id(d) if d is not None else None,
+            is_roi=isinstance(d, _RoiViewDocument),
+            has_undo=getattr(d, "can_undo", lambda: False)(),
+        )
+        return getattr(d, "undo", lambda: None)()
+
+    def redo(self):
+        d = self._current()
+        _debug_log_undo(
+            "LiveViewDocument.redo.call",
+            live_id=id(self),
+            resolved_type=type(d).__name__ if d is not None else None,
+            resolved_id=id(d) if d is not None else None,
+            is_roi=isinstance(d, _RoiViewDocument),
+            has_redo=getattr(d, "can_redo", lambda: False)(),
+        )
+        return getattr(d, "redo", lambda: None)()
+
+
+    # ---- generic fallback so existing attributes keep working ----
+    def __getattr__(self, name):
+        # Prefer the current resolved doc, then base_doc
+        d = object.__getattribute__(self, "_current")()
+        if hasattr(d, name):
+            return getattr(d, name)
+        return getattr(self._base, name)
+
+def _xisf_meta_to_fits_header(m: dict) -> fits.Header | None:
+    """
+    Best-effort: pull common WCS keys out of XISF FITSKeywords into a fits.Header.
+    Returns None if nothing usable found.
+    """
+    fk = m.get("FITSKeywords", {}) if isinstance(m, dict) else {}
+    if not fk:
+        return None
+
+    want = (
+        "WCSAXES", "CTYPE1", "CTYPE2", "CRPIX1", "CRPIX2",
+        "CRVAL1", "CRVAL2", "CD1_1", "CD1_2", "CD2_1", "CD2_2",
+        "CDELT1", "CDELT2", "PC1_1", "PC1_2", "PC2_1", "PC2_2",
+        "A_ORDER", "B_ORDER"
+    )
+
+    hdr = fits.Header()
+    found = False
+    for k in want:
+        vlist = fk.get(k)
+        if vlist and isinstance(vlist, list) and vlist[0].get("value") is not None:
+            hdr[k] = vlist[0]["value"]
+            found = True
+
+    # also pull SIP coeffs if present
+    for k, vlist in fk.items():
+        if k.startswith(("A_", "B_", "AP_", "BP_")) and vlist and vlist[0].get("value") is not None:
+            try:
+                hdr[k] = float(vlist[0]["value"])
+                found = True
+            except Exception:
+                pass
+
+    return hdr if found else None
+
+DEBUG_SAVE_DOCUMENT = False
+
+def debug_dump_metadata_print(meta: dict, context: str = ""):
+    if DEBUG_SAVE_DOCUMENT:
+        print(f"\n===== METADATA DUMP ({context}) =====")
+        if not isinstance(meta, dict):
+            print("  (not a dict) ->", type(meta))
+            print("====================================")
+            return
+
+        keys = sorted(str(k) for k in meta.keys())
+        print("  keys:", ", ".join(keys))
+
+        for key in keys:
+            val = meta[key]
+            if isinstance(val, fits.Header):
+                print(f"  {key}: fits.Header with {len(val.cards)} cards")
+            else:
+                print(f"  {key}: {val!r} ({type(val).__name__})")
+
+        print("===== END METADATA DUMP ({}) =====".format(context))
+
+class DocManager(QObject):
+    documentAdded = pyqtSignal(object)   # ImageDocument
+    documentRemoved = pyqtSignal(object) # ImageDocument
+    imageRegionUpdated = pyqtSignal(object, object)  # (doc, roi_tuple_or_None)
+    previewRepaintRequested = pyqtSignal(object, object)
+    
+    activeBaseChanged = pyqtSignal(object)  # emits ImageDocument | None
+
+    def __init__(self, image_manager=None, parent=None):
+        super().__init__(parent)
+        self.image_manager = image_manager
+        self._roi_doc_cache = {} 
+        self._docs: list[ImageDocument] = []
+        self._active_doc: ImageDocument | None = None
+        self._mdi: "QMdiArea | None" = None  # type: ignore
+        def _on_region_updated(doc, roi):
+            vw = self._active_view_widget()
+            if vw is not None:
+                try:
+                    if hasattr(vw, "refresh_from_docman") and callable(vw.refresh_from_docman):
+                        vw.refresh_from_docman(doc=doc, roi=roi)
+                    else:
+                        vw._render()
+                except Exception:
+                    pass
+
+        self.imageRegionUpdated.connect(_on_region_updated)
+        self._by_uid = {}
+        self._focused_base_doc: ImageDocument | None = None  # <— NEW
+
+        def _do_preview_repaint(doc, roi):
+            vw = self._active_view_widget()
+            if vw is not None:
+                try:
+                    if hasattr(vw, "refresh_from_docman") and callable(vw.refresh_from_docman):
+                        vw.refresh_from_docman(doc=doc, roi=roi)
+                    else:
+                        vw._render()
+                except Exception:
+                    pass
+        self.previewRepaintRequested.connect(_do_preview_repaint)
+
+    def get_document_for_view(self, view):
+        """
+        Given an ImageSubWindow widget, return either:
+        - the full base ImageDocument
+        - or a cached ROI-wrapper doc if a Preview/ROI tab is active
+        Works with both old (has_active_preview/current_preview_roi) and
+        new (_active_roi_tuple) view APIs. Falls back to view.document.
+        """
+        # 1) Resolve a base document from the view
+        base = (
+            getattr(view, "base_document", None)
+            or getattr(view, "_base_document", None)
+            or getattr(view, "document", None)
+        )
+        if base is None:
+            return None
+
+        # 2) Try to discover an ROI (support both APIs)
+        roi = None
+        try:
+            if hasattr(view, "has_active_preview") and callable(view.has_active_preview):
+                if view.has_active_preview():
+                    # preferred old API
+                    try:
+                        roi = view.current_preview_roi()  # (x,y,w,h)
+                    except Exception:
+                        roi = None
+        except Exception:
+            pass
+
+        if roi is None:
+            # new API candidate
+            for attr in ("_active_roi_tuple", "current_roi_tuple", "selected_roi", "roi"):
+                try:
+                    fn = getattr(view, attr, None)
+                    if callable(fn):
+                        r = fn()
+                        if r and len(r) == 4:
+                            roi = r
+                            break
+                except Exception:
+                    pass
+
+        # 3) If no ROI, return the base doc
+        if not roi:
+            return base
+
+        # 4) Cache and return a lightweight ROI view doc
+        try:
+            x, y, w, h = map(int, roi)
+            key = (id(base), id(view), (x, y, w, h))
+            roi_doc = self._roi_doc_cache.get(key)
+            if roi_doc is None:
+                roi_doc = self._build_roi_document(base, (x, y, w, h))
+                self._roi_doc_cache[key] = roi_doc
+            return roi_doc
+        except Exception:
+            # If anything about ROI construction fails, fall back
+            return base
+
+    def _invalidate_roi_cache(self, parent_doc, roi_tuple):
+        """Drop cached ROI docs that overlap an updated region of parent_doc."""
+        if not roi_tuple:
+            # full-image change -> drop all for this parent
+            dead = [k for k in self._roi_doc_cache.keys() if k[0] == id(parent_doc)]
+        else:
+            px, py, pw, ph = roi_tuple
+            def _overlaps(a, b):
+                ax, ay, aw, ah = a; bx, by, bw, bh = b
+                return not (ax+aw <= bx or bx+bw <= ax or ay+ah <= by or by+bh <= ay)
+            dead = []
+            for (parent_id, _view_id, aroi), _doc in list(self._roi_doc_cache.items()):
+                if parent_id != id(parent_doc):
+                    continue
+                if _overlaps(aroi, (px, py, pw, ph)):
+                    dead.append((parent_id, _view_id, aroi))
+        for k in dead:
+            self._roi_doc_cache.pop(k, None)
+
+    def get_document_by_uid(self, uid: str):
+        return self._by_uid.get(uid)
+
+
+    def _register_doc(self, doc):
+        import weakref
+        # Only ImageDocument needs the backref; tables can ignore it.
+        if hasattr(doc, "image") or hasattr(doc, "apply_edit"):
+            try:
+                doc._doc_manager = weakref.proxy(self)   # avoid cycles
+            except Exception:
+                doc._doc_manager = self                  # fallback
+            self._docs.append(doc)
+            if hasattr(doc, "uid"):
+                self._by_uid[doc.uid] = doc
+            self.documentAdded.emit(doc)
+
+    def _build_roi_document(self, base_doc, roi):
+        #print("[DocManager] Building ROI view document")
+        doc = _RoiViewDocument(base_doc, roi, name_suffix=" (Preview)")
+        try:
+            import weakref
+            doc._doc_manager = weakref.proxy(self)
+        except Exception:
+            doc._doc_manager = self
+
+        # Repaint the active view on ROI preview changes, but DO NOT invalidate cache.
+        try:
+            #print("[DocManager] Connecting ROI view document change signal")
+            import weakref
+            dm_ref = weakref.ref(self)
+            roi_tuple = tuple(map(int, roi))
+
+            def _on_roi_changed():
+                dm = dm_ref()
+                if dm is None:
+                    return
+                vw = dm._active_view_widget()
+                if vw is not None:
+                    try:
+                        # IMPORTANT: use the *parent* doc here, not the ROI wrapper
+                        base = getattr(doc, "_parent_doc", None) or doc
+                        if hasattr(vw, "refresh_from_docman") and callable(vw.refresh_from_docman):
+                            vw.refresh_from_docman(doc=base, roi=roi_tuple)
+                        else:
+                            vw._render()
+                    except Exception:
+                        pass
+
+            doc.changed.connect(_on_roi_changed)
+
+            #print("[DocManager] ROI view document change signal connected")
+        except Exception:
+            print("[DocManager] Failed to connect ROI view document change signal")
+            pass
+
+        return doc
+
+    def commit_active_preview_to_parent(self, metadata: dict | None = None, step_name: str = "Edit"):
+        doc = self.get_active_document()
+        if isinstance(doc, _RoiViewDocument):
+            doc.commit_to_parent(None, metadata=metadata or {}, step_name=step_name)
+            # after commit, force an immediate view repaint
+            vw = self._active_view_widget()
+            if vw is not None:
+                try:
+                    if hasattr(vw, "refresh_from_docman") and callable(vw.refresh_from_docman):
+                        vw.refresh_from_docman(doc=doc._parent_doc, roi=None)
+                    else:
+                        vw._render()
+                except Exception:
+                    pass
+
+    def wrap_document_for_view(self, view, base_doc: ImageDocument) -> LiveViewDocument:
+        """Return a live, ROI-aware proxy for this view."""
+        return LiveViewDocument(self, view, base_doc)
+
+    def open_path(self, path: str):
+        ext = os.path.splitext(path)[1].lower().lstrip('.')
+        norm_ext = _normalize_ext(ext)
+
+        lower_path = path.lower()
+        is_fits = lower_path.endswith((".fit", ".fits", ".fit.gz", ".fits.gz", ".fz"))
+        is_xisf = (norm_ext == "xisf")
+
+        primary_doc = None
+        created_any = False
+
+        # ---------- 1) Try the universal loader first (ALL formats) ----------
+        img = header = bit_depth = is_mono = None
+        meta = None
+        try:
+            # NEW: prefer metadata-aware return
+            out = legacy_load_image(path, return_metadata=True)
+            if out and len(out) == 5:
+                img, header, bit_depth, is_mono, meta = out
+            else:
+                img, header, bit_depth, is_mono = out
+        except TypeError:
+            # legacy_load_image older signature → fall back
+            try:
+                img, header, bit_depth, is_mono = legacy_load_image(path)
+            except Exception as e:
+                print(f"[DocManager] legacy_load_image failed (non-fatal if FITS/XISF): {e}")
+        except Exception as e:
+            print(f"[DocManager] legacy_load_image failed (non-fatal if FITS/XISF): {e}")
+
+        maybe_warn_raw_preview(path, header)
+
+        if img is not None:
+            if meta is None:
+                meta = {
+                    "file_path": path,
+                    "original_header": header,
+                    "bit_depth": bit_depth,
+                    "is_mono": is_mono,
+                    "original_format": norm_ext,
+                }
+
+                # NEW: attach WCS even for old loader
+                meta = attach_wcs_to_metadata(meta, header)
+
+            _snapshot_header_for_metadata(meta)
+
+            img = _normalize_image_01(img)
+            primary_doc = ImageDocument(img, meta)
+            self._register_doc(primary_doc)
+            created_any = True
+
+
+
+        # ---------- 2) FITS: enumerate HDUs (tables + extra images + ICC) ----------
+        if is_fits:
+            try:
+                with fits.open(path, memmap=True) as hdul:
+                    base = os.path.basename(path)
+
+
+                    for i, hdu in enumerate(hdul):
+                        name_up = (getattr(hdu, "name", "PRIMARY") or "PRIMARY").upper()
+                        if primary_doc is not None and (i == 0 or name_up == "PRIMARY"):
+
+                            continue
+
+                        ext_hdr = hdu.header
+                        try:
+                            en = str(ext_hdr.get("EXTNAME", "")).strip()
+                            ev = ext_hdr.get("EXTVER", None)
+                            extname = f"{en}[{int(ev)}]" if (en and isinstance(ev, (int, np.integer))) else (en or "")
+                        except Exception:
+                            extname = ""
+
+                        # --- Tables → TableDocument ---
+                        if isinstance(hdu, (fits.BinTableHDU, fits.TableHDU)):
+                            key_str = extname or f"HDU{i}"
+                            nice = key_str
+                            #print(f"[DocManager] HDU {i}: {type(hdu).__name__} '{nice}' → Table")
+
+                            # Optional CSV export
+                            csv_name = f"{os.path.splitext(path)[0]}_{key_str}.csv".replace(" ", "_")
+                            try:
+                                _ = _fits_table_to_csv(hdu, csv_name)
+                            except Exception as e_csv:
+                                print(f"[DocManager] Table CSV export failed ({nice}): {e_csv}")
+                                csv_name = None
+
+                            # Build in-app table
+                            try:
+                                rows, headers = _fits_table_to_rows_headers(hdu, max_rows=500000)
+                                tmeta = {
+                                    "file_path": f"{path}::{key_str}",
+                                    "original_header": ext_hdr,
+                                    "original_format": "fits",
+                                    "display_name": f"{base} {key_str} (Table)",
+                                    "doc_type": "table",
+                                    "table_csv": csv_name if (csv_name and os.path.exists(csv_name)) else None,
+                                }
+                                _snapshot_header_for_metadata(tmeta)
+                                tdoc = TableDocument(rows, headers, tmeta, parent=self.parent())
+                                self._register_doc(tdoc)
+                                try: tdoc.changed.emit()
+                                except Exception as e:
+                                    import logging
+                                    logging.debug(f"Exception suppressed: {type(e).__name__}: {e}")
+                                created_any = True
+                                #print(f"[DocManager] Added TableDocument: rows={len(rows)} cols={len(headers)} title='{tdoc.display_name()}'")
+                            except Exception as e_tab:
+                                print(f"[DocManager] Table HDU {nice} → in-app view failed: {e_tab}")
+                            continue  # IMPORTANT: don’t treat a table as an image
+
+                        # --- Not a table: ICC or image ---
+                        if hdu.data is None:
+                            #print(f"[DocManager] HDU {i} '{extname or f'HDU{i}'}' has no data — noted as aux")
+                            continue
+
+                        arr = np.asanyarray(hdu.data)
+                        en_up = (extname or "").upper()
+                        is_probable_icc = ("ICC" in en_up or "PROFILE" in en_up)
+
+                        # ICC ONLY if name suggests ICC/profile AND data is 1-D uint8
+                        if arr.ndim == 1 and arr.dtype == np.uint8 and is_probable_icc:
+                            try:
+                                icc_path = f"{os.path.splitext(path)[0]}_{extname or f'HDU{i}'}_.icc".replace(" ", "_")
+                                with open(icc_path, "wb") as f:
+                                    f.write(arr.tobytes())
+                                #print(f"[DocManager] Extracted ICC profile → {icc_path}")
+                                created_any = True
+                                continue
+                            except Exception as e_icc:
+                                print(f"[DocManager] ICC export failed: {e_icc} — will try as image")
+
+                        # Otherwise: treat as image doc
+                        try:
+                            if arr.dtype.kind in "ui":
+                                a = arr.astype(np.float32, copy=False)  # NO normalization
+                                # optional: if you want to record original scale:
+                                ext_depth = f"{arr.dtype.itemsize*8}-bit {'unsigned' if arr.dtype.kind=='u' else 'signed'}"
+                            else:
+                                a = arr.astype(np.float32, copy=False)  # floats preserved
+                                ext_depth = "32-bit floating point" if arr.dtype == np.float32 else "64-bit floating point"
+
+                            ext_mono = bool(a.ndim == 2 or (a.ndim == 3 and a.shape[2] == 1))
+                            key_str = extname or f"HDU {i}"
+                            disp = f"{base} {key_str}"
+
+                            aux_meta = {
+                                "file_path": f"{path}::{key_str}",
+                                "original_header": ext_hdr,
+                                "bit_depth": ext_depth,
+                                "is_mono": bool(ext_mono),
+                                "original_format": "fits",
+                                "image_meta": {"derived_from": path, "layer": key_str, "readonly": True},
+                                "display_name": disp,
+                            }
+
+                            # NEW: attach WCS from this HDU header
+                            aux_meta = attach_wcs_to_metadata(aux_meta, ext_hdr)
+
+                            _snapshot_header_for_metadata(aux_meta)
+                            a = _normalize_image_01(a)
+                            aux_doc = ImageDocument(a, aux_meta)
+
+                            self._register_doc(aux_doc)
+                            try: aux_doc.changed.emit()
+                            except Exception as e:
+                                import logging
+                                logging.debug(f"Exception suppressed: {type(e).__name__}: {e}")
+                            created_any = True
+
+                        except Exception as e_img:
+                            print(f"[DocManager] FITS HDU {i} image build failed: {e_img}")
+            except Exception as _e:
+                print(f"[DocManager] FITS HDU enumeration failed: {_e}")
+
+        # ---------- 3) XISF: create primary if needed, then enumerate extras ----------
+        if is_xisf:
+            try:
+                # helpers
+                def _bit_depth_from_dtype(dt: np.dtype) -> str:
+                    dt = np.dtype(dt)
+                    if dt == np.float32: return "32-bit floating point"
+                    if dt == np.float64: return "64-bit floating point"
+                    if dt == np.uint8:   return "8-bit"
+                    if dt == np.uint16:  return "16-bit"
+                    if dt == np.uint32:  return "32-bit unsigned"
+                    return "32-bit floating point"
+
+                def _to_float32_01(arr: np.ndarray) -> np.ndarray:
+                    a = np.asarray(arr)
+                    if a.dtype == np.float32:
+                        return a
+                    if a.dtype.kind in "iu":
+                        return (a.astype(np.float32) / np.iinfo(a.dtype).max).clip(0.0, 1.0)
+                    return a.astype(np.float32, copy=False)
+
+                def _to_float32_preserve(arr: np.ndarray) -> np.ndarray:
+                    a = np.asarray(arr)
+                    return a if a.dtype == np.float32 else a.astype(np.float32, copy=False)
+
+                xisf = XISFReader(path)
+                metas = xisf.get_images_metadata() or []
+                base = os.path.basename(path)
+
+                # If legacy did NOT create a primary, build image #0 now
+                if primary_doc is None and len(metas) >= 1:
+                    try:
+                        arr0 = xisf.read_image(0, data_format="channels_last")
+                        arr0_f32 = _to_float32_preserve(arr0)
+                        arr0_f32 = _normalize_image_01(arr0_f32)
+                        bd0 = _bit_depth_from_dtype(metas[0].get("dtype", arr0.dtype))
+                        is_mono0 = (arr0_f32.ndim == 2) or (arr0_f32.ndim == 3 and arr0_f32.shape[2] == 1)
+
+                        # Friendly label for #0
+                        label0 = metas[0].get("id") or "Image[0]"
+                        md0 = {
+                            "file_path": f"{path}::XISF[0]",
+                            "original_header": metas[0],  # will be sanitized
+                            "bit_depth": bd0,
+                            "is_mono": is_mono0,
+                            "original_format": "xisf",
+                            "image_meta": {"derived_from": path, "layer_index": 0, "readonly": True},
+                            "display_name": f"{base} {label0}",
+                        }
+                        # NEW: attach WCS if possible
+                        hdr0 = _xisf_meta_to_fits_header(metas[0])
+                        if hdr0 is not None:
+                            md0 = attach_wcs_to_metadata(md0, hdr0)
+
+                        _snapshot_header_for_metadata(md0)
+                        primary_doc = ImageDocument(arr0_f32, md0)
+                        self._register_doc(primary_doc)
+                        try: primary_doc.changed.emit()
+                        except Exception as e:
+                            import logging
+                            logging.debug(f"Exception suppressed: {type(e).__name__}: {e}")
+                        created_any = True
+
+                    except Exception as e0:
+                        print(f"[DocManager] XISF primary (index 0) open failed: {e0}")
+
+                # Add images 1..N-1 as siblings (even if primary came from legacy)
+                for i in range(1, len(metas)):
+                    try:
+                        m = metas[i]
+                        arr = xisf.read_image(i, data_format="channels_last")
+                        arr_f32 = _to_float32_preserve(arr)
+                        arr_f32 = _normalize_image_01(arr_f32)
+
+                        bd = _bit_depth_from_dtype(m.get("dtype", arr.dtype))
+                        is_mono_i = (arr_f32.ndim == 2) or (arr_f32.ndim == 3 and arr_f32.shape[2] == 1)
+
+                        # Friendly label: prefer id, else EXTNAME/EXTVER in FITSKeywords, else index
+                        label = m.get("id") or None
+                        if not label:
+                            try:
+                                fk = m.get("FITSKeywords", {})
+                                en = (fk.get("EXTNAME") or [{}])[0].get("value", "")
+                                ev = (fk.get("EXTVER")  or [{}])[0].get("value", "")
+                                if en:
+                                    label = f"{en}[{ev}]" if ev else en
+                            except Exception:
+                                pass
+                        if not label:
+                            label = f"Image[{i}]"
+
+                        md = {
+                            "file_path": f"{path}::XISF[{i}]",
+                            "original_header": m,  # snapshot; sanitized below
+                            "bit_depth": bd,
+                            "is_mono": is_mono_i,
+                            "original_format": "xisf",
+                            "image_meta": {"derived_from": path, "layer_index": i, "readonly": True},
+                            "display_name": f"{base} {label}",
+                        }
+                        hdri = _xisf_meta_to_fits_header(m)
+                        if hdri is not None:
+                            md = attach_wcs_to_metadata(md, hdri)
+
+                        _snapshot_header_for_metadata(md)
+                        sib = ImageDocument(arr_f32, md)
+                        self._register_doc(sib)
+                        try: sib.changed.emit()
+                        except Exception as e:
+                            import logging
+                            logging.debug(f"Exception suppressed: {type(e).__name__}: {e}")
+                        created_any = True
+
+                    except Exception as _e:
+                        print(f"[DocManager] XISF image {i} skipped: {_e}")
+            except Exception as _e:
+                print(f"[DocManager] XISF open/enumeration failed: {_e}")
+
+        # ---------- 4) Return sensible doc or raise ----------
+        if primary_doc is not None:
+            return primary_doc
+        if created_any:
+            return self._docs[-1]  # e.g., a table-only FITS or extra XISF image
+
+        raise IOError(f"Could not load: {path}")
+    
+    # --- Subwindow / ROI awareness -------------------------------------
+    def _active_subwindow(self):
+        """Return the active QMdiSubWindow (if any)."""
+        if self._mdi is None:
+            return None
+        try:
+            return self._mdi.activeSubWindow()
+        except Exception:
+            return None
+
+    def _active_view_widget(self):
+        """Return the active view widget (ImageSubWindow or TableSubWindow)."""
+        sw = self._active_subwindow()
+        if not sw:
+            return None
+        try:
+            return sw.widget()
+        except Exception:
+            return None
+
+    def _active_preview_roi(self):
+        """
+        Returns (x,y,w,h) if the active view is an ImageSubWindow with a selected Preview tab.
+        Else returns None.
+        """
+        #print("[DocManager] Checking for active preview ROI")
+        vw = self._active_view_widget()
+        if vw and hasattr(vw, "has_active_preview") and vw.has_active_preview():
+            try:
+                return vw.current_preview_roi()
+            except Exception:
+                return None
+        return None
+
+    def get_active_image(self, prefer_preview: bool = True):
+        """
+        Unified read: returns the ndarray a tool should operate on.
+        If a Preview tab is active and prefer_preview=True, return that crop.
+        Otherwise return the full document image.
+        """
+        doc = self.get_active_document()
+        if doc is None or doc.image is None:
+            return None
+        roi = self._active_preview_roi() if prefer_preview else None
+        if roi is None:
+            return doc.image
+        x, y, w, h = roi
+        return doc.image[y:y+h, x:x+w]
+
+
+    # --- Slot -> Document ---
+    def open_from_slot(self, slot_idx: int | None = None) -> "ImageDocument | None":
+        if not self.image_manager:
+            return None
+
+        if slot_idx is None:
+            slot_idx = getattr(self.image_manager, "current_slot", 0)
+
+        img = self.image_manager.get_image_for_slot(slot_idx)
+        if img is None:
+            return None
+
+        meta = {}
+        try:
+            meta = dict(self.image_manager._metadata.get(slot_idx, {}))
+        except Exception:
+            pass
+
+        meta.setdefault("file_path", f"Slot {slot_idx}")
+        meta.setdefault("bit_depth", "32-bit floating point")
+        meta.setdefault("is_mono", (img.ndim == 2))
+        meta.setdefault("original_header", meta.get("original_header"))  # whatever SASv2 had
+        meta.setdefault("original_format", "fits")
+
+        _snapshot_header_for_metadata(meta)
+
+        doc = ImageDocument(img, meta)
+        self._register_doc(doc)
+        return doc
+    
+    # --- Save ---
+    def _infer_bit_depth_for_format(self, img: np.ndarray, ext: str, current_bit_depth: str | None) -> str:
+        # Previous heuristic fallback (used only if no override provided).
+        if ext in ("png", "jpg"):
+            return "8-bit"
+        if ext in ("fits", "fit"):
+            return "32-bit floating point"
+        if ext == "tif":
+            if current_bit_depth in _ALLOWED_DEPTHS["tif"]:
+                return current_bit_depth
+            return "16-bit" if np.issubdtype(img.dtype, np.floating) else "8-bit"
+        if ext == "xisf":
+            return current_bit_depth if current_bit_depth in _ALLOWED_DEPTHS["xisf"] else "32-bit floating point"
+        return "32-bit floating point"
+
+    def save_document(
+        self,
+        doc: "ImageDocument",
+        path: str,
+        bit_depth: str | None = None,
+        *,
+        bit_depth_override: str | None = None,
+    ):
+        """
+        Save the given ImageDocument to 'path'.
+
+        bit_depth_override:
+            New-style explicit choice from a dialog.
+
+        bit_depth:
+            Legacy positional argument; still honored if override is None.
+        """
+        ext = _normalize_ext(os.path.splitext(path)[1])
+        img = doc.image
+        meta = doc.metadata or {}
+
+        # ── MASSIVE DEBUG: show everything we know coming in ───────────────
+        debug_dump_metadata_print(meta, context="save_document: BEFORE HEADER PICK")
+
+        # --- Decide final bit depth ---------------------------------------
+        requested = bit_depth_override or bit_depth or meta.get("bit_depth")
+
+        if requested:
+            allowed = _ALLOWED_DEPTHS.get(ext, set())
+            if allowed and requested not in allowed:
+                print(f"[save_document] Requested bit depth {requested!r} "
+                      f"not in allowed {allowed}, falling back to first.")
+                final_bit_depth = next(iter(allowed))
+            else:
+                final_bit_depth = requested
+        else:
+            final_bit_depth = self._infer_bit_depth_for_format(
+                img, ext, meta.get("bit_depth")
+            )
+
+
+
+        # --- Clip if needed for integer encodes ---------------------------
+        needs_clip = (
+            ext in ("png", "jpg", "jpeg", "tif", "tiff")
+            and final_bit_depth in ("8-bit", "16-bit", "32-bit unsigned")
+        )
+        if needs_clip:
+            print("[save_document] Clipping image to [0,1] for integer encode.")
+        img_to_save = np.clip(img, 0.0, 1.0) if needs_clip else img
+
+        # --- PICK THE HEADER EXPLICITLY -----------------------------------
+        # Priority:
+        #   1) wcs_header
+        #   2) fits_header
+        #   3) original_header
+        #   4) header
+        effective_header = None
+        for key in ("original_header", "fits_header", "wcs_header", "header"):
+            val = meta.get(key)
+            if isinstance(val, fits.Header):
+                effective_header = val
+
+                break
+
+        #if effective_header is None:
+        #    print("[save_document] WARNING: No fits.Header in metadata, "
+        #          "legacy_save_image will pick a default header.")
+        #else:
+        #    # Print first few cards so we can confirm we have the SIP stuff
+        #    print("[save_document] effective_header preview (first 25 cards):")
+        #    for i, card in enumerate(effective_header.cards):
+        #        if i >= 25:
+        #            print("  ... (truncated)")
+        #            break
+        #        print(f"  {card.keyword:8s} = {card.value!r}")
+
+        # ── Call the legacy saver ─────────────────────────────────────────
+
+
+        legacy_save_image(
+            img_array=img_to_save,
+            filename=path,
+            original_format=ext,
+            bit_depth=final_bit_depth,
+            original_header=effective_header,
+            is_mono=meta.get("mono", img.ndim == 2),
+            image_meta=meta.get("image_meta"),
+            file_meta=meta.get("file_meta"),
+            wcs_header=meta.get("wcs_header"),
+        )
+
+        # ── Update metadata in memory to match what we just wrote ─────────
+        meta["file_path"] = path
+        meta["original_format"] = ext
+        meta["bit_depth"] = final_bit_depth
+
+        if isinstance(effective_header, fits.Header):
+            meta["original_header"] = effective_header
+
+            # If you have this helper, keep it; if not, you can skip it
+            try:
+                _snapshot_header_for_metadata(meta)
+            except Exception as e:
+                print("[save_document] _snapshot_header_for_metadata error:", e)
+
+        doc.metadata = meta
+
+        # reset dirty flag
+        if hasattr(doc, "dirty"):
+            doc.dirty = False
+
+        if hasattr(doc, "changed"):
+            doc.changed.emit()
+
+    def duplicate_document(self, source_doc: ImageDocument, new_name: str | None = None) -> ImageDocument:
+        # DEBUG: log the source doc WCS before we touch anything
+        if _DEBUG_WCS:
+            try:
+                name = source_doc.display_name() if hasattr(source_doc, "display_name") else "<src>"
+            except Exception:
+                name = "<src>"
+            
+            _debug_log_wcs_context("  source.metadata", getattr(source_doc, "metadata", {}))
+
+        # COPY-ON-WRITE: Share the source image instead of copying immediately.
+        # The duplicate's apply_edit will copy when it first modifies the image.
+        # This saves memory when duplicates are created but not modified.
+        img_ref = source_doc.image  # Shared reference, no copy
+
+        meta = dict(source_doc.metadata or {})
+        base = source_doc.display_name()
+        dup_title = (new_name or f"{base}_duplicate")
+        # 🚫 strip any lingering emojis / link markers
+        dup_title = dup_title.replace("🔗", "").strip()
+        meta["display_name"] = dup_title
+
+        # Remove anything that makes the view look "linked/preview"
+        imi = dict(meta.get("image_meta") or {})
+        for k in ("readonly", "view_kind", "derived_from", "layer", "layer_index", "linked"):
+            imi.pop(k, None)
+        meta["image_meta"] = imi
+        for k in list(meta.keys()):
+            if k.startswith("_roi_") or k.endswith("_roi") or k == "roi":
+                meta.pop(k, None)
+
+        # NOTE: we intentionally DO NOT remove "roi_wcs_header" or "original_header"
+        # so that a ROI doc keeps its cropped WCS in the duplicate.
+
+        # Safe bit depth / mono flags
+        meta.setdefault("original_format", meta.get("original_format", "fits"))
+        if isinstance(img_ref, np.ndarray):
+            meta["is_mono"] = (img_ref.ndim == 2 or (img_ref.ndim == 3 and img_ref.shape[2] == 1))
+
+        _snapshot_header_for_metadata(meta)
+
+        dup = ImageDocument(img_ref, meta, parent=self.parent())
+        # Mark this duplicate as sharing image data with source
+        dup._cow_source = source_doc
+        self._register_doc(dup)
+
+        # DEBUG: log the duplicate doc WCS
+        if _DEBUG_WCS:
+            try:
+                dname = dup.display_name()
+            except Exception:
+                dname = "<dup>"
+
+            _debug_log_wcs_context("  duplicate.metadata", dup.metadata)
+
+        return dup
+
+    #def open_array(self, arr, metadata: dict | None = None, title: str | None = None) -> ImageDocument:
+    #    import numpy as np
+    ##    if arr is None:
+    #        raise ValueError("open_array: arr is None")
+    #    img = np.asarray(arr)
+    #    if img.dtype != np.float32:
+    #        img = img.astype(np.float32, copy=False)
+
+    #    meta = dict(metadata or {})
+    #    meta.setdefault("bit_depth", "32-bit floating point")
+    #    meta.setdefault("is_mono", img.ndim == 2)
+    #    meta.setdefault("original_header", meta.get("original_header"))
+    #    meta.setdefault("original_format", meta.get("original_format", "fits"))
+    #    if title:
+    #        meta.setdefault("display_name", title)
+
+    #    doc = ImageDocument(img, meta, parent=self.parent())
+    #    self._docs.append(doc)
+    #    self.documentAdded.emit(doc)
+    #    return doc
+
+    # convenient aliases used by your tool code
+    def open_array(self, img: np.ndarray, metadata: dict | None = None, title: str | None = None) -> "ImageDocument":
+        meta = dict(metadata or {})
+        if title:
+            meta["display_name"] = title
+        # normalize a few expected fields if missing
+        try:
+            if "is_mono" not in meta and isinstance(img, np.ndarray):
+                meta["is_mono"] = (img.ndim == 2 or (img.ndim == 3 and img.shape[2] == 1))
+        except Exception:
+            pass
+        meta.setdefault("bit_depth", meta.get("bit_depth", "32-bit floating point"))
+
+        _snapshot_header_for_metadata(meta)
+
+        doc = ImageDocument(img, meta, parent=self.parent())
+        self._register_doc(doc)
+        return doc
+
+    # (optional alias for old code)
+    open_numpy = open_array
+
+
+    def create_document(self, image, metadata: dict | None = None, name: str | None = None) -> ImageDocument:
+        return self.open_array(image, metadata=metadata, title=name)
+
+    def close_document(self, doc):
+        if doc in self._docs:
+            self._docs.remove(doc)
+            try:
+                if hasattr(doc, "uid"):
+                    self._by_uid.pop(doc.uid, None)
+            except Exception:
+                pass
+            
+            # Cleanup swap files
+            if hasattr(doc, "close"):
+                try:
+                    doc.close()
+                except Exception as e:
+                    print(f"[DocManager] Failed to close document {doc}: {e}")
+                    
+            self.documentRemoved.emit(doc)
+            
+    # --- Active-document helpers (NEW) ---------------------------------
+    def all_documents(self):
+        return list(self._docs)
+
+    def _find_main_window(self):
+        from PyQt6.QtWidgets import QMainWindow, QApplication
+        w = self.parent()
+        while w is not None and not isinstance(w, QMainWindow):
+            w = w.parent()
+        if w:
+            return w
+        for tlw in QApplication.topLevelWidgets():
+            if isinstance(tlw, QMainWindow):
+                return tlw
+        return None
+
+    def set_active_document(self, doc: ImageDocument | None):
+        if doc is not None and doc not in self._docs:
+            return
+        # ensure backref for legacy docs
+        if doc is not None and not hasattr(doc, "_doc_manager"):
+            try:
+                import weakref
+                doc._doc_manager = weakref.proxy(self)
+            except Exception:
+                doc._doc_manager = self
+        self._active_doc = doc
+
+    def set_mdi_area(self, mdi):
+        """Call this once from MainWindow after MDI is created."""
+        self._mdi = mdi
+        try:
+            mdi.subWindowActivated.connect(self._on_subwindow_activated)
+        except Exception:
+            pass
+
+    def _base_from_subwindow(self, sw):
+        """Best-effort: unwrap to the base ImageDocument bound to a subwindow."""
+        if sw is None:
+            return None
+        try:
+            w = sw.widget()
+            base = (getattr(w, "base_document", None)
+                    or getattr(w, "_base_document", None)
+                    or getattr(w, "document", None))
+            # unwrap ROI wrappers if any
+            p = getattr(base, "_parent_doc", None)
+            return p if isinstance(p, ImageDocument) else base
+        except Exception:
+            return None
+
+    def _on_subwindow_activated(self, sw):
+        # existing logic (keep it)
+        doc = None
+        try:
+            if sw is not None:
+                w = sw.widget()
+                doc = getattr(w, "document", None) or getattr(sw, "document", None)
+        except Exception:
+            doc = None
+        self.set_active_document(doc)
+
+        # NEW: compute focused *base* doc and emit change only when different
+        new_base = self._base_from_subwindow(sw)
+        if new_base is not self._focused_base_doc:
+            self._focused_base_doc = new_base
+            try:
+                self.activeBaseChanged.emit(new_base)
+            except Exception:
+                pass
+
+    def get_focused_base_document(self):
+        """
+        Returns the last *activated* subwindow's base ImageDocument (sticky),
+        ignoring hover/preview wrappers.
+        """
+        return self._focused_base_doc
+
+    def get_active_document(self):
+        """
+        Return the active document-like object.
+        If a Preview tab is selected on the active ImageSubWindow, return a cached
+        _RoiViewDocument so tools and the Preview tab share the same instance.
+        Otherwise return the real ImageDocument.
+        """
+        # Prefer cached (if set and still valid)
+        if self._active_doc is not None and self._active_doc in self._docs:
+            base_doc = self._active_doc
+        else:
+            base_doc = None
+            try:
+                if self._mdi is not None:
+                    sw = self._mdi.activeSubWindow()
+                    if sw is not None:
+                        w = sw.widget()
+                        base_doc = getattr(w, "document", None) or getattr(sw, "document", None)
+                        if base_doc is not None:
+                            self._active_doc = base_doc
+            except Exception:
+                pass
+            if base_doc is None:
+                base_doc = self._docs[-1] if self._docs else None
+
+        # Non-image docs just pass through
+        if base_doc is None or not isinstance(base_doc, ImageDocument) or base_doc.image is None:
+            return base_doc
+
+        # ✅ ROI-aware, CACHED preview doc
+        vw = self._active_view_widget()
+        if vw and hasattr(vw, "has_active_preview") and vw.has_active_preview():
+            try:
+                roi_doc = self.get_document_for_view(vw)  # <-- uses _roi_doc_cache
+                if isinstance(roi_doc, _RoiViewDocument):
+                    try:
+                        name_suffix = f" (Preview {vw.current_preview_name() or ''})"
+                        roi_doc.metadata["display_name"] = f"{base_doc.display_name()}{name_suffix}"
+                    except Exception:
+                        pass
+                return roi_doc
+            except Exception:
+                return base_doc
+
+        return base_doc
+
+
+
+    def update_active_document(self, updated_image, metadata=None, step_name: str = "Edit"):
+
+        view_doc = self.get_active_document()
+        if view_doc is None:
+            raise RuntimeError("No active document")
+
+        old_img = getattr(view_doc, "image", None)
+        old_shape = getattr(old_img, "shape", None)
+
+        img = np.asarray(updated_image)
+        if img.dtype != np.float32:
+            img = img.astype(np.float32, copy=False)
+
+        _debug_log_undo(
+            "DocManager.update_active_document.entry",
+            step_name=step_name,
+            view_doc_type=type(view_doc).__name__,
+            view_doc_id=id(view_doc),
+            is_roi=isinstance(view_doc, _RoiViewDocument),
+            old_shape=old_shape,
+            new_shape=getattr(img, "shape", None),
+        )
+
+        # --- Extract operation parameters (if any) from metadata --------
+        md = dict(metadata or {})
+        op_params = md.pop("__op_params__", None)
+
+        # If this is an ROI view doc, keep track of where this happened
+        roi_tuple = None
+        source_kind = "full"
+        if isinstance(view_doc, _RoiViewDocument):
+            roi_tuple = getattr(view_doc, "_roi", None)
+            source_kind = "roi"
+
+        # --- ROI preview branch: only update preview, no parent paste ----
+        if isinstance(view_doc, _RoiViewDocument):
+            # Update ONLY the preview; view repaint is driven by signals
+            view_doc.apply_edit(img, md, step_name)
+
+            # Record operation on the ROI doc itself
+            if hasattr(view_doc, "record_operation"):
+                try:
+                    view_doc.record_operation(
+                        step_name=step_name,
+                        params=op_params,
+                        roi=roi_tuple,
+                        source=source_kind,
+                    )
+                except Exception:
+                    pass
+            _debug_log_undo(
+                "DocManager.update_active_document.roi_after",
+                step_name=step_name,
+                view_doc_id=id(view_doc),
+                roi=getattr(view_doc, "_roi", None),
+                pundo_len=len(getattr(view_doc, "_pundo", [])),
+                predo_len=len(getattr(view_doc, "_predo", [])),
+            )
+            return
+
+        # --- Full image branch ------------------------------------------
+        if isinstance(view_doc, ImageDocument):
+            view_doc.apply_edit(img, md, step_name)
+            try:
+                self.imageRegionUpdated.emit(view_doc, None)
+            except Exception:
+                pass
+
+            _debug_log_undo(
+                "DocManager.update_active_document.full_after",
+                step_name=step_name,
+                view_doc_id=id(view_doc),
+                undo_len=len(getattr(view_doc, "_undo", [])),
+                redo_len=len(getattr(view_doc, "_redo", [])),
+                final_shape=getattr(view_doc.image, "shape", None),
+            )
+            # Record operation on the full document
+            if hasattr(view_doc, "record_operation"):
+                try:
+                    view_doc.record_operation(
+                        step_name=step_name,
+                        params=op_params,
+                        roi=None,
+                        source=source_kind,
+                    )
+                    
+                except Exception:
+                    pass
+        else:
+            raise RuntimeError("Active document is not an image")
+
+    def get_active_operation_log(self) -> list[dict]:
+        """
+        Return the operation log for the *currently active* document-like
+        (full image or ROI-preview). Empty list if none.
+        """
+        doc = self.get_active_document()
+        if doc is None:
+            return []
+        get_log = getattr(doc, "get_operation_log", None)
+        if callable(get_log):
+            try:
+                return get_log()
+            except Exception:
+                return []
+        return []
+
+
+
+    # Back-compat/aliases so tools can call any of these:
+    def update_image(self, updated_image, metadata=None, step_name: str = "Edit"):
+        self.update_active_document(updated_image, metadata, step_name)
+
+    def set_image(self, img, metadata=None, step_name: str = "Edit"):
+        self.update_active_document(img, metadata, step_name)
+
+    def apply_edit_to_active(self, img, step_name: str = "Edit", metadata=None):
+        self.update_active_document(img, metadata, step_name)