[gscan2pdf-announce] scantpaper v4.0.2 released

[Jeff <jf...@po...>]

Changes compared to v4.0.1:

* Fix a start-up warning for legacy boolean settings such as
   `post_save_hook` that gscan2pdf 2.x stored as an empty string for
   false. The value is now corrected and logged instead of being reported
   on every start, and the corrected `false` is written back on exit.
* A scan batch that ends because the document feeder is empty is no
   longer logged as an error with a traceback. An empty feeder is the
   normal end of a batch, so it is now recorded at INFO level; genuine
   scan errors still log at ERROR with their traceback.
* A saved PDF no longer silently re-encodes a lossless page image as
   JPEG during the PDF/A conversion. When the "Flate" option is used, or
   "G4" (or Automatic resolves a page to Group 4 or Flate), the image now
   stays Flate or Group 4 in the final file, on every supported OCRmyPDF
   version. A save whose pages are all JPEG keeps them as JPEG, so an
   already-lossy document is not inflated.
* A posterised grayscale page - a scan whose pixels use only a few
   distinct intensity levels, such as a document with a half-tone or
   line-art picture - is now stored and saved losslessly. On import it is
   kept as PNG instead of being re-encoded to JPEG, and "Automatic" PDF
   compression writes it with lossless Flate rather than JPEG. JPEG's
   transform assumes smooth tone, so for these pages it both smeared the
   levels and produced a larger file; the same 8-level page now saves
   smaller and without loss. Genuinely continuous-tone grayscale and
   colour pages still use JPEG.
* The misleading "None" option is gone from the PDF compression menu. It
   never produced uncompressed output: it resolved to exactly the same
   lossless Flate stream as the "Flate" option, so it promised something
   it did not deliver. A configuration that had selected it now loads as
   "Flate", keeping the same output with an honest label.
* The OCR "Advanced Tesseract engine parameters" field now says what it
   wants: engine parameters in `name=value` form - the values of
   tesseract's `-c` option - not command-line options. A pasted option
   such as `--user-words` or `--dpi` is refused with the equivalent
   parameter named (`user_words_file`, `user_defined_dpi`), and a leading
   `-c` is accepted so a line copied from tesseract's documentation
   works. A `user_words_file` or `user_patterns_file` that does not exist
   or cannot be read is reported before OCR starts instead of failing
   silently, and the resolved paths that will be applied are written to
   the log.
* The OCR settings now include a **Page segmentation mode** and an **OCR
   engine mode** selector, alongside the language. The segmentation mode
   tells tesseract how to lay out the page (Automatic, Single column,
   Single block, Single line, Single word, Single character, Sparse text,
   or Raw line), and the engine mode chooses the engine (Default, LSTM
   only, Legacy, or Combined). Both default to the previous behaviour and
   are stored with the other OCR options. Choosing an engine mode the
   installed language data cannot provide no longer fails with a
   misleading "invalid tessdata path"; the error names the engine mode
   instead. Because these two modes are set by the new controls,
   `tessedit_pageseg_mode` and `tessedit_ocr_engine_mode` are now
   rejected in the advanced field with a message pointing to the
   controls.
* Fixed about dialog
* The OCR options now include a **Tessdata directory** control, so a
   `tessdata_best`, `tessdata_fast`, or per-user language data set can be
   chosen without replacing the system models or exporting
   `TESSDATA_PREFIX` before launch. Leave it empty for the automatic
   directory; **Browse** picks a folder and **Default** clears the
   choice. The directory is resolved as the chosen path first, then
   `TESSDATA_PREFIX`, then the system default, and it must directly
   contain `.traineddata` files. A `TESSDATA_PREFIX` that names a broad
   parent such as `/usr/share` is now rejected with an explanatory error
   instead of making tesseract search the whole tree and hang while
   listing languages. Changing the directory rebuilds the language list
   and keeps the current language, marked "(not installed)" when the new
   directory lacks it. The selection is stored with the other OCR
   options. Tesseract 3's removed `m_data_sub_dir` parameter remains
   rejected.
* The OCR **Advanced Tesseract engine parameters** field and the text-
   layer editor now let **Tab** move focus to the next control, so OK (or
   Scan) stays reachable without the mouse; **Ctrl+Tab** and
   **Ctrl+Shift+Tab** move focus forward and backward. A literal tab
   character is no longer entered in these fields.
* The OCR dialog now returns to the height it had before the **Advanced
   Tesseract engine parameters** expander was opened when that expander
   is collapsed. The **Page Range** frame also keeps its natural height
   instead of stretching to fill surplus window height in any dialog that
   shows it.
* Rows in the scan, Clean up and Preferences dialogs now keep the height
   their content needs when the window is taller than the page. The
   surplus appears as empty space below the last row instead of
   stretching a frame, a checkbox, or a label-and-control row. This keeps
   the scan **# Pages** and **Source document** frames, the rotation
   controls, the Clean up options and the Preferences **General options**
   page at their natural heights.
* Dialog content no longer runs flush against the window edge, and rows
   are no longer packed together. GTK4 dropped the content-area inset
   that GTK3 derived from the theme, and a box's spacing now defaults to
   zero, so every dialog sets a consistent inset and spacing - including
   the ones opened from the Tools menu and the ones built directly on
   `Gtk.Dialog`.
* Dialog action buttons no longer touch each other or the window edge.
   GTK3's theme supplied the action area's button spacing and border;
   GTK4's action area is a plain box with zero spacing, so the app now
   applies a consistent inset and gap to every dialog with separate
   action buttons, including file choosers. A message dialog keeps its
   joined button bar.
* Clicking a word in the text or annotation layer now selects the word
   under the pointer on HiDPI displays and on pages whose horizontal and
   vertical resolutions differ. The click was mapped with an older
   transform than the one the layer is drawn with, so it could land on a
   neighbouring word or on the page, and a correction only began after
   clicking several times.
* A PDF whose text layer mixes upright and rotated slices now saves
   every word at its own size, so searching a word highlights the word
   found rather than a large surrounding block. A single tall rotated
   slice used to widen the grouping tolerance for the whole flat layer,
   which merged upright words from different rows into one oversized line
   and made the renderer draw them at that line's scale.
* The scan dialog now opens with every saved postprocessing setting. A
   saved alternate-rotation, image clean-up or user-defined-tool setting,
   a missing clean-up tool, or an empty user-defined-tool list no longer
   raises a `TypeError` while the dialog is being built. Applying
   Preferences after removing the last tool also leaves an open scan
   dialog usable.
* Restoring an empty crashed session no longer fails on the first scan
   with an internal error. It opens as a usable empty document, and the
   imported page becomes its first undo step. Reopening a session whose
   latest action removed every page now stays empty instead of showing
   the deleted pages again; undo still brings them back.
* Selecting a flatbed source now starts **# Pages** at one page instead
   of inheriting the ADF `All` default. An unbounded flatbed batch is
   only possible after choosing `All` explicitly, and the dialog warns
   that scanning continues until cancelled.
* Cancelling a scan between pages no longer queues the next page. If the
   scanner does not respond to cancel, scantpaper reports the failure and
   rebuilds the device session for the next scan; a device still held by
   the operating system is reported as busy instead of leaving the
   application unable to scan.

Available on Github:

https://github.com/carygravel/scantpaper/releases/tag/v4.0.2

Or via the PPA:

https://launchpad.net/~jeffreyratcliffe/+archive/ubuntu/ppa