A download built the book again each time. It now sends the kept build,
and builds only if the book changed after it. Stop tracking deletes the
built book. A cancelled import deletes its uploaded file, and the start
deletes the uploads that no job needs.
A job placed each new chapter with the chapter order from its start, so
a move by hand during the job went back. The store now places a new
chapter against the order that the story has at that moment, in one
transaction. A recheck of a chapter that the user deleted no longer
stops the job.
An error outside a job, such as a full disk, stopped the thread, and no
job ran until a restart. The health check now answers 503 when a job
thread stopped.
An unexpected error now sends a general message, and the log keeps the
text. A job no longer shows the path of its uploaded file. A cover is
cached only by the browser of the user.
A key of the chapters, rules and metadata of a story tells if its book
changed. A check without a change no longer builds the book again, also
for a book that waits for an upload. A held book goes out without a new
build. The Rebuild button still builds and sends in each case.
The story page already showed last check time. Add last update time
next to it, from the existing last_changed field.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The merge cannot always tell where new chapters go. For example, a site
can list an announcement before the chapters of an old book. You can now
select chapters and move them, in their order, to after another chapter
or to the start. You can also delete them with their revisions. A story
keeps at least one chapter. The book is then rebuilt. A deleted chapter
that the site still lists comes back with the next download.
The Move… and Delete… buttons open a dialog that names the selected
chapters before anything changes. A shift-click selects a range of
chapters.
The Select all checkbox moves into the first heading of the chapter
list. On a phone, that heading shows with the text "Select all".
Some revisions differ from others only by curly quotes, curly
apostrophes and the ellipsis. The script deletes those revisions and
keeps the one with the least smart punctuation. In revisions with other
text changes, it makes the punctuation straight and stores the new text
hash. It does not change the newest revision if it came from the site.
The script uses only the Python standard library, so you can run it on
the host against bindery.db. It is a dry run unless you give --apply,
and --apply makes a backup first.
A FanFicFare ePub records the site address of each chapter, so Bindery
can import it. Other ePubs, such as the author's own ebook of a story,
have no chapter links. The script asks which stored chapters come before
and after the gap, and which ePub chapters match them. It suggests
matches by title and puts the suggestion on the input line. It then adds
the ePub chapters between the matches, with their images, in the order
of the ePub.
Each added chapter gets a stable URL with the #bindery-import- marker
and the revision source "epub-import". A second run with the same ePub
does not add the same chapters again. --remove deletes the chapters that
the script added.
Each table had six columns, which squeezed down to unreadable widths on a
narrow screen and forced horizontal scrolling.
Below 600px, each row becomes a bordered card. The main value (the story
or chapter title) stays unlabelled at the top, and every other value gets
a small label in front of it, taken from its column heading.
The chapter table's "select all" checkbox lived in the table header,
which mobile now hides in favour of the per-row labels. It moves into
the sticky actions bar above the list instead, next to the buttons that
already act on the selection.
Each field was a row of a four-column table: name, lock, value and the
site's current value. On a narrow screen the columns squeezed the
inputs down to a few characters wide.
Each field is now a card with its name and lock on one line and the
input below at full width. The site's value shows only when it
differs, folded behind a line that opens on tap, instead of always
taking its own column. The lock is a labelled toggle instead of a
bare checkbox, and Save stays disabled until something changes.
On a wide screen, the label sits beside the field instead of above
it, and a single-line input no longer stretches to fill the card.
Plain HTML, CSS and JavaScript modules, with no build step. Hash routes
and relative URLs make it work under any path. The job list is polled:
every 2 seconds while a job runs, and less often otherwise.
Logins follow shelfmark: none, CWA accounts, or a reverse proxy header.
CWA accounts are checked against the OPDS feed, with a lockout after
repeated wrong passwords. An API key works in every mode, and changes
need an X-Requested-With header or the key. URL_BASE serves the app
under a path prefix.
Jobs download new chapters, recheck chapters, replace the book, update
metadata or rebuild from stored chapters. Locked metadata fields keep
their values. A scheduler queues updates. An update finds the library
book of a story through CWA, and fails with an error if it cannot, so
that it never makes a duplicate book.
Bindery mounts no CWA database. It searches the OPDS feed by title and
reads the story URL inside each candidate epub to find the library book.
Updates go to the ingest folder with a .cwa.json sidecar, so CWA replaces
the epub of the existing book.
FanFicFare adapters fetch metadata and chapters, and its epub writer
builds each book from the store. Environment variables configure the
app, and a default personal.ini is copied to the config folder on first
start.
The SQLite store keeps every chapter revision, so a chapter that the site
removes stays in the book. The merge puts new chapters after the chapter
that comes before them on the site. Diffs compare text, not HTML, and
replacement rules run on chapter HTML when a book is built.
tim
