The unordered_map with 2768 string keys (~100KB+) was causing OOM crashes
at beginTocPass() on ESP32-C3's limited ~380KB RAM.
Reverted createTocEntry() to use original O(n) spine file scan instead.
Kept the safe spineToTocIndex vector in buildBookBin() (only ~5.5KB).
Replace O(n²) lookups with O(n) preprocessing:
1. createTocEntry(): Build href->spineIndex map once in beginTocPass()
instead of scanning spine file for every TOC entry
2. buildBookBin(): Build spineIndex->tocIndex vector in single pass
instead of scanning TOC file for every spine entry
For 2768-chapter EPUBs, this reduces:
- TOC pass: from ~7.6M file reads to ~5.5K reads
- buildBookBin: from ~7.6M file reads to ~5.5K reads
Memory impact: ~80KB for href map (acceptable trade-off for 10x+ speedup)
Remove the call to loadAllFileStatSlims() which pre-loads all ZIP central
directory entries into memory. For EPUBs with 2000+ chapters (like webnovels),
this exhausts the ESP32-C3's ~380KB RAM and causes abort().
The existing loadFileStatSlim() function already handles individual lookups
by scanning the central directory per-file when the cache is empty. This is
O(n*m) instead of O(n), but prevents memory exhaustion.
Fixes#134
* origin:
fix: truncate chapter names that are too long (#422)
feat: dict based Hyphenation (#305)
fix: render U+FFFD replacement character instead of ? (#366)
fix: Invert colors on home screen cover overlay when recent book is selected (#390)
Adds KOReader Sync support (#232)
feat: Change keyboard "caps" to "shift" & Wrap Keyboard (#377)
fix: XTC 1-bit thumb BMP polarity inversion (#373)
## Summary
* Adds (optional) Hyphenation for English, French, German, Russian
languages
## Additional Context
* Included hyphenation dictionaries add approximately 280kb to the flash
usage (German alone takes 200kb)
* Trie encoded dictionaries are adopted from hypher project
(https://github.com/typst/hypher)
* Soft hyphens (and other explicit hyphens) take precedence over
dict-based hyphenation. Overall, the hyphenation rules are quite
aggressive, as I believe it makes more sense on our smaller screen.
---------
Co-authored-by: Dave Allie <dave@daveallie.com>
## Summary
- Adds KOReader progress sync integration, allowing CrossPoint to sync
reading positions with other
KOReader-compatible devices
- Stores credentials securely with XOR obfuscation
- Uses KOReader's partial MD5 document hashing for cross-device book
matching
- Syncs position via percentage with estimated XPath for compatibility
# Features
- Settings: KOReader Username, Password, and Authenticate options
- Sync from chapters menu: "Sync Progress" option appears when
credentials are configured
- Bidirectional sync: Can apply remote progress or upload local progress
---------
Co-authored-by: Dave Allie <dave@daveallie.com>
Let's start small by showing the ALT text of IMG. This is rudimentary,
but avoids those otherwise completely blank chapters.
I feel we will need this even when we can render images if that
rendering takes >1s - I would then prefer rendering optional and showing
the ALT text first.
## Summary
* **What is the goal of this PR?**: Fix the bug I reported in
https://github.com/daveallie/crosspoint-reader/issues/292
* **What changes are included?**: Instead of silently dropping table
content in EPUBs., replace with an italicized '[Table omitted]' message
where tables appear.
## Additional Context
* Add any other information that might be helpful for the reviewer
(e.g., performance implications, potential risks,
specific areas to focus on).
---
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**PARTIALLY **_
---------
Co-authored-by: Evan Fenner <evan@evanfenner.com>
Co-authored-by: Warp <agent@warp.dev>
Currently there is no visual indication whatsoever if something is in a
list. An `<li>` is essentially just another paragraph.
As a partial remedy for this, add a bullet character to the beginning of
`<li>` text blocks so that the user can see that they're list items.
This is incomplete in that an `<ol>` should also have a counter so that
its list items can get numbers instead of bullets (right now I don't
think we track if we're in a `<ul>` or an `<ol>` at all), but it's
strictly better than the current situation.
Co-authored-by: Maeve Andrews <maeve@git.mail.maeveandrews.com>
Currently, when Extra Paragraph Spacing is off, an em-space is added to
the beginning of each ParsedText even for blocks like headers that are
centered. This whitespace makes the centering slightly off. Change the
calculation here to only add the em-space for left/justified text.
Co-authored-by: Maeve Andrews <maeve@git.mail.maeveandrews.com>
## Summary
Skip BOM character (sometimes used in front of em-dashes) - they are not
part of the glyph set and would render `?` otherwise.
---
### AI Usage
Did you use AI tools to help write this code? _**YES**_
## Summary
* **What is the goal of this PR?** (e.g., Fixes a bug in the user
authentication module,
Display the book cover image in the **"Continue Reading"** card on the
home screen, with fast navigation using framebuffer caching.
* **What changes are included?**
- Display book cover image in the "Continue Reading" card on home screen
- Load cover from cached BMP (same as sleep screen cover)
- Add framebuffer store/restore functions (`copyStoredBwBuffer`,
`freeStoredBwBuffer`) for fast navigation after initial render
- Fix `drawBitmap` scaling bug: apply scale to offset only, not to base
coordinates
- Add white text boxes behind title/author/continue reading label for
readability on cover
- Support both EPUB and XTC file cover images
- Increase HomeActivity task stack size from 2048 to 4096 for cover
image rendering
## Additional Context
* Add any other information that might be helpful for the reviewer
(e.g., performance implications, potential risks, specific areas to
focus on).
- Performance: First render loads cover from SD card (~800ms),
subsequent navigation uses cached framebuffer (~instant)
- Memory: Framebuffer cache uses ~48KB (6 chunks × 8KB) while on home
screen, freed on exit
- Fallback: If cover image is not available, falls back to standard
text-only display
- The `drawBitmap` fix corrects a bug where screenY = (y + offset) scale
was incorrectly scaling the base coordinates. Now correctly uses screenY
= y + (offset scale)
## Summary
- Nav file in EPUB 3 file is a HTML file with relative hrefs
- If this file exists anywhere but in the same location as the
content.opf file, navigating in the book will fail
- Bump the book cache version to rebuild potentially broken books
## Additional Context
- Fixes https://github.com/daveallie/crosspoint-reader/issues/264
---
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code?
- [ ] Yes
- [ ] Partially
- [x] No
## Summary
* Remove HTML entity parsing
* This has been completely useless since the introduction of expat
* expat tries to parse all entities in the document, but only knows of
HTML ones
* Parsing will never end with HTML entities in the text, so the
additional step to parse them that we had went completely unused
* We should figure out the best way to parse that content in the future,
but for now remove that module as it generates a lot of heap allocations
with its map and strings
## Summary
* **What is the goal of this PR?** Add EPUB 3 support by implementing
native navigation document (nav.xhtml) parsing with NCX fallback,
addressing issue Fixes: #143.
* **What changes are included?**
- New `TocNavParser` for parsing EPUB 3 HTML5 navigation documents
(`<nav epub:type="toc">`)
- Detection of nav documents via `properties="nav"` attribute in OPF
manifest
- Fallback logic: try EPUB 3 nav first, fall back to NCX (EPUB 2) if
unavailable
- Graceful degradation: books without any TOC now load with a warning
instead of failing
## Additional Context
* The implementation follows the existing streaming XML parser pattern
using Expat to minimize RAM usage on the ESP32-C3
* EPUB 3 books that include both nav.xhtml and toc.ncx will prefer the
nav document (per EPUB 3 spec recommendation)
* No breaking changes - existing EPUB 2 books continue to work as before
* Tested on examples from
https://idpf.github.io/epub3-samples/30/samples.html
## Summary
* **What is the goal of this PR?**
Add a new user setting for paragraph alignment, instead of hard-coding
full justification.
* **What changes are included?**
One new line in the settings screen, with 4 options
(justify/left/center/right). Default is justified since that's what it
was already. I personally only wanted to disable justification and use
"left", but I included the other options for completeness since they
were already supported.
## Additional Context
Tested on my X4 and looks as expected for each alignment.
Co-authored-by: Maeve Andrews <maeve@git.mail.maeveandrews.com>
## Summary
* Update EpdFontFamily::Style to be u8 instead of u32 (saving 3 bytes
per word)
* Update layout width/height to be u16 from int
* Update page element count to be u16 from u32
* Update text block element count to be u16 from u32
* Bumped section bin version to version 8
## Summary
* Redesigned home screen with big option to continue reading and
slightly nicer options to navigate to core sections
* Attempt to use the cached EPUB details (title, author) if they exist,
otherwise fall back to file name
* Adjusted button hints on home screen, removed Back option and changed
left/right to up/down
## Additional Context
* Core of this work comes from @ChandhokTannay in
1d36a86ef1
This parses the guide section in the content.opf for text/start
references and jumps to this on first open of the book.
Currently, this behavior will be repeated in case the reader manually
jumps to Chapter 0 and then re-opens the book. IMO, this is an
acceptable edge case (for which I couldn't see a good fix other than to
drag a "first open" boolean around).
---------
Co-authored-by: Sam Davis <sam@sjd.co>
Co-authored-by: Dave Allie <dave@daveallie.com>
## Summary
* Show previous title for unnamed spines
* The spec is a little unclear, but there are plenty of cases where
chapters are split up in parts and should show the previous chapter's
title
* List TOC items instead of spine items in chapter select
* Bump `BOOK_CACHE_VERSION` to `2` to force regeneration of spine item's
TOC indexes
## Summary
* Swap to updated SDCardManager which uses SdFat
* Add exFAT support
* Swap to using FsFile everywhere
* Use newly exposed `SdMan` macro to get to static instance of
SDCardManager
* Move a bunch of FsHelpers up to SDCardManager
## Summary
* Consolidate chapter page data into single file
* Header structure of the file stays the same, following the page count,
we now put a LUT offset
* The page data is all then appended to this file
* Finally the LUT is appended to the end of the file, and the page count
is updated
* This will also significantly improve the duration of cache cleanup
which takes a while to scan the directory and cleanup content
* Remove page file version as it's all tied up into the section file now
* Bumped section file version to 7
* Moved section content into sub directory
* Updated docs
## Additional Context
* Benchmarks:
* Generating 74 pages of content from a chapter in Jade Legacy took:
* master: 6,229ms
* this PR: 1,305ms
* Speedup of 79%
* Generating 207 pages of content from Livesuit book:
* With progress bar UI updates:
* master: 24,250ms
* this PR: 8,063ms
* Speedup of 67%
* Without progress bar UI updates:
* master: 13,055ms
* this PR: 3,600ms
* Speedup of 72%
• What is the goal of this PR?
Implement a horizontal EPUB reading mode so books can be read in
landscape orientation (both 90° and 270°), while keeping the rest of the
UI in portrait.
• What changes are included?
◦ Rendering / Display
▪ Added an orientation model to GfxRenderer (Portrait, LandscapeNormal,
LandscapeFlipped) and made:
▪ drawPixel, drawImage, displayWindow map logical coordinates
differently depending on orientation.
▪ getScreenWidth() / getScreenHeight() return orientation‑aware logical
dimensions (480×800 in portrait, 800×480 in landscape).
◦ Settings / Configuration
▪ Extended CrossPointSettings with:
▪ landscapeReading (toggle for portrait vs. landscape EPUB reading).
▪ landscapeFlipped (toggle to flip landscape 180° so both horizontal
holding directions are supported).
▪ Updated settings serialization/deserialization to persist these fields
while remaining backward‑compatible with existing settings files.
▪ Updated SettingsActivity to expose two new toggles:
▪ “Landscape Reading”
▪ “Flip Landscape (swap top/bottom)”
◦ EPUB Reader
▪ In EpubReaderActivity:
▪ On onEnter, set GfxRenderer orientation based on the new settings
(Portrait, LandscapeNormal, or LandscapeFlipped).
▪ On onExit, reset orientation back to Portrait so Home, WiFi, Settings,
etc. continue to render as before.
▪ Adjusted renderStatusBar to position the status bar and battery
indicator relative to GfxRenderer::getScreenHeight() instead of
hard‑coded Y coordinates, so it stays correctly at the bottom in both
portrait and landscape.
◦ EPUB Caching / Layout
▪ Extended Section cache metadata (section.bin) to include the logical
screenWidth and screenHeight used when pages were generated; bumped
SECTION_FILE_VERSION.
▪ Updated loadCacheMetadata to compare:
▪ font/margins/line compression/extraParagraphSpacing and screen
dimensions; mismatches now invalidate and clear the cache.
▪ Updated persistPageDataToSD and all call sites in EpubReaderActivity
to pass the current GfxRenderer::getScreenWidth() / getScreenHeight() so
portrait and landscape caches are kept separate and correctly sized.
Additional Context
• Cache behavior / migration
◦ Existing section.bin files (old SECTION_FILE_VERSION) will be detected
as incompatible and their caches cleared and rebuilt once per chapter
when first opened after this change.
◦ Within a given orientation, caches will be reused as before. Switching
orientation (portrait ↔ landscape) will cause a one‑time re‑index of
each chapter in the new orientation.
• Scope and risks
◦ Orientation changes are scoped to the EPUB reader; the Home screen,
Settings, WiFi selection, sleep screens, and web server UI continue to
assume portrait orientation.
◦ The renderer’s orientation is a static/global setting; if future code
uses GfxRenderer outside the reader while a reader instance is active,
it should be aware that orientation is no longer implicitly fixed.
◦ All drawing primitives now go through orientation‑aware coordinate
transforms; any code that previously relied on edge‑case behavior or
out‑of‑bounds writes might surface as logged “Outside range” warnings
instead.
• Testing suggestions / areas to focus on
◦ Verify in hardware:
▪ Portrait mode still renders correctly (boot, home, settings, WiFi,
reader).
▪ Landscape reading in both directions:
▪ Landscape Reading = ON, Flip Landscape = OFF.
▪ Landscape Reading = ON, Flip Landscape = ON.
▪ Status bar (page X/Y, % progress, battery icon) is fully visible and
aligned at the bottom in all three combinations.
◦ Open the same book:
▪ In portrait first, then switch to landscape and reopen it.
▪ Confirm that:
▪ Old portrait caches are rebuilt once for landscape (you should see the
“Indexing…” page).
▪ Progress save/restore still works (resume opens to the correct page in
the current orientation).
◦ Ensure grayscale rendering (the secondary pass in
EpubReaderActivity::renderContents) still looks correct in both
orientations.
---------
Co-authored-by: Dave Allie <dave@daveallie.com>
## Summary
* **What is the goal of this PR?**
Improve reliability and user experience during chapter indexing by
adding retry logic for SD card operations and a visual progress bar.
* **What changes are included?**
- **Retry logic**: Add 3 retry attempts with 50ms delay for ZIP to SD
card streaming to handle timing issues after display refresh
- **Progress bar**: Display a visual progress bar (0-100%) during
chapter indexing based on file read progress, updating every 10% to
balance responsiveness with e-ink display limitations
## Additional Context
* **Problem observed**: When navigating quickly through books with many
chapters (before chapter titles finish rendering), the "Indexing..."
screen would appear frozen. Checking the serial log revealed the
operation had silently failed, but the UI showed no indication of this.
Users would likely assume the device had crashed. Pressing the next
button again would resume operation, but this behavior was confusing and
unexpected.
* **Solution**:
- Retry logic handles transient SD card timing failures automatically,
so users don't need to manually retry
- Progress bar provides visual feedback so users know indexing is
actively working (not frozen)
* **Why timing issues occur**: After display refresh operations, there
can be timing conflicts when immediately starting SD card write
operations. This is more likely to happen when rapidly navigating
through chapters.
* **Progress bar design**: Updates every 10% to avoid excessive e-ink
refreshes while still providing meaningful feedback during long indexing
operations (especially for large chapters with CJK characters).
* **Performance**: Minimal overhead - progress calculation is simple
byte counting, and display updates use `FAST_REFRESH` mode.
## Summary
* **What is the goal of this PR?** Fixes a bug where text disappears
after approximately 25 pages in long chapters during EPUB indexing.
* **What changes are included?**
- Removed the `MAX_LINES = 1000` hard limit in
`ParsedText::computeLineBreaks()`
- Added safer infinite loop prevention by checking if `nextBreakIndex <=
currentWordIndex` and forcing advancement by one word when stuck
## Additional Context
* **Root cause:** The `MAX_LINES = 1000` limit was introduced to prevent
infinite loops, but it truncates content in long chapters. For example,
a 93KB chapter that generates ~242 pages (~9,680 lines) gets cut off at
~1000 lines, causing blank pages after page 25-27.
* **Solution approach:** Instead of a hard line limit, I now detect when
the line break algorithm gets stuck (when `nextBreakIndex` doesn't
advance) and force progress by moving one word at a time. This preserves
the infinite loop protection while allowing all content to be rendered.
* **Testing:** Verified with a Korean EPUB containing a 93KB chapter -
all 242 pages now render correctly without text disappearing.
## Summary
* Use single unified cache file for book spine, table of contents, and
core metadata (title, author, cover image)
* Use new temp item store file in OPF parsing to store items to be
rescaned when parsing spine
* This avoids us holding these items in memory
* Use new toc.bin.tmp and spine.bin.tmp to build out partial toc / spine
data as part of parsing content.opf and the NCX file
* These files are re-read multiple times to ultimately build book.bin
## Additional Context
* Spec for file format included below as an image
* This should help with:
* #10
* #60
* #99
## Summary
* Fix incorrect justification of last line in paragraph
* `words` is changing size due to the slice, so `isLastLine` would
rarely be right, either removing justification mid-paragraph, or
including it in the last line.
## Additional Context
* Introduced in #73
## Summary
* Extract EPUB TOC into temp file before parsing
* Streaming ZIP -> XML parser uses up a lot of memory as we're
allocating inflation buffers while also holding a few copies of the
buffer in different forms
* Instead, but streaming the inflated file down to the SD card (like we
do for HTML parsing, we can lower memory usage)
## Additional Context
* This should help with
https://github.com/daveallie/crosspoint-reader/issues/60 and
https://github.com/daveallie/crosspoint-reader/issues/10. It won't
remove those class of issues completely, but will allow for many more
books to be opened.
Still a bit raw, but gets the time required to determine the size of
each chapter (for reading progress) down from ~25ms to 0-1ms.
This is done by keeping the zipArchive open (so simple ;)).
Probably we don't need to cache the spine sizes anymore then...
---------
Co-authored-by: Dave Allie <dave@daveallie.com>
## Summary
* Build out lines for pages when holding over 750 buffered words
* Should fix issues with parsing long blocks of text causing memory
crashes
## Problem
Three Epub getter functions can throw exceptions:
- `getCumulativeSpineItemSize()`: No bounds check before
`.at(spineIndex)`
- `getSpineItem()`: If spine is empty and index invalid, `.at(0)` throws
- `getTocItem()`: If toc is empty and index invalid, `.at(0)` throws
## Fix
- Add bounds check to `getCumulativeSpineItemSize()`, return 0 on error
- Add empty container checks to `getSpineItem()` and `getTocItem()`
- Use static fallback objects for safe reference returns on empty
containers
Changed `lib/Epub/Epub.cpp`.
## Test
- Defensive additions - follows existing bounds check patterns
- No logic changes for valid inputs
- Manual device testing appreciated
## Problem
- `getBookSize()` calls `getCumulativeSpineItemSize(getSpineItemsCount()
- 1)` which passes -1 when spine is empty
- `calculateProgress()` then divides by zero when book size is 0
## Fix
- Return 0 from `getBookSize()` if spine is empty
- Return 0 from `calculateProgress()` if book size is 0
## Testing
- Builds successfully with `pio run`
- Affects: `lib/Epub/Epub.cpp`
