Rework the reader around a clean BookInfo interface and lazy chapter loading.
Types/parser:
- BookInfo now splits into loadStructure (cheap: spine refs + TOC tree) and renderChapter (one chapter on demand). EpubEnv carries the parsed archive.
- resolveZipPath collapses ./.. and strips #fragments (normalise does not), so relative hrefs like ../Images/x.jpg resolve; used for chapters and images.
- decodeEntry tolerates encodings: UTF-8/UTF-16 BOMs, Latin-1 fallback instead of throwing on invalid bytes.
- Skip spine items with no manifest entry instead of throwing an error.
- Scan the zip for a .opf if container.xml is missing.
- Image inlining rewrites the src URL string, not the whole <img> tag (scalpel
re-serialises tags, so a whole-tag replace silently missed self-closing imgs).
Navigation:
- Hierarchical TOC: ncxToc (EPUB 2), navToc (EPUB 3), guideToc (<guide> fallback). Nested scrapers use atDepth 1 to take only direct children.
UI:
- Streaming: render the opened chapter on demand, cache it, preload neighbours so arrow paging is instant with a placeholder that covers uncached jumps.
- Sidebar shows the TOC tree with CSS styling. Clicks scroll to #fragments.
- wrapHtml wraps each body in a full HTML document for reliable WebKit render.
Test: 'svitak-test <book> <idx>' dumps one rendered chapter. cabal test stanza fixed to list all needed modules/deps.
Marko Andjelic
2011d638b0