English · Русский

Idea, requirements and all testing on real hardware — @Hinduc0der. Code written by Claude (Anthropic) under his direction.

A MIDlet for MIDP 2.0 / CLDC 1.1 that reads .md (and .txt) straight off the memory card and renders the markup itself. Targets the Nokia 5300 (240x320), Sony Ericsson K750 (176x220) and anything else with Java and JSR-75. Layout is derived from the screen size, so there are no per-resolution builds.

The interface comes in two languages, as two separate builds: dist/MDReader-en.jar + .jad and dist/MDReader-ru.jar + .jad. Every released version of both is kept in dist/versions/.

1. Pick your language and copy both files of that build, the .jar and the .jad, into the same folder on the memory card.
2. On the phone, open the .jad (on S40 — via Gallery / Applications) and accept the install.
3. The first time you open a file the phone will ask for permission to access data — that is JSR-75. To stop it asking every time, set the application permission Data access: ask once (on S40 — Applications → Options → Data access).

The MIDlet is unsigned, so the phone treats it as untrusted. That is expected; it is simply being careful about reading the card.

On touch devices: tap the top of the screen for the previous page, the bottom for the next one, the middle for the menu.

The menu has: table of contents (the file's headings), jump to a percentage, open another file, settings (font, theme, encoding, status bar, paragraph joining) and auto-scroll.

Reading position and settings are stored in RMS; the last 20 books are remembered by path. On startup the last book reopens exactly where you stopped.

Headings #…######, **bold**, *italic*, `code`, fenced ``` blocks and 4-space-indented blocks, > blockquotes, `-`/`*`/`+` and numbered lists, `---` rules, links `text` and images — the caption is shown. Tables and HTML are rendered as plain text.

Encoding is detected automatically (UTF-8 versus Windows-1251) and can be forced in the settings. The decoder is hand-written: the system one on S40 cannot be trusted.

You need JDK 8, WTK 2.5.2 and, for a release build, ProGuard. Paths are set at the top of build.bat.

build.bat                plain build, Russian, for the emulator
build.bat en             plain build, English
build.bat release        release of BOTH languages + version archive
build.bat release en     release of one language only

The output is dist/MDReader-<lang>.jar and .jad — always the latest build, overwritten every time. On top of that, every release build is archived:

dist/versions/MDReader-0.1.2.jar        + .jad    (up to 0.1.2 — Russian only)
dist/versions/MDReader-0.2.0-ru.jar    + .jad
dist/versions/MDReader-0.2.0-en.jar    + .jad

In an archived pair the .jad points at its own versioned .jar, so any old version can be installed by copying that pair into one folder. Debug builds are not archived — the archive only holds what actually went onto a phone.

Bump the version in build.bat (set VERSION=) before any build you put on a phone: otherwise the phone may not notice the update, and the previous archive entry gets overwritten (the build warns about this). What changed in each version is in CHANGELOG.md.

"C:\WTK2.5.2_01\bin\emulator.exe" -Xdevice:DefaultColorPhone -Xdomain:maximum ^
  -classpath build\pre mdreader.MDReaderMIDlet

-Xdevice:DefaultGrayPhone gives a 176x220 screen, like the K750. The emulator reads files from %USERPROFILE%\j2mewtk\2.5.2\appdb\temp.<device>N\filesystem\root1\ (launched via -classpath that directory is temporary and recreated each run). For a more honest picture, KEmulator nnmod works too.

Be warned: the emulator uses desktop Windows fonts and lies about memory. Two bugs in 0.1.1 were invisible there and only showed up on a real Nokia.

The file is never read into memory as a whole: a page is identified by a pair of «line offset + visual line number inside it», and Book keeps an 8 KB window. On a K750 with its one megabyte of heap that is not optional.

test/RootsTest.java runs on desktop Java and covers the root-merging logic:

javac -d build\test src\mdreader\Roots.java test\RootsTest.java
java -cp build\test RootsTest

CLDC 1.1 has no resource bundles, so the language is chosen at build time: build.bat hands javac the Strings.java for the language being built, out of i18n/ru/ or i18n/en/. Every user-visible string is a static final field there, which means ProGuard inlines them and the class disappears from the JAR — the English build carries no Russian text and vice versa.

Both files must declare the same set of fields. If one language is missing a string, that language simply fails to compile, which is exactly the check we want.

To add a language, copy i18n/en/mdreader/Strings.java to i18n/<code>/mdreader/, translate the values, and run build.bat release <code>.

MIT, see LICENSE.

The whole thing was built through conversation: a human sets the task, defines the constraints (target devices, memory limits, the old ReadManiac as the reference point) and tests the builds on real phones; Claude writes the code, drives the build and works out what the findings mean. Some things would never have surfaced otherwise — Symbian listed every drive twice in the file browser, and the auto-scroll marker rendered as an empty box. The emulator reproduces neither; both are only visible on an actual Nokia. The log of those findings is in CHANGELOG.md.