[vitorjordao]

Fixes #8.

eeprom_write_bytes reads a block back before a partial write only when the write starts inside it, or ends inside a different last block. A write that starts on a block boundary and ends inside the same block skips the read-back. Example: eeprom_write_bytes(src, 136, 3), i.e. block 17, bytes 0..2. If that block isn't cached yet, bytes 3..7 come from uninitialized cache memory, and the flush writes them to the EEPROM.

block 17   [ 10 | 11 | 12 | ?? | ?? | ?? | ?? | ?? ]   <- cache
             \__________/   \____________________/
             written (len=3) never fetched, flushed

The fix fetches the last block whenever the write ends inside it (end_offset != 0). When the last block is also the first one, eeprom_cache_ensure_block_valid returns early on the second call because the block is already valid.

Tests

Two new testrom tests, test_eeprom_read_bytes and test_eeprom_write_bytes. Each one goes through seven spans, written as (offset in the block, length):

• inside one block: (5,1), (1,3), (0,3)
• to the end of a block: (3,5)
• across a boundary: (6,4), (0,11)
• one whole block: (0,8)

For each span, the tests:

1. seed blocks the cache hasn't touched yet with known bytes, using raw Joybus commands;
2. call eeprom_read_bytes / eeprom_write_bytes, then eeprom_wait_idle;
3. check the hardware contents, again over raw Joybus.

They check that only the span changed, including the blocks on either side, and that a read writes nothing outside its destination. The tests skip without EEPROM, like test_eepromfs, so the ROM has to be built with a save type: make -C tests N64_ROM_SAVETYPE=eeprom4k.

ares v148, testrom_emu.z64, N64_ROM_SAVETYPE=eeprom4k.

Before the fix (f26e1c1 with the new tests):

**** Starting test: test_eeprom_read_bytes
**** Starting test: test_eeprom_write_bytes
write_bytes(start=77, len=1)
write_bytes(start=105, len=3)
write_bytes(start=136, len=3)

ASSERTION FAILED (test_eeprom.c:134):
[1112000000] != [11128b8c8d]
     ^^              ^^  idx: 11
write_bytes(136, 3) changed the wrong bytes

**** Starting test: test_eepromfs
...
Passed: 220 out of 222 (7 skipped)

After the fix (same result with eeprom16k):

**** Starting test: test_eeprom_read_bytes
**** Starting test: test_eeprom_write_bytes
**** Starting test: test_eepromfs
...
Passed: 221 out of 222 (7 skipped)

The remaining failure is test_timer_ticks. It fails the same way before and after this change, and it is unrelated.