mef3io 1.1.2

Released 2026-08-05

A bug-fix release. Thanks to @SweetVlad for the first two fixes below and the faster overlapping-block reads, found on long-term iEEG recordings written by other software.

Fixes

• Wrong sample values with parallel reads. In files from other recorders a data block can overlap the previous one by a few samples. Parallel decoding could then write those samples from two threads at once and return wrong values, differently from read to read. Reads are now identical for any thread count. Files written by mef3io were not affected.
• Valid sessions refused as corrupted. Some writers leave extra bytes at the end of a channel's metadata file (.tmet). mef3io included them in the checksum, rejected the file, and with it the whole session. These sessions now open; real damage inside the metadata is still reported.
• Text fields (units, descriptions, subject fields) that are too long are now shortened cleanly. Previously a long value could be cut in the middle of a non-ASCII character such as µ, and the session could no longer be opened.
• Annotation (record) types must be exactly four ASCII characters, such as "Note" or "EDFA", and a record carrying data its type cannot store is refused. Previously such a record could be written with its text silently dropped.
• Faster reads of files with heavily overlapping blocks: blocks whose samples are entirely covered by later blocks are no longer decoded.