Resolved: hhsuitedb.py UTF-8 decode crash + _cs219 misses from long/complex names

[Jiachen2000]

Status

Resolved on our side — fixed by (1) making ffindex.py/hhsuitedb.py use binary I/O with latin-1 and no name truncation, and (2) normalizing a few problematic, overly long A3M basenames.

Original Symptoms

UnicodeDecodeError in ffindex.read_index() when hhsuitedb.py reads .ffindex.
Successful _a3m/_hhm but missing _cs219 files; later hhsearch -d <db> fails with could not open ..._cs219.ffdata.
Sporadic failures tied to very long/complex basenames (multiple double underscores, long family names).

Root Causes

1. Scripts treated .ffindex/.ffdata as UTF-8 text; they’re effectively binary tab tables (non-UTF-8 bytes possible).
2. Index writing truncated names with "{name:.64}", causing downstream mismatches.
3. cstranslate --ffindex appears sensitive to very long basenames.

What We Changed (local patches)

ffindex.py
Read/write binary; decode/encode lines with latin-1 (1:1 byte mapping).
Removed 64-char truncation; write full names.
Actually sort entries (entries.sort(...)); use mmap.ACCESS_READ.
Use splitlines(); open output index in binary.

hhsuitedb.py
New robust read_ffindex(path) (binary + latin-1; ignore malformed short lines).
write_subset_index(...) writes exact, untruncated names in latin-1 with \n.
This stabilizes subset builds used by ffindex_apply_mpi and cstranslate.

After these changes, _a3m/_hhm build reliably and _cs219 builds for almost all families.

Filename Fix

Simplifiing input .a3m filenames can help with some cases.