Changelog -- hyph-malagasy
============================

0.3.1-dev (2026-09-02)
------------------------
Added a French-language user guide (doc/hyph-malagasy-guide-utilisation.pdf)
covering installation, both compilation paths (LuaTeX dynamic loading
and pdfTeX INITEX format), pattern regeneration/validation, running
tests, and contributing corrections. Repository layout harmonised:
validate.py and extract_hyphenation.py now live under source/ (were
previously in a separate scripts/ directory in the development
repository), matching the flat CTAN structure. No changes to the
patterns themselves.

0.3.0-dev (2026-09-02)
------------------------
The 29-word corpus was reviewed by a native Malagasy speaker (using
docs/relecture-locuteurs-natifs.md, a jargon-free review sheet). All
29 words were confirmed correct, with no corrections proposed. This
resolves the "not reviewed by a native speaker" caveat for the
current corpus (Academia Malgache validation is still pending, and
the corpus remains intentionally small). No changes to the patterns
themselves.

0.2.1-dev (2026-09-02)
------------------------
Flattened archive structure at the request of the CTAN team (dropped
the "generic/hyph-malagasy" path segments -- files now sit directly
under tex/, doc/, source/). Added an English summary to README, and
made explicit in README/description/announcement that this material
is not yet ready for integration into TeX distributions. No changes
to the patterns themselves.

0.2.0-dev (2026-09-02)
------------------------
Fixed a cross-contamination bug found during testing (words absent
from the corpus were getting spurious hyphens). Pattern generation
now emits one whole-word-anchored pattern per corpus word instead of
all substrings, which structurally eliminates cross-contamination.
Corpus went from 623 raw substrings to 29 whole-word patterns.
Re-validated by real compilation (pdfTeX + LuaTeX): corpus words
still hyphenate identically; out-of-corpus words stay intact.
Trade-off: no generalisation to unseen words.

0.1.0-dev (2026-09-02)
------------------------
First public preview.

- Initial corpus of 33 annotated words (Liang notation), covering
  Malagasy prenasalised consonant clusters (mp, mb, nk, ng, nd, ndr,
  nj, nts) and digraphs/trigraphs (ts, tr, dr), several common
  prefixes (an-/am-, man-, mank-, mi-, fi-, faha-, voa-, mpi-), and
  two reduplication cases.
- Generation (generate_patterns.py), validation (validate.py) and
  extraction (extract_hyphenation.py) scripts.
- Real pdfTeX (\showhyphens) and LuaTeX (post_linebreak_filter
  callback) tests.
- Known limitation at the time: cross-contamination between patterns
  (fixed in 0.2.0-dev, see above).
- Corpus and reference files not yet reviewed by a native Malagasy
  speaker or validated by the Academia Malgache.