Most of the bugs in this file have been reported via official bug tracker, which currently is email box <[email protected]>. For full listing of fixed bugs and changes, see ChangeLog.
The automatic handling of flag diacritics, i.e. parsing symbols of form @?.*@ as special epsilons with side effects is only implemented in hfst-compose-intersect, hfst-compose, hfst-lookup and hfst-optimized-lookup. Handling them as epsilons for other tools that may require them to be treated as epsilons does not work. Also, the support in hfst-compose and hfst-compose-intersect is only partial, as it only supports the use of flag diacritics on left hand side transducer or lexicon.
The handling of # or .#. as automatic border of automaton symbol is only supported in hfst-twolc, hfst-lexc and hfst-compose-intersect. The support of # as special symbol in hfst-lexc only applies to using lexc transducers on left hand side of hfst-compose-intersect.
- Library should provide package-config .pc-files and stable API and reliable version numbering. (External library requirement, e.g. for Voikko)
- Library headers should not reveal underlying variables or structures, nor cause all compilations to report warnings from bugs of underlying libraries.
- Transducer reading will segfault with files which have a header that is potentially valid HFST, SFST or OpenFST transducer header but rest of the file is faulty.
- hfst-lookup tokenizes left-to-right longest match, hfst-optimized-lookup ditto except with ASCII where all ASCII are one arc.
- Swapped ? and ?:? parsing of xerox regular expressions in r266
- Fixed handling of empty transducers in intersecting composition in r254
- Multiple lookup formatting fixes and lookup input format for apertium in r242
- Fixed multiple include bugs of hfst string parsers in r232
Fixed flag diacritic support in hfst-diff-test in r196
Fixed optimized lookup transducer not containing epsilons in r193
Fixed D.X in optimized lookup in r183
Fixed flag diacritic support in hfst-diff-test in r196
Fixed optimized lookup transducer not containing epsilons in r193
Fixed D.X in optimized lookup in r183
Better handling of R.X and D.X flag diacritics in r178
Fixed platform dependent end-state parsing bugs in r157
Possibility to treat flag diacritics as epsilons in hfst-compose partially in r146
Handle # as special symbol in lexc and twolc, mostly fixed in r120
Mac OS X gettext / autopoint failures, resolved in r150 by removal of gettext
libtool version 1 vs. 2 macros, partially resolved in r127
hfst-twolc escaped and low ascii symbol issues, Resolved in r97, r125
hfst-lookup does not allow non-ascii in readline on Mac OS X, resolved in r85 by removal of readline
hfst-fst2string -u -n -w
hfst-lookup formats: Resolved in r63:69, r76:77
“hfst-optimized-lookup not accepting file input” Resolved in documentation
(FIXED in development branch) The currently (elsewhere) distributed German transducer, german.hfst, fails under hfst-lookup with
- terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted
The optimized lookup version, german.hfst.ol, works fine however.