2 * Fixes error handling in HashMatcher (1.6.6)
3 * Adds kShortestDelta for operations dependent on shortest-distance (1.6.6)
4 * Adds Python methods for (un)pickling and (de)serializing FSTs (1.6.6)
5 * Adds constructive variants of Invert and Project (1.6.6)
6 * Increases code sharing in MemoryPool/MemoryArena (1.6.6)
7 * Improves consistency of matcher FST ownership (1.6.6)
8 * Adds non-trivial A* estimator class (1.6.6)
9 * Prevents unreachable code generation in libfstscript (1.6.5)
10 * Adds move constructors for non-trivial weight types (1.6.5)
11 * Standardizes method names for tuple weight types (1.6.5)
12 * Eliminates undefined behavior in weight hashing (1.6.5)
13 * Optimizes binary search in SortedMatcher (1.6.5)
14 * Adds SetWeight (1.6.5)
15 * Fixes typing error in Python FAR reader (1.6.4)
16 * Removes restriction that Prune argument have commutative weights (1.6.3)
17 * Improves configuration of CompositeWeight readers and writers (1.6.3)
18 * Improves accuracy of ShortestDistance summation (1.6.3)
19 * SetFinal now "moves" its weight argument (1.6.3)
20 * Exposes ArcIterator and EncodeMapper flags in Python (1.6.3)
21 * Properly sets return codes in FST binaries (1.6.3)
22 * Eliminates StringWeight macros (1.6.3)
23 * Finalizes most virtual method overrides (1.6.2)
24 * Fixes missing includes of <fst/log.h> (1.6.1)
25 * Adds float format support to FST drawing (1.6.1)
26 * Extensive modernization for C++11 style (1.6.0)
27 * Many classes and constants moved into an internal namespace (1.6.0)
28 * Adds HashMatcher (1.6.0)
29 * Adds Member method to SymbolTable (1.6.0)
30 * Adds the "special" extension and the fstspecial binary; this is similar to
31 fstconvert but accepts arguments for specifying special labels (phi, rho,
32 and sigma) of FSTs (1.6.0)
33 * Exposes allow_negative_label option for Python symbol tables (1.6.0)
36 * Added p-subsequential determinization (1.5.0)
37 * Generalized epsilon normalization to non-functional case (1.5.0)
38 * Added general gallic (plus is union) semiring (1.5.0)
39 * Added FST compression extension (1.5.0)
40 * Added Python extension (1.5.0)
41 * Added multiple pushdown transducer (MPDT) support (1.5.0)
42 * Fixed Isomorphic function (1.5.0)
43 * Added final method to matchers (1.5.0)
44 * Fixed various compiler issues (1.5.0)
45 * Fixed missing Isomorphic components (1.5.0)
46 * Added UnionWeight (1.5.0)
47 * Added InputEpsilonMapper and OutputEpsilonMapper arc mappers (1.5.1)
48 * Added TrivialComposeFilter for more efficient composition when one
49 of the arguments is epsilon-free (1.5.1)
50 * Added properties bits kUnweightedCycles and kWeightedCycles (1.5.1)
51 * Added missing const qualification to (1.5.1):
52 - SymbolTableIterator access
53 - EncodeMapper writing to file
54 - EncodeMapper SymbolTable access
55 * Replaced internal custom reference-counting (RefCounter) with
56 C++11 smart pointers where possible, and fixed associated
57 reference-counting bugs (1.5.1)
58 * When calling DeleteStates on a MutableFst with a shared impl, the impl
59 is set to a new empty impl rather than copying and deleting (1.5.1)
60 * Prepended `Pdt` to the Expand libraries and classes in the PDT
61 extension, and prepended `MPdt` to the Expand libraries and classes
62 in the MPDT extension, so that both can be used in the same compilation
64 * Added option to PDT Replace for compiling a strongly-regular RTN into a
65 bounded-stack PDT (1.5.1)
66 * Improved symbol table support for PDT Replace, including automatic
67 generation of parentheses symbols (1.5.1)
68 * Improvements to scripting API (1.5.1):
69 - Added methods for FST access and mutation
70 - Added additional checks for arc/weight compatibility
71 - WeightClass::One and WeightClass::Zero now require a specified weight
72 type at time of construction
73 - Improved VectorFstClass constructors
74 - Added linear-time check for cyclic dependencies in Replace
75 - Added EncodeMapperClass, a template-free box for an EncodeMapper
76 * Improvements to the binaries (1.5.1):
77 - Fixed no-op --precision flag to fstdraw (1.5.1)
78 - Fixed no-op --file_list_input flag to farcreate (1.5.1)
79 * Improvements to the Python extension (1.5.1):
80 - Added methods for creating an empty mutable FST
81 - Added methods for FST access via state and arc iteration
82 - Added FST compilation from arclists (cf. fstcompile)
83 - Added FST printing and drawing
84 - Added FarReader and FarWriter classes.
85 * FarReader's GetFst method now returns a pointer (1.5.2)
86 * Fixed FSTERROR macro (1.5.2)
87 * Fixed build flags for dlopen (1.5.2)
88 * Consolidated Python extension into single module (1.5.2)
89 * Python add_arc now takes an Arc object (1.5.2)
90 * Adds optional minimization of non-deterministic FSTs (1.5.3)
91 * Mutation methods of the Python Fst object now support chaining (1.5.3)
92 * Scripting API and Python weight objects now support semiring arithmetic
94 * Adds RemoveSymbol method to SymbolTable (1.5.4)
95 * Prevents underflow when using LogProbArcSelector in random generation
97 * Makes random weight generators a single template class (1.5.4)
98 * Makes weight Properties constexpr where possible (1.5.4)
99 * Adds check for error when opening files when compiling strings into FARs
101 * Adds routines for parsing string flags to the scripting API (1.5.4)
104 * Port to C++11 (1.4.0)
105 * Disambiguate function added (1.4.0)
106 * Isomorphic function added (1.4.0)
107 * Matcher interface augmented with Priority method.
108 * Special matchers (rho/sigma/phi) can match special symbols
109 on both input FSTs in composition/intersection provided at each
110 state pair they only match one side (1.4.0)
111 * Added ExplicitMatcher to suppress implicit matches (e.g. epsilon
113 * Linear{Tagger,Classifier}Fst extensions added (1.4.0).
114 * Generalized state-reachable to work when input is cyclic (so long as no
115 final state is in a cycle). This ensures label-reachable (and hence label
116 lookahead) works with cyclic input (1.4.0)
117 * Added Condense to build the condensation graph (SCCs condensed to single
118 states) of an FST (1.4.0).
119 * Added an option to Reverse to specify whether a super-initial state
120 should always be created (1.4.0).
121 * Fixed bugs in FirstCacheStore, PowerWeight, and StringCompiler (1.4.0).
122 * Changed SymbolTable to use faster data structure (1.4.0).
123 * Added 'min' disambiguation in determinizaton to keep only the minimum
124 output in a non-functional transducer when plus=min/max
125 (flag --disambiguate_output) (1.4.1)
126 * Compiler issues in linear-fst fixed (1.4.1)
129 * Support for non-fatal exits on errors: (1.3.1)
130 - Added FLAGS_fst_error_fatal: FST errors are
131 fatal if true (default); o.w. return objects flagged as bad:
133 prop. true, FST weights - not a Member().
134 - Added kError property bit signifying bad FST
135 - Added NoWeight() method to FST weight requirements that returns
136 weight that is not a Member().
137 * Various improvements to the FAR extensions (1.3.1)
138 - a single FST is now a FAR type
139 - FLAGS_initial_symbols: Uses the symbol table from the
140 first FST in the archive for all entries"
141 - Input/output to standard input/output for some FAR and arc types
142 * --with-icu configuration option no longer needed (1.3.1)
143 * Improved flags usage esp. if use SET_FLAGS not SetFlags/InitFst (1.3.2)
144 * Added 'fst' as possible far writer type (1.3.2)
145 * phi matcher can now accept 0 as the phi label (1.3.2)
146 * Added ngram-fst extension (1.3.2)
147 * Improved performance of PDT composition (1.3.3)
148 * Memory-map support (1.3.3)
149 * Fixed cross-FST serialization issues (1.3.3)
150 * Fixed NGramFst off-by-one issue (1.3.3)
151 * farextract now allows one to specify a list of comma-separated keys,
152 including key ranges (1.3.3)
153 * Fixed bug in PDT replace that could cause close paren IDs to collide
154 with open paren IDs (1.3.4)
157 * Added lookahead matching and filtering for faster composition
158 * Added EditFst for mutation of o.w. immutable FSTs
159 * Added script sub-namespace defining type FstClass, a non-templated
160 Fst<Arc> to hold the arc template type internally. This and FST
161 operations on it allow easier I/O and scripting at the cost of some
163 * Added per-arc-iterator control of Fst caching.
164 * Added PowerWeight and Power Arc.
165 * Added SparsePowerWeight and SparsePowerArc (1.2.4)
166 * Added SignedLogWeight and SignedLogArc (1.2.4)
167 * Added ExpectationWeight and ExpectationArc (1.2.4)
168 * Added AStarQueue, PruneQueue and NaturalPruneQueue disciplines (1.2.6)
169 * Added Log64Weight and Log64Arc to FST library throughout, including
170 support throughout scripts/bins/dsos (1.2.8)
171 * Added delayed RandGenFst that outputs tree of paths weighted
173 * Added fstsymbols shell-level command
174 * Added total weight removal option to pushing
175 * Changed methods for symbol table mutation:
176 use MutableInputSymbols()/MutableOutputSymbols().
177 * Numerous efficiency improvements esp in composition, replace, and caching
178 * Made "fstmap" handle semiring conversion by adding "to_std", "to_log"
179 and "to_log64" as supported 'map_type' arguments (1.2.8).
180 * Made the destructive implementation of RmEpsilon skip over states
181 admitting no non-epsilon incoming transition (1.2.8).
182 * Fixed numerous bugs (1.2 through 1.2.9) including:
183 - improper types of some approximation deltas
184 - sub-optimal hashing functions
185 - issues in internal reuse of shortest distance
186 - hashing bug in FloatWeight
187 - bug in shortest path queue
188 - symbol table checksumming issues
189 - various C++ standards issues
190 - Visit() behavior when visitation aborted
191 - Decode() hash performance bug (1.2.1)
192 - EditFst::Copy(bool) method when the boolean parameter is true (1.2.7)
193 - SymbolTable memory leak in Invert() (1.2.8)
194 - Added escaping of " and \ in labels in fstdraw, needed for dot to
195 function properly (1.2.8)
196 - Fixed handling of final weight of start state in fstpush (1.2.8)
197 - Added FST_LL_FORMAT to fix 64-bit integer printf issues (1.2.9)
198 - Fixed missing <functional> includes (1.2.9)
199 - Fixed reused local variable names (1.2.9)
200 - Fixed passing string by reference in FstDraw args (1.2.9)
201 * Added extensions directories including:
202 - finite-state archive (FAR) utilities,
203 added stlist format supporting writing/reading to/from standard out/in
204 at the library-level (1.2.8)
207 - pushdown transducers (improved in 1.2.1 through 1.2.7).
208 * Added StateMap/StateMapFst; renamed Map/MapFst to ArcMap/ArcMapFst;
209 map/MapFst retained (but deprecated) (1.2.9)
210 * Deleted ArcSum() and ArcMerge; use StateMap w/ ArcSumMapper and
211 ArcUniqueMapper (1.2.9).
212 * Incremented version of ConstFst/CompactFsts to stop memory alignment
213 that fails on pipes. Made old version raises errors when read on
215 * Improved determinize hash (1.2.9)
216 * Removed stdio uses (1.2.10)
217 * Fixed library ordering issues esp. with newer GNU build tools (1.2.10)
220 * Added compat.h to src/include/fst to fix missing defines
221 * Fixed bug in acyclic minimization that led to non-minimal
222 (but equivalent) results
223 * Fixed missing FST typedef in various matchers in matcher.h
224 so that they can be cascaded
225 * Opened file streams binary where appropriate
227 OpenFst: Release 1.0 (Additions to beta version):
228 * Matcher class added for matching labels at FST states. Includes
229 special matchers for sigma (any), rho ('rest'), and phi ('fail')
231 * Composition generalized with arbitrary filters, matchers, and state
233 * Sequence and matching composition filters provided. (see compose.h,
234 compose-filter.h, matcher.h, state-table.h)
235 * Unique n-best (see shortest-path.h)
236 * Pruning in determinization and epsilon removal (see determinize.h,
239 * Compact Fsts for space-efficient representation (see compact-fst.h)
240 * New Weight classes:
243 * Miscellaneous bug fixes