2 * TokenType is now a scoped enum (1.7.8)
3 * pywrapfst is now Python 3-only (1.7.8)
4 * fstproject now has --project_type flag (1.7.8)
5 * BitmapIndex is now 2x faster for Select0/Select1 (1.7.8)
6 * Property testing is now thread-safe (1.7.7)
7 * Modernizes random generation (1.7.7)
8 * Adds MakeArcMapFst (1.7.6)
9 * Adds RealWeight and Real64Weight (1.7.6)
10 * Adds a new, idiomatic SymbolTable iterator interface (1.7.6)
11 * Improves symbol table lifetime management in pywrapfst (1.7.6)
12 * Improves the design of the FST class hierarchy in pywrapfst (1.7.6)
13 * Removes unnecessary template parameters in constructors (1.7.5)
14 * Converts RmEpsilonFstOptions from class to struct (1.7.5)
15 * Eliminates redundant checks in Minimize (1.7.5)
16 * CompactFst is now templated on Compactor rather than ArcCompactor (1.7.4)
17 * Removes harmful constexpr specifications in the FAR extension (1.7.4)
18 * Improved script API support for EncodeMapper (1.7.4)
19 * New header format for the EncodeMapper (1.7.4)
20 * Many cleanups to the n-gram extension (1.7.4)
21 * Improved C++17 compatibility shims (1.7.4)
22 * Overloads Arc constructors with default weight argument (1.7.3)
23 * Fixes RmEpsilon and Union property checking bugs (1.7.3)
24 * Makes Isomorphic more robust to nondeterminism (1.7.3)
25 * Adds default weight argument to SetFinal (1.7.3)
26 * Cleans up low-level logging (1.7.3)
27 * Adds power-weight mappers (1.7.3)
28 * Adds expander cache (1.7.3)
29 * Fixes bug with coinaccessible states in NaturalAStarEstimate (1.7.2)
30 * Optionally allows building with Bazel (1.7.2)
31 * Simplifies string printing interface (1.7.2)
32 * Marks weight converters const (1.7.2)
33 * Adds NoMatchComposeFilter (1.7.2)
34 * Removed static assertions that trigger bugs in GCC (1.7.1)
35 * Evaluates many weight operations at compile-time (1.7.1)
36 * Improved use of move semantics, especially in cache-backed FSTs (1.7.0)
37 * Adds configure-time test for float equality reflexivity (1.7.0)
38 * Removes volatile qualifiers from float weights (1.7.0)
39 * Protections for signedness in string compiler (1.7.0)
40 * Adds additional overloads to Equals (1.7.0)
41 * Clean-up to weight constructors (1.7.0)
44 * Optimized label lookup in SymbolTable (1.6.9)
45 * Fixed HashMatcher issues with SetState() and Find() consistency (1.6.8)
46 * Fixed PROGRAM_FLAGS documentation string in binaries (1.6.8)
47 * Fixed handling of symbol tables in EpsNormalize (1.6.8)
48 * Fixed error reporting when FST arc type unknown (1.6.8)
49 * The first_path option to ShortestPath is now optimal for A* (1.6.7)
50 * Renames SymbolTable::kNoSymbol to kNoSymbol (1.6.7)
51 * Exposes PowerMapper to the scripting API (1.6.7)
52 * Fixes linking of the special SOs (1.6.7)
53 * Adds kShortestDelta for operations dependent on shortest-distance (1.6.6)
54 * Adds Python methods for (un)pickling and (de)serializing FSTs (1.6.6)
55 * Adds constructive variants of Invert and Project (1.6.6)
56 * Increases code sharing in MemoryPool/MemoryArena (1.6.6)
57 * Improves consistency of matcher FST ownership (1.6.6)
58 * Fixes error handling in HashMatcher (1.6.6)
59 * Adds non-trivial A* estimator class (1.6.6)
60 * Prevents unreachable code generation in libfstscript (1.6.5)
61 * Adds move constructors for non-trivial weight types (1.6.5)
62 * Standardizes method names for tuple weight types (1.6.5)
63 * Eliminates undefined behavior in weight hashing (1.6.5)
64 * Optimizes binary search in SortedMatcher (1.6.5)
65 * Adds SetWeight (1.6.5)
66 * Fixes typing error in Python FAR reader (1.6.4)
67 * Removes restriction that Prune argument have commutative weights (1.6.3)
68 * Improves configuration of CompositeWeight readers and writers (1.6.3)
69 * Improves accuracy of ShortestDistance summation (1.6.3)
70 * SetFinal now "moves" its weight argument (1.6.3)
71 * Exposes ArcIterator and EncodeMapper flags in Python (1.6.3)
72 * Properly sets return codes in FST binaries (1.6.3)
73 * Eliminates StringWeight macros (1.6.3)
74 * Finalizes most virtual method overrides (1.6.2)
75 * Adds float format support to FST drawing (1.6.1)
76 * Fixes missing includes of <fst/log.h> (1.6.1)
77 * Adds the "special" extension and the fstspecial binary; this is similar to
78 fstconvert but accepts arguments for specifying special labels (phi, rho,
79 and sigma) of FSTs (1.6.0)
80 * Exposes allow_negative_label option for Python symbol tables (1.6.0)
81 * Many classes and constants moved into an internal namespace (1.6.0)
82 * Extensive modernization for C++11 style (1.6.0)
83 * Adds Member method to SymbolTable (1.6.0)
84 * Adds HashMatcher (1.6.0)
87 * Generalized epsilon normalization to non-functional case (1.5.0)
88 * Added multiple pushdown transducer (MPDT) support (1.5.0)
89 * Added general gallic (plus is union) semiring (1.5.0)
90 * Added p-subsequential determinization (1.5.0)
91 * Fixed missing Isomorphic components (1.5.0)
92 * Added FST compression extension (1.5.0)
93 * Added final method to matchers (1.5.0)
94 * Fixed various compiler issues (1.5.0)
95 * Fixed Isomorphic function (1.5.0)
96 * Added Python extension (1.5.0)
97 * Added UnionWeight (1.5.0)
98 * Added missing const qualification to (1.5.1):
99 - SymbolTableIterator access
100 - EncodeMapper writing to file
101 - EncodeMapper SymbolTable access
102 * Added TrivialComposeFilter for more efficient composition when one
103 of the arguments is epsilon-free (1.5.1)
104 * Added InputEpsilonMapper and OutputEpsilonMapper arc mappers (1.5.1)
105 * Added properties bits kUnweightedCycles and kWeightedCycles (1.5.1)
106 * Replaced internal custom reference-counting (RefCounter) with C++11 smart
107 pointers where possible, and fixed reference-counting bugs (1.5.1)
108 * When calling DeleteStates on a MutableFst with a shared impl, the impl
109 is set to a new empty impl rather than copying and deleting (1.5.1)
110 * Prepended Pdt to the Expand libraries and classes in the PDT
111 extension, and prepended MPdt to the Expand libraries and classes
112 in the MPDT extension, so that both can be used in the same compilation
114 * Added option to PDT Replace for compiling a strongly-regular RTN into a
115 bounded-stack PDT (1.5.1)
116 * Improved symbol table support for PDT Replace, including automatic
117 generation of parentheses symbols (1.5.1)
118 * Improvements to scripting API (1.5.1):
119 - Added methods for FST access and mutation
120 - Added additional checks for arc/weight compatibility
121 - WeightClass::One and WeightClass::Zero now require a specified weight
122 type at time of construction
123 - Improved VectorFstClass constructors
124 - Added linear-time check for cyclic dependencies in Replace
125 - Added EncodeMapperClass, a template-free box for an EncodeMapper
126 * Improvements to the binaries (1.5.1):
127 - Fixed no-op --precision flag to fstdraw (1.5.1)
128 - Fixed no-op --file_list_input flag to farcreate (1.5.1)
129 * Improvements to the Python extension (1.5.1):
130 - Adds methods for creating an empty mutable FST
131 - Adds methods for FST access via state and arc iteration
132 - Adds FST compilation from arclists (cf. fstcompile)
133 - Adds FST printing and drawing
134 - Adds FarReader and FarWriter classes.
135 * Consolidated Python extension into single module (1.5.2)
136 * FarReader's GetFst method now returns a pointer (1.5.2)
137 * Python add_arc now takes an Arc object (1.5.2)
138 * Fixed build flags for dlopen (1.5.2)
139 * Fixed FSTERROR macro (1.5.2)
140 * Scripting API and Python weight objects now support semiring arithmetic
142 * Mutation methods of the Python Fst object now support chaining (1.5.3)
143 * Adds optional minimization of non-deterministic FSTs (1.5.3)
144 * Adds check for error when opening files when compiling strings into FARs
146 * Prevents underflow when using LogProbArcSelector in random generation
148 * Adds routines for parsing string flags to the scripting API (1.5.4)
149 * Makes random weight generators a single template class (1.5.4)
150 * Makes weight Properties constexpr where possible (1.5.4)
151 * Adds RemoveSymbol method to SymbolTable (1.5.4)
154 * Port to C++11 (1.4.0)
155 * Isomorphic function added (1.4.0)
156 * Disambiguate function added (1.4.0)
157 * Matcher interface augmented with Priority method
158 * Special matchers (rho/sigma/phi) can match special symbols
159 on both input FSTs in composition/intersection provided at each
160 state pair they only match one side (1.4.0)
161 * Added ExplicitMatcher to suppress implicit matches (e.g. epsilon
163 * Linear{Tagger,Classifier}Fst extensions added (1.4.0).
164 * Generalized state-reachable to work when input is cyclic (so long as no
165 final state is in a cycle). This ensures label-reachable (and hence label
166 lookahead) works with cyclic input (1.4.0)
167 * Added Condense to build the condensation graph (SCCs condensed to single
168 states) of an FST (1.4.0).
169 * Added an option to Reverse to specify whether a super-initial state
170 should always be created (1.4.0).
171 * Fixed bugs in FirstCacheStore, PowerWeight, and StringCompiler (1.4.0).
172 * Changed SymbolTable to use faster data structure (1.4.0).
173 * Added 'min' disambiguation in determinizaton to keep only the minimum
174 output in a non-functional transducer when plus=min/max
175 (flag --disambiguate_output) (1.4.1)
176 * Compiler issues in linear-fst fixed (1.4.1)
179 * Support for non-fatal exits on errors: (1.3.1)
180 - Added FLAGS_fst_error_fatal: FST errors are
181 fatal if true (default); o.w. return objects flagged as bad:
183 prop. true, FST weights - not a Member().
184 - Added kError property bit signifying bad FST
185 - Added NoWeight() method to FST weight requirements that returns
186 weight that is not a Member().
187 * Various improvements to the FAR extensions (1.3.1)
188 - a single FST is now a FAR type
189 - FLAGS_initial_symbols: Uses the symbol table from the
190 first FST in the archive for all entries"
191 - Input/output to standard input/output for some FAR and arc types
192 * --with-icu configuration option no longer needed (1.3.1)
193 * Improved flags usage esp. if use SET_FLAGS not SetFlags/InitFst (1.3.2)
194 * Added 'fst' as possible far writer type (1.3.2)
195 * phi matcher can now accept 0 as the phi label (1.3.2)
196 * Added ngram-fst extension (1.3.2)
197 * Improved performance of PDT composition (1.3.3)
198 * Memory-map support (1.3.3)
199 * Fixed cross-FST serialization issues (1.3.3)
200 * Fixed NGramFst off-by-one issue (1.3.3)
201 * farextract now allows one to specify a list of comma-separated keys,
202 including key ranges (1.3.3)
203 * Fixed bug in PDT replace that could cause close paren IDs to collide
204 with open paren IDs (1.3.4)
207 * Added lookahead matching and filtering for faster composition
208 * Added EditFst for mutation of o.w. immutable FSTs
209 * Added script sub-namespace defining type FstClass, a non-templated
210 Fst<Arc> to hold the arc template type internally. This and FST
211 operations on it allow easier I/O and scripting at the cost of some
213 * Added per-arc-iterator control of Fst caching.
214 * Added PowerWeight and Power Arc.
215 * Added SparsePowerWeight and SparsePowerArc (1.2.4)
216 * Added SignedLogWeight and SignedLogArc (1.2.4)
217 * Added ExpectationWeight and ExpectationArc (1.2.4)
218 * Added AStarQueue, PruneQueue and NaturalPruneQueue disciplines (1.2.6)
219 * Added Log64Weight and Log64Arc to FST library throughout, including
220 support throughout scripts/bins/dsos (1.2.8)
221 * Added delayed RandGenFst that outputs tree of paths weighted
223 * Added fstsymbols shell-level command
224 * Added total weight removal option to pushing
225 * Changed methods for symbol table mutation:
226 use MutableInputSymbols()/MutableOutputSymbols().
227 * Numerous efficiency improvements esp in composition, replace, and caching
228 * Made "fstmap" handle semiring conversion by adding "to_std", "to_log"
229 and "to_log64" as supported 'map_type' arguments (1.2.8).
230 * Made the destructive implementation of RmEpsilon skip over states
231 admitting no non-epsilon incoming transition (1.2.8).
232 * Fixed numerous bugs (1.2 through 1.2.9) including:
233 - improper types of some approximation deltas
234 - sub-optimal hashing functions
235 - issues in internal reuse of shortest distance
236 - hashing bug in FloatWeight
237 - bug in shortest path queue
238 - symbol table checksumming issues
239 - various C++ standards issues
240 - Visit() behavior when visitation aborted
241 - Decode() hash performance bug (1.2.1)
242 - EditFst::Copy(bool) method when the boolean parameter is true (1.2.7)
243 - SymbolTable memory leak in Invert() (1.2.8)
244 - Added escaping of " and \ in labels in fstdraw, needed for dot to
245 function properly (1.2.8)
246 - Fixed handling of final weight of start state in fstpush (1.2.8)
247 - Added FST_LL_FORMAT to fix 64-bit integer printf issues (1.2.9)
248 - Fixed missing <functional> includes (1.2.9)
249 - Fixed reused local variable names (1.2.9)
250 - Fixed passing string by reference in FstDraw args (1.2.9)
251 * Added extensions directories including:
252 - finite-state archive (FAR) utilities,
253 added stlist format supporting writing/reading to/from standard out/in
254 at the library-level (1.2.8)
257 - pushdown transducers (improved in 1.2.1 through 1.2.7).
258 * Added StateMap/StateMapFst; renamed Map/MapFst to ArcMap/ArcMapFst;
259 map/MapFst retained (but deprecated) (1.2.9)
260 * Deleted ArcSum() and ArcMerge; use StateMap w/ ArcSumMapper and
261 ArcUniqueMapper (1.2.9).
262 * Incremented version of ConstFst/CompactFsts to stop memory alignment
263 that fails on pipes. Made old version raises errors when read on
265 * Improved determinize hash (1.2.9)
266 * Removed stdio uses (1.2.10)
267 * Fixed library ordering issues esp. with newer GNU build tools (1.2.10)
270 * Added compat.h to src/include/fst to fix missing defines
271 * Fixed bug in acyclic minimization that led to non-minimal
272 (but equivalent) results
273 * Fixed missing FST typedef in various matchers in matcher.h
274 so that they can be cascaded
275 * Opened file streams binary where appropriate
277 OpenFst: Release 1.0 (Additions to beta version):
278 * Matcher class added for matching labels at FST states. Includes
279 special matchers for sigma (any), rho ('rest'), and phi ('fail')
281 * Composition generalized with arbitrary filters, matchers, and state
283 * Sequence and matching composition filters provided. (see compose.h,
284 compose-filter.h, matcher.h, state-table.h)
285 * Unique n-best (see shortest-path.h)
286 * Pruning in determinization and epsilon removal (see determinize.h,
289 - Compact FSTs for space-efficient representation (see compact-fst.h)
290 * New Weight classes:
293 * Miscellaneous bug fixes