main branch superseded with cm

git-svn-id: https://pet.opendfki.de/repos/pet/main@743 4200e16c-5112-0410-ac55-d7fb557a720a

CHANGELOG

@@ -1,3 +1,23 @@
+specific to chart-mapping branch:
+  - activate all rules after lex-exhaustive parsing (not just syntactic rules)
+    for compatibility with LKB
+  - regular expression literals introduced with slashes in TDL
+  - assert that fss in unify(), unify_restrict() and unify_np() are valid
+  - added preliminary XML-RPC server mode
+  - added chart mapping engine:
+    * chart mapping rules are resource-sensitive rewrite rules
+    * chart mapping rules operate on a generalized chart
+    * two phases: token mapping and lexical filtering
+  - new generic instantiation mode `default-les=all': try to instantiate
+    all available generics for all available input items. selection should
+    then be accomplished by constraints on the input fs
+  - new tokenizer for the XML-based feature structure chart format (FSC)
+  - input fs can optionally be unified into a lexical item's fs
+  - input items can now be described by input feature structures
+  - case-sensitive string and YY tokenizer
+  - memory limit is respected during unpacking
+  - added unit tests for types
+
 v0.99.??
   - added the ParseNodes class to be able to produce nice small parse trees as
     in the online demo, and an item printer to test it

Makefile.am

@@ -54,9 +54,9 @@ cheapdoc:
 	$(DOXYGEN) doxyconfig.cheap
 
 logon:
-	strip flop/flop cheap/cheap
-	/bin/cp flop/flop $$LOGONROOT/uio/bin/linux.x86.32/flop
-	/bin/cp cheap/cheap $$LOGONROOT/uio/bin/linux.x86.32/cheap
+	strip flop/flop cheap/cheap 
+	/bin/cp flop/flop $$LOGONROOT/uio/bin/linux.x86.32/tflop
+	/bin/cp cheap/cheap $$LOGONROOT/uio/bin/linux.x86.32/tcheap
 
 endif
 

README

@@ -110,6 +110,13 @@ privileges. Note that the header files are often distributed in separate
 packages, usually ending in `-dev' or `-devel' (e.g. `libboost-dev'). Make
 sure that these packages are installed, too.
 
+If you install your own libraries in non-standard directories, make sure
+that all these shared libraries and their dependencies can be found when
+PET's configure is run. There are several ways to achieve this (e.g., using
+ldconfig or the LD_LIBRARY_PATH environment variable, cf. the Program Library
+HOWTO). The dependencies of a program or library can be inspected with ldd.
+http://tldp.org/HOWTO/Program-Library-HOWTO/.
+
 Required:
 
 Boost provides free portable C++ libraries for various applications.
@@ -120,7 +127,7 @@ http://www.boost.org/
 
 Optional:
 
-If you want UniCode support in PET, you need the icu package freely
+If you want UniCode support in PET, you need the ICU package freely
 available from IBM.
 Current GNU/Linux distributions usually provide this library as a pre-built
 package. Make sure that the header files are installed, too (see above).

TODO

@@ -1,4 +1,6 @@
-+- make << operator for tItemPrinters to allow modifier stream like usage
+- use logging framework for chart-mapping
+
+- make << operator for tItemPrinters to allow modifier stream like usage
    - check if making the stream an argument of the print function decreases
      performance, if not, remove the global pointer to the stream used and
      replace it with a method argument 
@@ -79,6 +81,9 @@
         used for unification and mark it with a special value. After some 
         thinking, i would say that this might not be feasible.
 
+- extension of TDL syntax to type extension and comments (ticket #1 by Emily
+  Bender)
+
 - remove all uses of negative values as `marker'-values, especially where casts
   from integers to pointers and back are involved. 
 

cheap/Makefile.am

@@ -36,7 +36,7 @@ bin_PROGRAMS = cheap
 #bin_PROGRAMS += mtest
 
 cheap_SOURCES = cheap.cpp cheap.h
-cheap_LDADD = libcheap.la
+cheap_LDADD   = libcheap.a
 
 #dumpgram_SOURCES = dumpgram.cpp \
 #                   $(top_srcdir)/common/errors.h \
@@ -47,18 +47,20 @@ cheap_LDADD = libcheap.la
 
 # Convenience library (makes building several executables easier and is
 # a tiny step towards a general pet library):
-noinst_LTLIBRARIES = libcheap.la
+noinst_LIBRARIES = libcheap.a
 
 # The include-file dependencies are automatically computed by Automake
 # (`Automatic dependency tracking'). Thus, the headers do not need to be listed
 # for compilation, but they need to be found for making a distribution.
 # Note that the headers in $(top_srcdir)/common are always included
 # (cf. $(top_srcdir)/common/Makefile.am) and are therefore not listed here.
-libcheap_la_SOURCES = \
+libcheap_a_SOURCES = \
 	agenda.h \
 	chart.cpp chart.h \
+	chart-mapping.cpp chart-mapping.h \
 	cheaptimer.h \
 	fs.cpp fs.h \
+	fs-chart.cpp fs-chart.h \
 	grammar.cpp grammar.h \
 	input-modules.cpp input-modules.h \
 	item-printer.cpp item-printer.h \
@@ -79,7 +81,9 @@ libcheap_la_SOURCES = \
 	task.cpp task.h \
 	tsdb++.h tsdb++.cpp \
 	mrs.h mrs.cpp \
+	mrs-printer.h mrs-printer.cpp \
 	vpm.h vpm.cpp \
+	pcfg.h pcfg.cpp \
 	yy-tokenizer.cpp yy-tokenizer.h \
 	$(top_srcdir)/common/bitcode.cpp \
 	$(top_srcdir)/common/chunk-alloc.cpp \
@@ -102,21 +106,21 @@ libcheap_la_SOURCES = \
 
 if TOMABECHI_UNIFIER
 DEFS += -DDAG_TOMABECHI
-libcheap_la_SOURCES += dag-tomabechi.cpp dag-tomabechi.h failure.cpp failure.h \
+libcheap_a_SOURCES += dag-tomabechi.cpp dag-tomabechi.h failure.cpp failure.h \
                       qc.h qc.cpp
 endif
 if SIMPLE_UNIFIER
 DEFS += -DDAG_SIMPLE -DWROBLEWSKI2
-libcheap_la_SOURCES += $(top_srcdir)/common/dag-simple.cpp
+libcheap_a_SOURCES += $(top_srcdir)/common/dag-simple.cpp
 endif
 
 if ECL
-libcheap_la_SOURCES += petecl.c petecl.h
+libcheap_a_SOURCES += petecl.c petecl.h
 endif
 if ECLPREPROC
 if ICU
-libcheap_la_SOURCES += eclpreprocessor.h eclpreprocessor.cpp #fspp.h fspp.cpp
-cheap_LDADD += ../fspp/libfspp.la -lpreprocessor #fix_me
+libcheap_a_SOURCES += eclpreprocessor.h eclpreprocessor.cpp #fspp.h fspp.cpp
+cheap_LDADD += ../fspp/libfspp.a -lpreprocessor ../fspp/libfspp.a #fix_me
 LDFLAGS += -L../fspp
 CPPFLAGS += -I../fspp
 else
@@ -125,16 +129,16 @@ else
 endif
 endif
 if ECLMRS
-libcheap_la_SOURCES += cppbridge.cpp cppbridge.h petmrs.h petmrs.c
+libcheap_a_SOURCES += cppbridge.cpp cppbridge.h petmrs.h petmrs.c
 cheap_LDADD += libmrs.a
 endif
 
 # Add ecl libraries to be deleted on "make clean" ...
 # Eric Nichols <[email protected]>, Jun. 18, 2005
 MOSTLYCLEANFILES = libmrs.a
 
-../fspp/libfspp.la:
-	make -C ../fspp libfspp.la
+../fspp/libfspp.a:
+	make -C ../fspp libfspp.a
 
 libmrs.a:
 	( \
@@ -147,34 +151,39 @@ libmrs.a:
 	) | $(ECL)
 
 if XML
-libcheap_la_SOURCES += pic-handler.cpp pic-handler.h pic-states.h \
+libcheap_a_SOURCES += pic-handler.cpp pic-handler.h pic-states.h \
                       pic-tokenizer.h pic-tokenizer.cpp \
                       xmlparser.cpp xmlparser.h \
-                      smaf-tokenizer.h smaf-tokenizer.cpp
+                      smaf-tokenizer.h smaf-tokenizer.cpp \
+                      fsc-tokenizer.cpp fsc-tokenizer.h
 endif
 
 if ICU
-libcheap_la_SOURCES += unicode.h unicode.cpp
+libcheap_a_SOURCES += unicode.h unicode.cpp
+endif
+
+if XMLRPC_C
+libcheap_a_SOURCES += server-xmlrpc.cpp server-xmlrpc.h
 endif
 
 # if LEXDB
-# libcheap_la_SOURCES += psqllex.h psqllex.cpp
+# libcheap_a_SOURCES += psqllex.h psqllex.cpp
 # endif
 
 if YY
-libcheap_la_SOURCES += yy.h yy.cpp
+libcheap_a_SOURCES += yy.h yy.cpp
 endif
 
 #if EXTDICT
-#libcheap_la_SOURCES += extdict.cpp extdict.h
+#libcheap_a_SOURCES += extdict.cpp extdict.h
 #endif
 
 #if RCU
 # QC_PATH_COMP could be done robustly, ie with weights.  The aim would be
 # to find the paths that most quickly find out the biggest losses.
 #DEFS += -DROBUST
-#libcheap_la_SOURCES += rcu-types.h rcu-types.cpp rcu-dag-node.h
-#nodist_libcheap_la_SOURCES += dag-chunk-alloc.h dag-chunk-alloc.cpp \
+#libcheap_a_SOURCES += rcu-types.h rcu-types.cpp rcu-dag-node.h
+#nodist_libcheap_a_SOURCES += dag-chunk-alloc.h dag-chunk-alloc.cpp \
 #                             type-chunk-alloc.cpp type-chunk-alloc.h \
 #                             type-alloc.h type-alloc.cpp
 # Not done for fair comparison

cheap/agenda.h

@@ -26,32 +26,139 @@
 #include <queue>
 #include <vector>
 
+#include "item.h"
+#include "options.h"
+#include "task.h"
+
+
 /** agenda: a priority queue adapter */
-template <typename T, typename LESS_THAN > class agenda {
+template <typename T, typename LESS_THAN > class abstract_agenda {
 public:
 
-  agenda() : _A() {}
+  virtual ~abstract_agenda() {}
+
+  virtual void push(T *t) = 0;
+  virtual T * top() = 0;
+  virtual T * pop() = 0;     // Responsibility to delete the task is for the caller. 
+  virtual bool empty() = 0;
+  virtual void feedback (T *t, tItem *result) = 0; 
+
+};
+
+
+template <typename T, typename LESS_THAN > class exhaustive_agenda : public abstract_agenda<T, LESS_THAN > {
+public : 
 
-  ~agenda() { while(!this->empty()) delete this->pop(); }
+  exhaustive_agenda() : _A() {}
+  ~exhaustive_agenda() { while(!this->empty()) delete this->pop(); }
 
-  /** Push \a t onto agenda */
   void push(T *t) { _A.push(t); }
-
-  /** Return the topmost (best) element from the agenda */
-  T * top() { return _A.top(); }
+  T * top()       { return _A.top(); }
+  T * pop()       { T *t = top(); _A.pop(); return t; }
+  bool empty()    { return _A.empty(); }
+  void feedback (T *t, tItem *result) {}
+
+private:
+
+  std::priority_queue<T *, std::vector<T *>, LESS_THAN> _A;
+};
+
+
 
-  /** Remove the topmost element from the agenda and return it */
-  T * pop() { T *t = top(); _A.pop(); return t; }
 
-  /** Test if agenda is empty */
-  bool empty() { return _A.empty(); }
 
-  /** Return the number of tasks in the agenda */
-  int size() { return _A.size();}
+/*
+ * LOCAL CAP AGENDA
+ */
+
+template <typename T, typename LESS_THAN > class local_cap_agenda : public abstract_agenda<T, LESS_THAN > {
+/* This class provides functionality to define a per-cell cap on the number of tasks to be executed. */
+
+public : 
+
+  local_cap_agenda(int cell_size, int max_pos) : _A(), _popped((max_pos+1)*(max_pos+1)), _max_pos(max_pos), _cell_size(cell_size),
+                                                 _exec((max_pos+1)*(max_pos+1)), _succ((max_pos+1)*(max_pos+1)), _pass((max_pos+1)*(max_pos+1)) { 
+  }
+  ~local_cap_agenda();
+
+  void push(T *t) {
+    _A.push(t); 
+  }
+  T * top();
+  T * pop(); 
+  bool empty() { return top() == NULL; }
+  void feedback (T *t, tItem *result);
 
 private:
 
   std::priority_queue<T *, std::vector<T *>, LESS_THAN> _A;
+  std::vector<int> _popped;
+  int _max_pos;
+  int _cell_size;
+
+  std::vector<int> _exec;
+  std::vector<int> _succ;
+  std::vector<int> _pass; 
 };
 
+
+/*
+ * (implementation here, due to use of templates)
+ */
+
+template <typename T, class LESS_THAN>
+local_cap_agenda<T, LESS_THAN>::~local_cap_agenda() {
+  while (!_A.empty()) {
+    T* t = _A.top();
+    delete t;
+    _A.pop();
+  }
+}
+
+template <typename T, class LESS_THAN>
+T * local_cap_agenda<T, LESS_THAN>::top() {
+  T* t;
+  bool found = false;
+  while (!found) {
+    if (!_A.empty()) {
+      t = _A.top();
+      if (t->phrasal() && _popped[t->start()*(_max_pos+1) + t->end()] >= _cell_size) {
+        // This span reached the limit, so continue searching for a new task. 
+        // Inflectional and lexical rules are always carried out. 
+        delete t;
+        _A.pop();
+      } else {
+        found = true;
+      }
+    } else {
+      t = NULL;
+      break;
+    }
+  }
+  return t;
+}
+
+template <typename T, class LESS_THAN>
+T * local_cap_agenda<T, LESS_THAN>::pop() { 
+  T *t = top(); 
+  if (t != NULL) { 
+    _A.pop(); 
+  }
+  return t; 
+}
+
+template <typename T, class LESS_THAN>
+void local_cap_agenda<T, LESS_THAN>::feedback (T *t, tItem *result) { 
+  if (t->phrasal()) {
+    if (get_opt_int("opt_chart_pruning_strategy") == 0) {
+      _popped[t->start()*(_max_pos+1) + t->end()]++;
+    } else if (get_opt_int("opt_chart_pruning_strategy") == 1 && (result != 0)) {
+      _popped[t->start()*(_max_pos+1) + t->end()]++;
+    } else if (get_opt_int("opt_chart_pruning_strategy") == 2 && (result != 0) && t->yields_passive()) {
+      _popped[t->start()*(_max_pos+1) + t->end()]++;
+    }    
+  }
+}
+
+
 #endif