Update Makefile.

- include copy of valgrind-python.supp that adds suppresions for reads
  of size 8
- make faulthandler module optional
- fix bug with empty terminals and latex output.
- warn that specifying training set size as fraction is not possible
  with --rerun

Makefile

@@ -20,41 +20,46 @@ docs:
 	cd docs && make html
 
 test: all inplace
-	py.test --doctest-modules discodop/ tests/unittests.py && \
-	python -tt -3 tests.py && \
-	cd tests/ && \
-	sh run.sh
+	py.test --doctest-modules discodop/ tests/unittests.py \
+	&& python -tt -3 tests.py \
+	&& cd tests/ \
+	&& sh run.sh
 
-test3:
-	rm -rf sample/
+test3: clean
 	python3 setup.py install --user
+	cp --update build/lib.*/discodop/*.so discodop/
 	PYTHONIOENCODING=utf-8 python3 -bb -tt tests.py
 
-debug:
-	python-dbg setup.py build_ext --inplace --debug --pyrex-gdb
-
-testdebug: debug valgrind-python.supp
-	valgrind --tool=memcheck --leak-check=full --num-callers=30 \
-		--suppressions=valgrind-python.supp python-dbg -tt -3 tests.py
-
-valgrind-python.supp:
-	wget http://svn.python.org/projects/python/trunk/Misc/valgrind-python.supp
-
 inplace: discodop
 	# python setup.py build_ext --inplace
 	cp --update build/lib.*/discodop/*.so discodop/
 
 install: discodop docs
 
+debug:
+	# NB: debug build requires all external modules to be compiled
+	# with debug symbols as well (e.g., install python-numpy-dbg package)
+	python-dbg setup.py build_ext --inplace --debug --pyrex-gdb
+
+debugvalgrind: debug inplace
+	valgrind --tool=memcheck --leak-check=full --num-callers=30 \
+		--suppressions=valgrind-python.supp --show-leak-kinds=definite \
+		python-dbg tests.py
+
+valgrind: inplace
+	valgrind --tool=memcheck --leak-check=full --num-callers=30 \
+		--suppressions=valgrind-python.supp --show-leak-kinds=definite \
+		python tests.py
+
 # pylint: R=refactor, C0103 == Invalid name
 lint: inplace
 	# Any files with more than 999 lines?
 	cd discodop; wc -l *.py *.pyx *.pxi *.pxd | egrep '[0-9]{4,}'
 	# Docstrings without single line summaries?
 	cd discodop; egrep -n '""".*[^.\"\\)]$$' *.pxd *.pyx *.py || echo 'none!'
 	pep8 --ignore=E1,W1 \
-		discodop/*.py web/*.py tests/*.py && \
-	pep8 --ignore=E1,W1,F,E901,E225,E227,E211 \
-		discodop/*.pyx discodop/*.pxi && \
-	pylint --indent-string='\t' --disable=R,bad-continuation,invalid-name \
+		discodop/*.py web/*.py tests/*.py \
+	&& pep8 --ignore=E1,W1,F,E901,E225,E227,E211 \
+		discodop/*.pyx discodop/*.pxi \
+	&& pylint --indent-string='\t' --disable=R,bad-continuation,invalid-name \
 		discodop/*.py web/*.py tests/*.py

discodop/bit.pxd

@@ -168,7 +168,7 @@ cdef inline int iteratesetbits(ULong *vec, int slots,
 		``idx`` should be initialized to 0, and ``cur`` to the first element of
 		the bit array ``vec``, i.e., ``cur = vec[idx]``.
 	:returns: the index of a set bit, or -1 if there are no more set
-		bits. The result of calling stopped iterator is undefined.
+		bits. The result of calling a stopped iterator is undefined.
 
 	e.g.::
 
@@ -191,7 +191,7 @@ cdef inline int iteratesetbits(ULong *vec, int slots,
 
 cdef inline int iterateunsetbits(ULong *vec, int slots,
 		ULong *cur, int *idx):
-	""" Like ``iteratesetbits``, but return indices of zero bits. """
+	"""Like ``iteratesetbits``, but return indices of zero bits."""
 	cdef int tmp
 	while not ~cur[0]:
 		idx[0] += 1

discodop/parser.py

@@ -261,8 +261,11 @@ def workerfunc(func):
 	@wraps(func)
 	def wrapper(*args, **kwds):
 		"""Apply decorated function."""
-		import faulthandler
-		faulthandler.enable()  # Dump information on segfault.
+		try:
+			import faulthandler
+			faulthandler.enable()  # Dump information on segfault.
+		except ImportError:
+			pass
 		# NB: only concurrent.futures on Python 3.3+ will exit gracefully.
 		try:
 			return func(*args, **kwds)

discodop/runexp.py

@@ -126,6 +126,9 @@ def startexp(
 				corpusfmt, traincorpus, binarization, punct, functions,
 				morphology, removeempty, transformations,
 				relationalrealizational)
+	elif isinstance(traincorpus.numsents, float):
+		raise ValueError('need to specify number of training set sentences, '
+				'not fraction, in rerun mode.')
 
 	testset = treebank.READERS[corpusfmt](
 			testcorpus.path, encoding=testcorpus.encoding,
@@ -946,8 +949,11 @@ def test():
 
 def main(argv=None):
 	"""Parse command line arguments."""
-	import faulthandler
-	faulthandler.enable()
+	try:
+		import faulthandler
+		faulthandler.enable()
+	except ImportError:
+		pass
 	if argv is None:
 		argv = sys.argv
 	if len(argv) == 1:

discodop/treedraw.py

@@ -533,7 +533,8 @@ def tikzmatrix(self, nodecolor='blue', leafcolor='red',
 		PDF can be produced with pdflatex. Uses TiKZ matrices meaning that
 		nodes are put into a fixed grid. Where the cells of each column all
 		have the same width."""
-		result = ['%% %s\n%% %s' % (self.tree, ' '.join(self.sent)),
+		result = ['%% %s\n%% %s' % (
+			self.tree, ' '.join(a or '' for a in self.sent)),
 			r'''\begin{tikzpicture}[scale=1, minimum height=1.25em,
 			text height=1.25ex, text depth=.25ex,
 			inner sep=0mm, node distance=1mm]''',
@@ -579,7 +580,8 @@ def tikznode(self, nodecolor='blue', leafcolor='red', funccolor='green'):
 
 		Nodes are drawn with the \\node command so they can have arbitrary
 		coordinates."""
-		result = ['%% %s\n%% %s' % (self.tree, ' '.join(self.sent)),
+		result = ['%% %s\n%% %s' % (
+			self.tree, ' '.join(a or '' for a in self.sent)),
 			r'''\begin{tikzpicture}[scale=0.75, minimum height=1.25em,
 			text height=1.25ex, text depth=.25ex,
 			inner sep=0mm, node distance=1mm]''',

tests.py

@@ -3,9 +3,7 @@
 import sys
 from doctest import testmod, NORMALIZE_WHITESPACE, REPORT_NDIFF
 from operator import itemgetter
-# do not import from current directory, since our compiled modules will only be
-# installed system-wide or in the user's equivalent.
-sys.path.pop(0)
+
 MODULES = """bit coarsetofine demos disambiguation estimates eval fragments
 		_fragments gen grammar lexicon kbest plcfrs pcfg tree treedist treedraw
 		treebank treebanktransforms treetransforms runexp""".split()