BF: fix and test for bug in PAR header reading

Parsing should accept empty values in general header.
Window values appear to be floating point.
Add image kindly provided by cccbauer from UMass medical school.

Author: matthew-brett

COPYING

@@ -204,3 +204,8 @@ The files::
 
 are from https://github.com/yarikoptic/nitest-balls1, also released under the
 the PDDL version 1.0 available at http://opendatacommons.org/licenses/pddl/1.0/
+
+    nibabel/nibabel/tests/data/umass_anonymized.PAR
+
+is courtesy of the University of Massachusetts Medical School, also released
+under the PDDL.

nibabel/parrec.py

@@ -185,8 +185,9 @@
     ('rescale intercept', float),
     ('rescale slope', float),
     ('scale slope', float),
-    ('window center', int,),
-    ('window width', int,),
+    # Window center, width recorded as integer but can be float
+    ('window center', float,),
+    ('window width', float,),
     ('image angulation', float, (3,)),
     ('image offcentre', float, (3,)),
     ('slice thickness', float),
@@ -244,6 +245,9 @@ class PARRECError(Exception):
     pass
 
 
+# Value after colon may be absent
+GEN_RE = re.compile(r".\s+(.*?)\s*:\s*(.*)")
+
 def _split_header(fobj):
     """ Split header into `version`, `gen_dict`, `image_lines` """
     version = None
@@ -276,7 +280,6 @@ def _split_header(fobj):
     return version, gen_dict, image_lines
 
 
-GEN_RE = re.compile(r".\s+(.*?)\s*:\s+(.*)")
 
 
 def _process_gen_dict(gen_dict):

nibabel/tests/data/README.rst

@@ -6,3 +6,6 @@ Nibabel data files
 
 ``phantom_truncated.PAR`` is a slightly edited copy of
 ``phantom_EPI_asc_CLEAR_2_1.PAR``.
+
+``umass_anonymized.PAR`` courtesy of Github user ``cccbauer``, data from the
+University of Massachusetts medical school.