Apply modified version of Andrew Dalke's patch in #874 to create a quicker-loading numpy.

2 files changed, 39 insertions, 14 deletions

diff --git a/numpy/lib/_datasource.py b/numpy/lib/_datasource.py
index 1201f3d7e..3055bf47a 100644
--- a/numpy/lib/_datasource.py
+++ b/numpy/lib/_datasource.py
@@ -36,21 +36,39 @@ __docformat__ = "restructuredtext en"
 
 import os
 from shutil import rmtree
-from urlparse import urlparse
 
-# TODO: .zip support, .tar support?
-_file_openers = {None: open}
-try:
-    import bz2
-    _file_openers[".bz2"] = bz2.BZ2File
-except ImportError:
-    pass
-try:
-    import gzip
-    _file_openers[".gz"] = gzip.open
-except ImportError:
-    pass
+# Using a class instead of a module-level dictionary
+# to reduce the inital 'import numpy' overhead by 
+# deferring the import of bz2 and gzip until needed
 
+# TODO: .zip support, .tar support?
+class _FileOpeners(object):
+    def __init__(self):
+        self._loaded = False
+        self._file_openers = {None: open}
+    def _load(self):
+        if self._loaded:
+            return
+        try:
+            import bz2
+            self._file_openers[".bz2"] = bz2.BZ2File
+        except ImportError:
+            pass
+        try:
+            import gzip
+            self._file_openers[".gz"] = gzip.open
+        except ImportError:
+            pass
+        self._loaded = True
+
+    def keys(self):
+        self._load()
+        return self._file_openers.keys()
+    def __getitem__(self, key):
+        self._load()
+        return self._file_openers[key]
+
+_file_openers = _FileOpeners()
 
 def open(path, mode='r', destpath=os.curdir):
     """Open ``path`` with ``mode`` and return the file object.
@@ -179,6 +197,9 @@ class DataSource (object):
 
     def _isurl(self, path):
         """Test if path is a net location.  Tests the scheme and netloc."""
+        
+        # We do this here to reduce the 'import numpy' initial import time.
+        from urlparse import urlparse
 
         # BUG : URLs require a scheme string ('http://') to be used.
         #       www.google.com will fail.
@@ -276,6 +297,9 @@ class DataSource (object):
             `open` : Method that downloads and opens files.
 
         """
+        # We do this here to reduce the 'import numpy' initial import time.
+        from urlparse import urlparse
+
 
         # TODO:  This should be more robust.  Handles case where path includes
         #        the destpath, but not other sub-paths. Failing case:
diff --git a/numpy/lib/format.py b/numpy/lib/format.py
index 4192e1225..32a23ae18 100644
--- a/numpy/lib/format.py
+++ b/numpy/lib/format.py
@@ -57,7 +57,6 @@ means there is 1 element) by dtype.itemsize.
 """
 
 import cPickle
-import struct
 
 import numpy
 from numpy.lib.utils import safe_eval
@@ -180,6 +179,7 @@ def write_array_header_1_0(fp, d):
         This has the appropriate entries for writing its string representation
         to the header of the file.
     """
+    import struct
     header = ["{"]
     for key, value in sorted(d.items()):
         # Need to use repr here, since we eval these when reading
@@ -230,6 +230,7 @@ def read_array_header_1_0(fp):
     """
     # Read an unsigned, little-endian short int which has the length of the
     # header.
+    import struct
     hlength_str = fp.read(2)
     if len(hlength_str) != 2:
         msg = "EOF at %s before reading array header length"