diff --git a/waflib/Node.py b/waflib/Node.py
index c1429590..4ac1ea8a 100644
--- a/waflib/Node.py
+++ b/waflib/Node.py
@@ -629,7 +629,7 @@ class Node(object):
 
 	def ant_glob(self, *k, **kw):
 		"""
-		Finds files across folders:
+		Finds files across folders and returns Node objects:
 
 		* ``**/*`` find all files recursively
 		* ``**/*.class`` find all files ending by .class
@@ -638,14 +638,51 @@ class Node(object):
 		For example::
 
 			def configure(cfg):
-				cfg.path.ant_glob('**/*.cpp') # finds all .cpp files
-				cfg.root.ant_glob('etc/*.txt') # matching from the filesystem root can be slow
-				cfg.path.ant_glob('*.cpp', excl=['*.c'], src=True, dir=False)
+				# find all .cpp files
+				cfg.path.ant_glob('**/*.cpp')
+				# find particular files from the root filesystem (can be slow)
+				cfg.root.ant_glob('etc/*.txt')
+				# simple exclusion rule example
+				cfg.path.ant_glob('*.c*', excl=['*.c'], src=True, dir=False)
 
-		For more information see http://ant.apache.org/manual/dirtasks.html
+		For more information about the patterns, consult http://ant.apache.org/manual/dirtasks.html
+		Please remember that the '..' sequence does not represent the parent directory::
+
+			def configure(cfg):
+				cfg.path.ant_glob('../*.h') # incorrect
+				cfg.path.parent.ant_glob('*.h') # correct
+
+		The Node structure is itself a filesystem cache, so certain precautions must
+		be taken while matching files in the build or installation phases.
+		Nodes objects that do have a corresponding file or folder are garbage-collected by default.
+		This garbage collection is usually required to prevent returning files that do not
+		exist anymore. Yet, this may also remove Node objects of files that are yet-to-be built.
+
+		This typically happens when trying to match files in the build directory,
+		but there are also cases when files are created in the source directory.
+		Run ``waf -v`` to display any warnings, and try consider passing ``remove=False``
+		when matching files in the build directory.
+
+		Since ant_glob can traverse both source and build folders, it is a best practice
+		to call this method only from the most specific build node::
+
+			def build(bld):
+				# traverses the build directory, may need ``remove=False``:
+				bld.path.ant_glob('project/dir/**/*.h')
+				# better, no accidental build directory traversal:
+				bld.path.find_node('project/dir').ant_glob('**/*.h') # best
+
+		In addition, files and folders are listed immediately. When matching files in the
+		build folders, consider passing ``generator=True`` so that the generator object
+		returned can defer computation to a later stage. For example::
+
+			def build(bld):
+				bld(rule='tar xvf ${SRC}', source='arch.tar')
+				bld.add_group()
+				gen = bld.bldnode.ant_glob("*.h", generator=True, remove=True)
+				# files will be listed only after the arch.tar is unpacked
+				bld(rule='ls ${SRC}', source=gen, name='XYZ')
 
-		The nodes that correspond to files and folders that do not exist are garbage-collected.
-		To prevent this behaviour in particular when running over the build directory, pass ``remove=False``
 
 		:param incl: ant patterns or list of patterns to include
 		:type incl: string or list of strings