def get_func_code(func):
""" Attempts to retrieve a reliable function code hash.
The reason we don't use inspect.getsource is that it caches the
source, whereas we want this to be modified on the fly when the
function is modified.
Returns
-------
func_code: string
The function code
source_file: string
The path to the file in which the function is defined.
first_line: int
The first line of the code in the source file.
Notes
------
This function does a bit more magic than inspect, and is thus
more robust.
"""
source_file = None
try:
code = func.__code__
source_file = code.co_filename
if not os.path.exists(source_file):
# Use inspect for lambda functions and functions defined in an
# interactive shell, or in doctests
source_code = ''.join(inspect.getsourcelines(func)[0])
line_no = 1
if source_file.startswith('<doctest '):
source_file, line_no = re.match(
r'\<doctest (.*\.rst)\[(.*)\]\>', source_file).groups()
line_no = int(line_no)
source_file = '<doctest %s>' % source_file
return source_code, source_file, line_no
# Try to retrieve the source code.
with open_py_source(source_file) as source_file_obj:
first_line = code.co_firstlineno
# All the lines after the function definition:
source_lines = list(islice(source_file_obj, first_line - 1, None))
return ''.join(inspect.getblock(source_lines)), source_file, first_line
except:
# If the source code fails, we use the hash. This is fragile and
# might change from one session to another.
if hasattr(func, '__code__'):
# Python 3.X
return str(func.__code__.__hash__()), source_file, -1
else:
# Weird objects like numpy ufunc don't have __code__
# This is fragile, as quite often the id of the object is
# in the repr, so it might not persist across sessions,
# however it will work for ufuncs.
return repr(func), source_file, -1
def _check_previous_func_code(self, stacklevel=2):
"""
stacklevel is the depth a which this function is called, to
issue useful warnings to the user.
"""
# First check if our function is in the in-memory store.
# Using the in-memory store not only makes things faster, but it
# also renders us robust to variations of the files when the
# in-memory version of the code does not vary
try:
if self.func in _FUNCTION_HASHES:
# We use as an identifier the id of the function and its
# hash. This is more likely to falsely change than have hash
# collisions, thus we are on the safe side.
func_hash = self._hash_func()
if func_hash == _FUNCTION_HASHES[self.func]:
return True
except TypeError:
# Some callables are not hashable
pass
# Here, we go through some effort to be robust to dynamically
# changing code and collision. We cannot inspect.getsource
# because it is not reliable when using IPython's magic "%run".
func_code, source_file, first_line = self.func_code_info
func_id = _build_func_identifier(self.func)
try:
old_func_code, old_first_line =\
extract_first_line(
self.store_backend.get_cached_func_code([func_id]))
except (IOError, OSError): # some backend can also raise OSError
self._write_func_code(func_code, first_line)
return False
if old_func_code == func_code:
return True
# We have differing code, is this because we are referring to
# different functions, or because the function we are referring to has
# changed?
_, func_name = get_func_name(self.func, resolv_alias=False,
win_characters=False)
if old_first_line == first_line == -1 or func_name == '<lambda>':
if not first_line == -1:
func_description = ("{0} ({1}:{2})"
.format(func_name, source_file,
first_line))
else:
func_description = func_name
warnings.warn(JobLibCollisionWarning(
"Cannot detect name collisions for function '{0}'"
.format(func_description)), stacklevel=stacklevel)
# Fetch the code at the old location and compare it. If it is the
# same than the code store, we have a collision: the code in the
# file has not changed, but the name we have is pointing to a new
# code block.
if not old_first_line == first_line and source_file is not None:
possible_collision = False
if os.path.exists(source_file):
_, func_name = get_func_name(self.func, resolv_alias=False)
num_lines = len(func_code.split('\n'))
with open_py_source(source_file) as f:
on_disk_func_code = f.readlines()[
old_first_line - 1:old_first_line - 1 + num_lines - 1]
on_disk_func_code = ''.join(on_disk_func_code)
possible_collision = (on_disk_func_code.rstrip() ==
old_func_code.rstrip())
else:
possible_collision = source_file.startswith('<doctest ')
if possible_collision:
warnings.warn(JobLibCollisionWarning(
'Possible name collisions between functions '
"'%s' (%s:%i) and '%s' (%s:%i)" %
(func_name, source_file, old_first_line,
func_name, source_file, first_line)),
stacklevel=stacklevel)
# The function has changed, wipe the cache directory.
# XXX: Should be using warnings, and giving stacklevel
if self._verbose > 10:
_, func_name = get_func_name(self.func, resolv_alias=False)
self.warn("Function {0} (identified by {1}) has changed"
".".format(func_name, func_id))
self.clear(warn=True)
return False
