List of possible bugs found by automated code analysis #17012

I ran an automated code analysis by https://www.deepcode.ai/ and the AI found the following bugs. I'm not sure if all of them are serious or important but if two or three code improvements can be made, that's something at least.

• 1.

  numpy/numpy/core/code_generators/genapi.py

  Line 309 in e3c5213

  with open(filename, 'w') as fid:

  "Use binary mode in open [:309] (current mode is w [:309] ) to avoid encoding-related issues for written file, on Windows or with Python 3."
  • See Change mode of open in func write_file() #17054 (withdrawn)

• 2.

  numpy/numpy/lib/polynomial.py

  Line 162 in e3c5213

  if NX.all(NX.sort(roots) == NX.sort(roots.conjugate())):

  "Use sorted(). sort [:162] sorts in-place and returns None. == [:162] expects a value."
  • np.sort returns the sorted array and the comparison works elementwise. The sorted builtin applied to a 1D array returns a list and errors for multidimensional arrays.

• 3.

  numpy/numpy/core/src/umath/umathmodule.c

  Line 203 in e3c5213

  strcpy(newdocstr, docstr);

  "The result of malloc [:202] , which may return null flows [:202, :203] to the first argument [:203] of strcpy [:203] . This could result in undefined behavior. Consider adding a check for nullness."
  • Fixed in MAINT: Improve error handling in umathmodule setup #17014

• 4.

  numpy/numpy/core/src/multiarray/nditer_api.c

  Line 2228 in e3c5213

  (transfersize/reduce_innersize <=

  "This division [:2228] may result in a division by zero error because 0 [:2046] is a possible divisor."
  • @seberg: I can't see this actually happening. If this would be 0, the buffer and thus the iteration size should be 0, in which case it is invalid to actually iterate.

• 5.

  numpy/numpy/linalg/lapack_lite/f2c_lapack.c

  Line 708 in e3c5213

  } else {

  "If statement has identical branches."
  • @charris: Generated code

• 6.

  numpy/numpy/linalg/lapack_lite/f2c_lapack.c

  Line 1071 in e3c5213

  if (ret_val == 1) {

  "The expression will always evaluate to true because both sides always hold the same value."
  • @charris: Generated code

• 7.

  numpy/numpy/random/src/pcg64/pcg64.h

  Line 234 in e3c5213

  uint64_t threshold = -bound % bound;

  "Applying unary minus to an unsigned expression. Consider casting the operand to a signed expression before applying minus [:234] ."

• 8. https://github.com/numpy/numpy/blob/e3c52130add2ed5f7f0b9442b3b2742c09eb2de0/benchmarks/benchmarks/common.py#L28-34 "Accessing attribute append [:32] on possibly None value. None flows [:32] as determined by the following check [:31] . Consider adding a check."

  • @eric-wieser: this is clearly incorrect analysis when looking at the full context

• 9.

  numpy/numpy/distutils/fcompiler/gnu.py

  Line 402 in e3c5213

  h = hashlib.sha1()

  " hashlib.sha1 [:402] is insecure. Consider changing it to a secure hashing algorithm (e.g. SHA256)."

• 10.

  numpy/numpy/f2py/crackfortran.py

  Line 917 in e3c5213

  groupname[groupcounter] = 'program'

  "Trying to store [:917] a value in element of immutable type str [:182] (from str literal "" [:182] ) will lead to a crash."

• 11.

  numpy/numpy/polynomial/chebyshev.py

  Line 682 in e3c5213

  if len(c) == 1 and c[0] == 0:

  "Trying to call len [:682] on an object of type 0 (0) will lead to a TypeError."

• 12.

  numpy/numpy/distutils/command/build_ext.py

  Line 562 in e3c5213

  for lib in list(libraries):

  "There is no need to call list [:562] on an object [:562] that is already constructed as a list [:557] ."
  • See MAINT: Delete obsolete conversion to list #17052

• 13.

  numpy/numpy/distutils/misc_util.py

  Line 1903 in e3c5213

  for line in open(branch_cache_fn, 'r'):

  "Missing close for open [:1903] , add close or use a with block."
  • See MAINT: Change open() to with open() #17051

• 14.

  numpy/numpy/distutils/fcompiler/gnu.py

  Line 26 in e3c5213

  if is_win64():

  "If statement has identical branches."
  • See MAINT: Remove _EXTRAFLAGS variable #17050

• 15.

  numpy/numpy/distutils/system_info.py

  Line 718 in e3c5213

  found = map(lambda opt: self.cp.has_option(self.section, opt), options)

  "Use comprehensions instead of list [:719] ( map [:718] (lambda: x ...))"
  • See MAINT: Replace lambda function by list comprehension #17055

• 16.

  numpy/numpy/lib/polynomial.py

  Line 1020 in e3c5213

  truepoly = (isinstance(u, poly1d) or isinstance(u, poly1d))

  "Duplicate expressions on both sides of a logical operator [:1020] is probably a mistake."
  • See BUG: fix typo in polydiv that prevented promotion to poly1d #17053

• 17.

  numpy/numpy/polynomial/hermite_e.py

  Line 426 in e3c5213

  prd = np.empty(len(c) + 1, dtype=c.dtype)

  "Trying to load attribute dtype [:426] on an object of primitive type int [:485] (from int literal 0 [:485] ) will lead to an AttributeError."

• 18.

  numpy/tools/openblas_support.py

  Line 227 in e3c5213

  os.mkdir(target)

  "Use os.makedirs instead of os.mkdir [:227] because the given path may require creating the parent directories."
  • See MAINT: Change os.mkdir to os.makedirs #17061 (withdrawn)

• 19.

  numpy/tools/refguide_check.py

  Line 391 in e3c5213

  dep_in_ref = set(only_ref).intersection(deprecated)

  "There is no need to call set [:391] on an object [:391] that is already constructed as a set [:319] "
  • See MAINT: Remove obsolete conversion to set #17063

• 20.

  numpy/numpy/core/src/multiarray/iterators.c

  Line 517 in e3c5213

  Py_DECREF(ret);

  "Use Py_XDECREF instead of Py_DECREF [:517] "
  • As pointed out by Eric, obviously cannot be NULL, its checked a few lines earlier.

• 21.

  numpy/numpy/core/src/multiarray/methods.c

  Line 2603 in e3c5213

  err = PyString_FromString("unable to convert ");

  " PyString_FromString [:2603] does not work with Python 3.5, use PyUnicode_FromString instead"
  • See MAINT, BUG: Remove uses of PyString_FromString. #17068 for fixes.

• 22.

  numpy/numpy/core/src/multiarray/strfuncs.c

  Line 89 in e3c5213

  ostring = PyString_AsString(sp);

  "Use PyBytes_AsString instead of PyString_AsString [:89] for Python 3 compatibility"
  • See MAINT: Make arrayprint str and repr the ndarray defaults. #17141 for fix.

• 23.

  numpy/numpy/f2py/src/fortranobject.c

  Line 268 in e3c5213

  if (n < 0) {

  "Numerical operation lessThan [:268] used with boolean operand equals [:267] . Consider making the conversion from bool to integer explicit."

• 24.

  numpy/numpy/core/src/multiarray/nditer_constr.c

  Line 829 in e3c5213

  memset(axes_dupcheck, 0, NPY_MAXDIMS);

  "Be aware that an optimizing compiler might be removing this call to memset [:829] , since this axes_dupcheck array [:829] does not seem to be used after the call to memset [:829]."

• 25.

  numpy/numpy/random/src/legacy/legacy-distributions.c

  Line 41 in e3c5213

  if (shape == 1.0) {

  "Equals comparison of floating point numbers. Floating point numbers are only considered equal if their bit representation is equal. Consider using an epsilon comparison instead."

• 26.

  numpy/numpy/f2py/tests/src/array_from_pyobj/wrapmodule.c

  Line 152 in e3c5213

  s = PyString_FromString("This module 'wrap' is auto-generated with f2py (version:2_1330).\nFunctions:\n"

  "PyString_FromString [:152] does not work with Python 3.5, use PyUnicode_FromString instead"
  • See MAINT, BUG: Remove uses of PyString_FromString. #17068 for fixes.

• 27.

  numpy/numpy/f2py/tests/src/array_from_pyobj/wrapmodule.c

  Line 59 in e3c5213

  Py_DECREF(tmp);

  "Use Py_XDECREF instead of Py_DECREF [:59] "

• 28.

  numpy/numpy/lib/_datasource.py

  Line 267 in e3c5213

  return ext in _file_openers.keys()

  "No need to call keys [:267] , directly check with the in operator."
  • See MAINT: Simplify check for key in dictionary #17064 (needs refactoring/rewriting)

• 29.

  numpy/numpy/core/tests/test_array_coercion.py

  Line 494 in e3c5213

  return super().__len__()

  "Use len() instead of calling len [:494] directly"

• 30.

  numpy/numpy/core/tests/test_numeric.py

  Line 1376 in e3c5213

  a = rand(3, 5, 8)

  "Use predictable random with seed or secure random."

• 31.

  numpy/numpy/core/tests/test_scalarmath.py

  Line 496 in e3c5213

  assert_(not np.float32(1) == None)

  "Use the is operator instead of == [:496] to compare to None [:496] , because == [:496] calls eq which is slower and error-prone."

• 32.

  numpy/numpy/core/tests/test_shape_base.py

  Line 158 in e3c5213

  hstack(map(lambda x: x, np.ones((3, 2))))

  "Use comprehensions instead of map [:158] ."
  • See MAINT: Convert lambda func to list comprehension #17065 (withdrawn)

• 33.

  numpy/numpy/lib/_datasource.py

  Line 580 in e3c5213

  DataSource.__del__(self)

  "Use del statement instead of calling del [:580] directly"

  • @charris The current code is recommended when both the base and derived class define the __del__ method. That said, it isn't clear that an explicit __del__ is needed, the inherited version should serve.

    Note del x doesn’t directly call x.del() — the former decrements the reference count
    for x by one, and the latter is only called when x’s reference count reaches zero.

• 34.

  numpy/numpy/lib/tests/test__datasource.py

  Line 350 in e3c5213

  ds.__del__()

  "Use del statement instead of calling del [:350] directly"
  • @charris Calling the function directly insures that it is immediately run at that spot in the test rather than possibly
    waiting for the reference count to go to zero. Both options should work here, but I think the direct call is better
    because it is more explicit.

• 35.

  numpy/numpy/lib/tests/test_io.py

  Line 2581 in e3c5213

  assert_('x' in z.keys())

  "No need to call keys [:2581] , directly check with the in operator."