Bugs in the latest GPAW¶

See here: Bugs!

Handling segfaults¶

Segmentation faults are probably the hardest type of runtime error to track down, but they are also quite common during the unstable part of the release cycle. As a rule of thumb, if you get a segfault, start by checking that all array arguments passed from Python to C functions have the correct shapes and types.

Apart from running GPAW in debug mode, please familiarize yourself with the debugging tools for the Python and C code.

If you experience segfaults or unexplained MPI crashes when running GPAW in parallel, it is recommended to try a custom installation with a debugging flag in siteconfig.py:

define_macros += [('GPAW_MPI_DEBUG', 1)]

Common sources of bugs¶

General:
- Elements of NumPy arrays are C ordered, BLAS and LAPACK routines expect Fortran ordering.
Python:
- Always give contiguous arrays to C functions. If x is contiguous with dtype=complex, then x.real is non-contiguous of dtype=float.
- Giving array arguments to a function is a carte blanche to alter the data:
```
def double(a):
    a *= 2
    return a

x = np.ones(5)
print(double(x))  # x[:] is now 2.
```
- Forgetting a n += 1 statement in a for loop:
```
n = 0
for thing in things:
    thing.do_stuff(n)
    n += 1
```
  Use this instead:
```
for n, thing in enumerate(things):
    thing.do_stuff(n)
```
- Indentation errors like this one:
```
if ok:
    x = 1.0
else:
    x = 0.5
    do_stuff(x)
```
  where do_stuff(x) should have been reached in both cases. Emacs: always use C-c > and C-c < for shifting in and out blocks of code (mark the block first).
- Don’t use mutables as default values:
```
class A:
    def __init__(self, a=[]):
        self.a = a  # all instances get the same list!
```
- There are subtle differences between x == y and x is y.
- If H is a numeric array, then H - x will subtract x from all elements - not only the diagonal, as in Matlab!
C:
- Try building GPAW from scratch.
- Typos like if (x = 0) which should have been if (x == 0).
- Remember break in switch-case statements.
- Check malloc-free pairs. Test for memory leaks by repeating the call many times.
- Remember to update reference counts of Python objects.
- Never put function calls inside assert’s. Compiling with -DNDEBUG will remove the call.