[Numpy-discussion] The future of ndarray.diagonal()

[Matthew Brett]

Hi,

On Sun, Jan 4, 2015 at 4:22 PM, Konrad Hinsen
<konrad.hinsen at fastmail.net> wrote:
> 2) Changes that yield to different results for unmodified legacy code
> should never be allowed.

I think this is a very reasonable rule.

Case in point - I have some fairly old code in
https://github.com/matthew-brett/transforms3d .

I haven't updated this code since 2011.   Now I test it, I get the
following warning:

======================================================================
ERROR: Failure: ValueError (assignment destination is read-only)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/mb312/.virtualenvs/scipy-devel/lib/python2.7/site-packages/nose/loader.py",
line 251, in generate
    for test in g():
  File "/Users/mb312/dev_trees/transforms3d/transforms3d/tests/test_affines.py",
line 74, in test_rand_de_compose
    T, R, Z, S = func(M)
  File "/Users/mb312/dev_trees/transforms3d/transforms3d/affines.py",
line 298, in decompose
    Z[0] *= -1
ValueError: assignment destination is read-only

If I had waited until 1.10 (or whatever) - I would have had to hope
that my tests were good enough to pick this up, otherwise anyone using
this code would be subject to some very strange bugs.

Cheers,

Matthew