BUG: integrate: Fix crash with repeated t values in odeint. #8822
Conversation
WarrenWeckesser
added
scipy.integrate
defect
|
|
||
| dt = np.diff(t) | ||
| if not((dt >= 0).all() or (dt <= 0).all()): | ||
| raise ValueError("The values in t must be monotonically increasing " |
There was a problem hiding this comment.
I guess, since we allow for repeated values, this should read as must either all nondecreasing or all nonincreasing but not a mixture. or something along those lines.
There was a problem hiding this comment.
By every definition that I have seen, the statement "t is monotonically increasing" means t[i] <= t[i+1] for all i, so repeated values are allowed.
There was a problem hiding this comment.
But I guess it wouldn't hurt to state explicitly that repeated values are allowed.
There was a problem hiding this comment.
I also thought that "monotonically increasing" was < and "monotonically nondecreasing was <= but upon reading the definitions this does indeed appear mistaken! I agree adding a (t[i] <= t[i+1] for all i) sort of parenthetical is worth it.
There was a problem hiding this comment.
I added a note to the docstring and the error message about allowing repeated values.
|
Looks good to me apart from a tiny detail. |
Two changes to odeint: * Require `t` to be monotonically increasing or monotonically decreasing. Repeated values are allowed. * If the initial value in `t` is repeated, don't bother calling LSODA for those times. Just copy the initial condition to the output. This fixes the problem where an input such as `t = np.zeros(10)` would abort the Python interpreter (because of how the Fortran subroutine LSODA handled the repeated calls with the same value of t). Closes scipygh-8217.
WarrenWeckesser
force-pushed
the
odeint-crash-fix
branch
from
af7d9e1
to
0ec9261
Compare
|
Thanks, issues seemed addressed. |
Two changes to odeint:
tto be monotonically increasing or monotonicallydecreasing. Repeated values are allowed.
tis repeated, don't bother callingLSODA for those times. Just copy the initial condition to the
output.
This fixes the problem where an input such as
t = np.zeros(10)would abort the Python interpreter (because of how the Fortran
subroutine LSODA handled the repeated calls with the same value
of t).
Closes gh-8217.