MAINT/TST: Tidy dtype indexing #9947
Conversation
…tions Benefits of new structure: * Checking of fields doesn't happen twice * No longer any fear of possible recursion * Makes it easy to add sq_item later Also tidies initialization of the PySequenceMethods struct
09c4dd5 to
9271d81
Compare
| static PyObject * | ||
| _subscript_by_index(PyArray_Descr *self, Py_ssize_t i) | ||
| { | ||
| PyObject *name = PySequence_GetItem(self->names, i); |
There was a problem hiding this comment.
Using PySequence_GetItem here means we don't have to implement our own handling of negative indices
| int value = PyArray_PyIntAsInt(op); | ||
| int orig_value = value; | ||
|
|
||
| Py_ssize_t i = PyArray_PyIntAsIntp(op); |
There was a problem hiding this comment.
Changed to Py_ssize_t to avoid compiler warnings.
| return NULL; | ||
| } | ||
| name = PyTuple_GET_ITEM(self->names, value); | ||
| retval = descr_subscript(self, name); |
There was a problem hiding this comment.
This recursion was a strange way to handle this, and caused the field check to occur twice.
Really that's just because we're too lazy to call EDIT: Actually, this is not about laziness - the problem is that if you hit an exception while trying to build an exception message in the C code, you end up with no stack trace, and have no idea what actually went wrong. |
|
Don't worry about it - or at least not just for this piece - it would seem something that could be cleaned up more generally. This one looks all OK and tests passed, so I'll merge. |
Previously
descr_subscriptwould recurse, which was needlessly confusing.Extracted from #8814, which fell into the trap of combining ENH with MAINT.