diff options
| author | Sebastian Berg <sebastian@sipsolutions.net> | 2021-12-15 16:58:34 -0600 |
|---|---|---|
| committer | Sebastian Berg <sebastian@sipsolutions.net> | 2021-12-15 17:01:10 -0600 |
| commit | 6c5a9ac9681dba754728d40e3d2ea0fe4c10abcc (patch) | |
| tree | f03fc828eeb2238f09f2527f54c982f62029c491 | |
| parent | 43c113dd7aa36ef833315035858ea98a7b4732c5 (diff) | |
| download | numpy-6c5a9ac9681dba754728d40e3d2ea0fe4c10abcc.tar.gz | |
MAINT: Check for buffer interface support rather than try/except
This checks for buffer interface support using `PyObject_CheckBuffer`
up-front. That seems to shave off about 100ns e.g. when importing
pandas DataFrames (which do not support the buffer interface, but
do support `__array__`.
The reason is that building a TypeError and then clearing it takes
time.
It felt like a clear, easy win, since the check should be so fast
that checking twice (in the case that it is indeed a buffer) does
not really matter.
But, if nobody likes to think about it just close :).
| -rw-r--r-- | numpy/core/src/multiarray/ctors.c | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/numpy/core/src/multiarray/ctors.c b/numpy/core/src/multiarray/ctors.c index 487f0d4a9..9b86fca4d 100644 --- a/numpy/core/src/multiarray/ctors.c +++ b/numpy/core/src/multiarray/ctors.c @@ -1301,9 +1301,10 @@ _array_from_array_like(PyObject *op, * We skip bytes and unicode since they are considered scalars. Unicode * would fail but bytes would be incorrectly converted to a uint8 array. */ - if (!PyBytes_Check(op) && !PyUnicode_Check(op)) { + if (PyObject_CheckBuffer(op) && !PyBytes_Check(op) && !PyUnicode_Check(op)) { PyObject *memoryview = PyMemoryView_FromObject(op); if (memoryview == NULL) { + /* TODO: Should probably not blanket ignore errors. */ PyErr_Clear(); } else { |