Xgboost: Python: Booster.load_model() should not require buffer writeability

[aldanor]

It is impossible to load a model from a buffer (e.g. bytestring) in Python, due to a bug where ctypes is used incorrectly to get a const char* pointer:

>>> bst.load_model(b'123')
TypeError: underlying buffer is not writable

---

Here's a short example of what it is basically trying to do:

>>> b = b'123'
>>> (ctypes.c_char * 3).from_buffer(buf)
TypeError: underlying buffer is not writable

[aldanor]

See also: #542

[aldanor]

Basically, instead of using .from_buffer(), you should use .from_buffer_copy() which only requires readability.

[aldanor]

However, apparently, there is some problem with XGBoosterLoadModelFromBuffer() as it segfaults if you provide it with the dump buffer.

Here's a modified version that uses read-only buffer:

import ctypes
import xgboost
import xgboost.core

def xgb_load_model(buf):
    if isinstance(buf, str):
        buf = buf.encode()
    bst = xgboost.core.Booster()
    n = len(buf)
    length = xgboost.core.c_bst_ulong(n)
    ptr = (ctypes.c_char * n).from_buffer_copy(buf)
    xgboost.core._check_call(
        xgboost.core._LIB.XGBoosterLoadModelFromBuffer(bst.handle, ptr, length)
    )  # segfault
    return bst

[hcho3]

Consolidating to #3439. This issue should be re-opened if you or others decide to actively work on implementing this feature.