The addition of this code block to ColorbarBase._ticker in matplotlib 1.5.x:

        if isinstance(locator, ticker.LogLocator):
            eps = 1e-10
            b = b[(b <= intv[1] * (1 + eps)) & (b >= intv[0] * (1 - eps))]
        else:
            eps = (intv[1] - intv[0]) * 1e-10
            b = b[(b <= intv[1] + eps) & (b >= intv[0] - eps)]

can cause a numpy array scalar to be passed to ColorbarBase._locate but when that happens the np.searchsorted call in _locate returns an numpy.int64 scalar, which causes the indexing operations:

        i0[itop] -= 1
        ii[itop] -= 1
        i0[ibot] += 1
        ii[ibot] += 1

to generate the exception: 'numpy.int64' object does not support item assignment.

My quick and dirty patch for this was to add this line before the np.searchsorted call in _locate:

        if np.ndim(xn) == 0: xn = np.array([xn[()]])

I don't propose this as a formal patch since it may not be the most efficient fix and this issue may be broader than what I am seeing.

Specs:

• Windows 10 64-bit
• Python 2.7.11 64-bit (python.org installer)
• matplotlib 1.5.1 (pip install)
• Regression: OK in 1.4.3