> This may make this change untenable as no-one will want a series of
> use_feature_xxx flags, one for each change we introduce to glob.glob.
> Unless we throw in every known expansion / matching right now and have a
> single use-extended-features flag.

This is another question.  glob.glob(pattern, glob.GLOB_XXX | glob.GLOB_YYY) looks not much shorter than glob.glob(pattern, use_xxx=True, use_yyy=True).  And not all features can be configured by only one bit.  See os.access() and os.utime() as examples for many options.  I hope that before 3.4 feature freeze we can change the interface, if needed.

I also believe that this feature may not be regarded as mature without any sort of escaping (which also should be off by default for compatibility).  And we need glob.escape() function for escaping arbitrary path.

Since it is quite a complicated function, we need more examples in the documentation, which show different corner cases, nesting  and interaction with wildcards.