PERF: Index._shallow_copy doesn't copy ._engine

[topper-123]

>>> idx = pd.Index(np.arange(100_000))
>>> %timeit idx.get_loc(99_999)
774 ns ± 26 ns per loop
>>> %timeit idx._shallow_copy().get_loc(99_999)
3.57 ms ± 56.8 µs per loop

The same performance issue can be seen on other index types, e.g. CategoricalIndex and MultiIndex.

Problem description

The reason for the above diferences is that _shallow_copy does not copy over the ._engine attribute to the new index and the _engine is expensive to recreate.

Indexes are immutable, and likewise - to my understanding - are the ._engine attribute of indexes. The ._engine is quite expensive to create and if it has been created on the original index, I think it should be possible to reuse it on the new index, saving ther time needed to create a new and identical ._engine.

_shallow_copy is used in a few places internally in pandas, so there seems to be potential for some speedups for several pandas merhods by copying the _engine over to newly-copied indexes.

Possibly I'm missing some finer details here, e.g. don't know what the ._engine.clear_mappings is for and it seems from its name to be destructive, but overall it seems to be possible to make a change to copy this over.

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 79663fb
python : 3.7.3.final.0
python-bits : 64
OS : Windows
OS-release : 10
machine : AMD64
processor : Intel64 Family 6 Model 78 Stepping 3, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None

pandas : 0.25.0.dev0+1363.g79663fb66
numpy : 1.16.4
pytz : 2019.1
dateutil : 2.8.0
pip : 19.1.1
setuptools : 41.0.1
Cython : 0.29.13
pytest : 5.0.1
hypothesis : 4.28.2
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : 7.6.1
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : None
matplotlib : None
numexpr : 2.6.9
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
xarray : None
xlrd : None
xlwt : None
xlsxwriter : None

[TomAugspurger]

I suspect we might want to copy over other cached attributes in Index._cache (is_unique, inferred_type, etc) if this proves safe to do.

[jbrockmendel]

shallow_copy generally has new values passed to it. My intuition is that would make it impossible to share an engine. What am i missing?

[topper-123]

shallow_copy generally has new values passed to it. My intuition is that would make it impossible to share an engine. What am i missing?

I was thinking implicitly that reusing _engine would only be done when parameter values is None and the old values will be reused. If the values are reused, the _engine could be reused, if the engine is immutable, wouldn't you agree?