Handling of numbers under oracle is slow
|Reported by:||Shai Berger||Owned by:||Shai Berger|
|Component:||Database layer (models, ORM)||Version:||master|
|Cc:||Anssi Kääriäinen, Ian Kelly||Triage Stage:||Accepted|
|Has patch:||yes||Needs documentation:||no|
|Needs tests:||no||Patch needs improvement:||yes|
With current code, for all number columns, the Oracle backend asks cx_Oracle to return the numbers as strings, and then transforms them back to numbers on the way out. All this is done because Oracle supports higher precision (and a wider range of integers) than the Python
float type; some values, if they are to be reported accurately, must be returned as
decimal.Decimals, and cx_Oracle doesn't apparently do that on its own.
With cx_Oracle 5, though, it is possible to use a much better approach, on a per-column (rather than per-row) basis, selecting the right output type when there is enough information to do so, and making most of the decisions once-per-query instead of once-per-row.
An original patch for this was written by Ian Kelly; I improved it and made a pull-request in the stretch towards 1.5, but there was no ticket for it and so it was never merged. I hope we can finally merge it into 1.6.
There was a little saga of discussions around this (mostly Anssi, Ian and myself, https://groups.google.com/d/topic/django-developers/4BNkJyGez9A/discussion). Some other issues are also mentioned there. But this is the main one.
So now there is a ticket. Things changed, and the pull-request needed updating. The new one is at https://github.com/django/django/pull/1071.