FIX: hdfstore queries of the form where=[('date', '>=', datetime(2013,1,...

[hhuuggoo]

FIX: hdfstore queries of the form where=[('date', '>=', datetime(2013,1,1)), ('date', '<=', datetime(2014,1,1))] were broken
- modified Expr.parse_back_compat to check for tuples, in w, and unpack into w, op, value
- modified Expr.__init__ to modify the where list/tuple with the parsed result

[jreback]

hmm

did u check master with this; I thought I had fixed this already

[jreback]

would also need some tests for this

[hhuuggoo]

It broke for me in master, I'll add some tests though. But first - is this the right way to fix it? I don't know the pandas codebase that well

[jreback]

where u made the change is fine
their are already a bunch of tests for the back compat formats
so just add them their (u can add as another test function)
need to make sure that fix doesn't break anything

[jreback]

this is fine...can you add a release note (in bug section of 0.14; reference this PR number)

[jreback]

@hhuuggoo can you add a release note, otherwise looks ready to go

[jreback]

can you rebase...

and add a test that does an assert_produces_warning...that validates that you are in fact showing the warning int he correct place

thanks

[hhuuggoo]

I did it - however something wierd happens. The test is fine if ran with others, however if I run it on it's own using the following:

nosetests -s pandas/io/tests/test_pytables.py:TestHDFStore.test_backwards_compat_without_term_object

AssertionError: Caused unexpected warning(s): ['PendingDeprecationWarning'].

I see some comments about this in the test as well, so it seems like it's a persistent problem?

[hhuuggoo]

@jreback btw, do you know why the query syntax changed? Is there some discussion of this online I can read?

[jreback]

that deprecation warning is a weird thing...I have seen it using assert_produces_warning - and I can't narrow down why it actually happens

what do you mean why the query syntax changed? The original 'syntax' was just a hack...you could do anything beyond anding and too much verbiage to do anything...

is their a problem?

[jreback]

there was a whole very long discussion on this...search for 'eval' in the issues....

[hhuuggoo]

@jreback, no problem for me, but personally I prefer the old syntax, because it's easier to programatically build up queries using lists and tuples, rather than doing string operations. Not a problem, I can always pull out the backwards compat code in my application if I really need to.

anyways I think this PR is done?

[jreback]

@hhuuggoo thanks!

its not that difficult to put in a combiner to handle built up queries (of say list of sub-queries)...
you may need a wrapper class (which basically does what the older code did, but ends up stringifying)

I think that is done now (but maybe only if you pass Terms...can't remember)

interested?