Support 'ORDER BY' about datajoint-matlab HOT 10 CLOSED

Yeah, I find myself implementing the following pattern quite often, which could be much easier if ORDER BY was used:

result = fetch(rel, '*');
result = dj.struct.order(result, 'foo');
a = [result.a];
b = [result.b];
c = [result.c];

instead of, for example, a simpler [a, b, c] = fetchsorted(rel, 'foo', 'a', 'b', 'c').

Sorting should be limited to happen at fetch time. Anything else would probably break (or at least complicate a lot) the relational model.

from datajoint-matlab.

We could use the fact that all attributes in DataJoint are lower case. Upper case strings will be interpreted as parameter names. So Alex's code above will look like:

[a,b,c] = rel.fetchn(rel, 'a', 'b', 'c', 'SORT BY', {'foo'})

Or we could use a structure to pass parameters into fetch:

opt.sortby = {'foo'}
[a,b,c] = rel.fetchn(rel, 'a', 'b', 'c', opt)

Preferences? I think the second way looks more graceful.

from datajoint-matlab.

okay, this will let us to implement limits more gracefully too:

clear opt
opt.sortby = {'foo'};
opt.start = 1000;
opt.limit = 100;
[a,b,c] = rel.fetchn('a','b','c', opt)

from datajoint-matlab.

I prefer you last suggestion. If sortby could also be a string in case of a single attribute that would be even better :-)
And please let's keep the old way of doing limits for a while before throwing it out.

from datajoint-matlab.

Wouldn't the first suggestion be better? Aas the string we specify is appened at the end were passed we could easily write our own:

[a,b,c] = rel.fetchn(rel, 'a', 'b', 'c', 'SORT BY "a" LIMIT 500')

from datajoint-matlab.

All fetch operators now handle the last argument differently if it begins with "ORDER BY" or "LIMIT ". In both cases, it is simply appended to the SELECT statement.

Examples:

[a,b,c] = rel.fetchn('a','b','c', 'ORDER BY a*b DESC LIMIT 3 OFFSET 5')
[a,b,c] = rel.fetchn('a','b','c', 'LIMIT 3 OFFSET 5')

I will update the documentation after a bit of testing.

from datajoint-matlab.

The old fetch*(..., 1) syntax for limits is broken now...

from datajoint-matlab.

I wasn't familiar with that syntax. That just limits the length by one without ordering? That seems like it's essentially "return a random key" given we aren't meant to make assumptions about ordering.

from datajoint-matlab.

Yes, but often a random key is just fine (e.g. if you know that a bunch of rows all have the same value) and I'm using it at several places and don't feel like refactoring all my code again. That's why I specifically asked to not break that syntax in the comment above :(

from datajoint-matlab.

I just enabled the old syntax for backward compatibility. I only did it for limits, not offsets. So fetch(..., 10, 10) will not work, but fetch (...., 10) will work.

I would argue that it makes more sense to go back and replace all instances of fetch(....., 1) with fetch(....., 'LIMIT 1') for better readability anyway.

Relevant http://xkcd.com/1172/

from datajoint-matlab.