Extended search functionality

[joahim]

tripper/tripper/datadoc/dataset.py

Line 1399 in 17c0093

def search(

The existing search function does not provide means to:

1. search for a value across all objects (case with missing predicate value)
2. search for existence of predicate (case with missing object value)

While 2. may not be needed (or could even be hard for users to understand), 1. is valuable for users since it facilitates sort of "full text search" where the user does not necessarily need to know the predicate value but knows the object value (e.g. title).

The API of the search call could be modified to support this in the following way: instead of a dictionary to pass in parameters, use a list of tuples (predicate, object). Passing None for predicate value supports case 1 above. Passing None for object value supports case 2 above. Using this approach would also allow for cases where a user wants to search for multiple objects for single predicate (e.g. dc:title=Foo | dc:title=Bar).

Additionally, the existing criterias and regex parameters could be merged into one by using a list of tuples (predicate, object, regex) where regex: bool = False would signify whether to use regular expression search or not for a given criteria pair.

[jesper-friis]

How would you like the syntax for these cases? Case 2. is easy, there we can just provide None as the value to the criteria argument.

How about case 1? We can only have one None key in a dict.
One possibility could by to map none to a list of objects: criteria={None: ["obj1", "obj2"]}. This could also be extended to other predicates.

[joahim]

As mentioned above, my preference is for a list of tuples which, in my view, best matches the intent (compare [(None, "obj1")] vs {None: ["obj1", "obj2"]}). Also, a list is an order preserving structure and a dict is not, which could become useful for a stateless clients when returning errors, for example.

This type is also easily extendable into triples (predicate, object, use_regex), eliminating somewhat artificial split between two kinds of queries in the current API (prioritize query vs. search method).

[francescalb]

I added the possibility to using None as either key or value above, as well as adding criteria as a list of tuples. However, I just extended on the current implementation. Do we need to reimplement to have list of tuples as main implementation? @jesper-friis @joahim , input welcome.