[query] Full support for manipulating multivalue properties

[dantleech]

Adds support for updating multivalue properties with the UPDATE query:

UPDATE: This PR now includes support for evaluating functions as values over the previous syntax for updating multivalues properties. Currently only array functions are implemented.

UPDATE: This PR now uses functions exclusively for setting and manipulating array valyes.

The functions are based on the Postgres array functions.

array_remove($fieldSpec, 'ValueToRemove') // remove by value
array_replace($fieldSpec, 'ValueToReplace', 'Replacement'); // replace by valyue
array_append($fieldSpec, 'ValueToAppend); // append value

Other functions:

array('asdf', 'asdf', 'asdf') // returns an array value (i.e. `SET a.tags = array('foo', 'bar')`)
array_set($fieldSpec, 3, 'foo') // set index 3 to "foo"

UPDATE [nt:unstructured] AS a SET a.tags = array_replace(a.tags, 'Trains', 'Rockets') WHERE a.tags = 'Trains'
UPDATE [nt:unstructured] AS a SET a.tags = array('Rockets', 'Dragons') WHERE a.tags = 'Trains'
UPDATE [nt:unstructured] SET tags = array_replace(tags, 'Planes', 'Rockets') WHERE tags = 'Planes'
UPDATE [nt:unstructured] AS a SET a.tags = array_remove(a.tags, 'Planes') WHERE a.tags = 'Planes'
UPDATE [nt:unstructured] AS a SET a.tags = array_set(a.tags, 0, NULL) WHERE a.tags = 'Planes'
UPDATE [nt:unstructured] AS a SET a.tags = array_append(a.tags, 'Kite') WHERE a.tags = 'Planes'
UPDATE [nt:unstructured] AS a SET a.tags = array_set(a.tags, 1, 'Kite'), a.tags = array_set(a.tags, 2, 'foobar') WHERE a.tags = 'Planes'

See also documentation: phpcr/phpcr-docs#17

[dantleech]

Depends on https://github.com/phpcr/phpcr-utils/pull/121/files
and is related to: #81

[dantleech]

//cc @dbu

[dbu]

nice one!

how exactly do the first examples work? do they replace Trains, or all tags? (i would expect to replace all). and setting an index to NULL removes the value from the list, right? i think you can't store null as a value. and a.tags = NULL would remove the whole list?
what is the difference between test 1 and 3? you search for Trains and then Planes, but set the same thing.

[dbu]

i was wondering if using an absolute index as you do is valid. from reading i think it is:
http://www.day.com/specs/jcr/2.0/10_Writing.html#10.4.2.5%20Multi-value%20Properties%20and%20Null

however, i can't really think of a good use case for that. what is much more likely is that i want to replace the value i search for with something new. would there be a way to do either a complete string replacement or even a regex pattern replace that would apply to the specific value in the list? (i am thinking of things like: update all Plane tags to Airplane to standardize tags)

[dantleech]

• The first replaces the index containing Planes with Rockets
• The second replaces the entire value (as we specify an array in SET)
• One andf three are indeed the same :P
• Setting to NULL will remove the value at the matching index.

I think the use case for the index thing would be very specific but the use case for the [] syntax is generally useful, and given that, parsing an index is easy.

If you specify SET foo = "bar" WHERE foo = "baz" then this will replace the single value matching baz in the value of the multivalue property.

[dantleech]

Same as Replace a multivalie index by value

[dbu]

Ah cool, sounds useful. What happens if i do "where a.tags = 'a' and a.tags = 'b'"? Should we not allow the replace syntax in that case? Or come up with something like regexp do for match groups?

[dantleech]

Hmm, actually now you say that I am unsure about this syntax for replacing values, we are mixing the two sides of the query.

Maybe something like:

// @ Planes
UPDATE [nt:foo] SET a.tags[@"Planes"] = 'Bikes' WHERE a.tags = 'Planes';

// quoted value means select index by value
UPDATE [nt:foo] SET a.tags["Planes"] = 'Bikes' WHERE a.tags = 'Planes';

// two squre brackets means select index by value
UPDATE [nt:foo] SET a.tags[["Planes"]] = 'Bikes' WHERE a.tags = 'Planes';

This is pushing the syntax a bit of course -- are there any precedents for this type of thing in other languages?

[dbu]

i am -1 on the second option, because it looks like "Planes" would be the index in a hashmap, while it is actually a value. the [[ look a bit weird. i think the @ best shows the meaning. maybe perl already has something like this, being the language of cryptic shortcuts?

[dantleech]

See discussion: #81

[dantleech]

Updated to use functions.

[dantleech]

Note it would be good to also implement functions like CONCAT, JOIN, LOWERCASE, etc. In fact we could implement all of the MySQL / Postgres functions.

The problem is that while we could implement them the SELECT statement SELECT LOWERCASE(a.title) it would be really pushing it to implement this on the criteria WHERE LOWERCASE(a.title) = 'asdf'.

Using functions in SELECT could also be a good strategy for retrieving the node path and score, e.g. SELECT PATH(a), SCORE(a) FROM [fooar] AS a.

[dbu]

needed or not?

[dantleech]

technically yes. but I can't find a way to make the assertion work without refactoring the code .. maybe that wouldn't be such a bad idea.

[dantleech]

I removed this test in the end. This is one of the things I don't like about PHPSpec, its not that flexible.

[dbu]

do you have open questions or is this ready? functions for select statements are in conflict with staying compatible with jcr, and they would not work with jackalope-jackrabbit as we need to run the search on the server.

[dantleech]

I think it needs a cleanup :) Functions would never be used with the SELECT, only with UPDATE.

[dantleech]

Updated. Will merge it later.

[dantleech]

Hmm. Updated the description of this PR, I have now removed the [] operators and all array operations are now effected as functions.

[dantleech]

I wonder if array_set is a good name, i.e.

SET a.tags = array_set(a.tags, 'foo', 'bar');

Maybe:

array_replace_index(a.tags, 0, 'foo')

Would make more sense.

[dbu]

you could also use array_replace in both cases, and if its an int its the index, if its a string its the value to replace. hm, sounds really confusing though when you accidentally pass a number as string...

[dantleech]

yeah I think we should keep setting indexes to values and replacing values with values should remain as separate functions.

[dantleech]

using array() does work..

[dantleech]

ping @dbu - what do you think? arrays are now set with a function (foo = array('val1', 'val2')), and I am not sure about the naming of the function indicated above..

[dantleech]

Merged. I think the function name and syntax are OK.

[dbu]

would this be overwriting or insert at that position?

[dantleech]

overwriting

[dbu]

sorry, only found the time to review now. i would suggest these names:

• array_insert_at (numeric index, any location. what if location > length?)
• array_append (add to end. i guess rather than prepend, one would use insert_at(0) )
• array_remove_at (remove by index)
• array_replace (replace all occurences of value)
• array_remove (remove all occurences of value)
  you could now also do array_insert_after|before if you want. its less useful though.

[dantleech]

I was probably a bit to hasty in merging as the tests are now failing, will have another look, insert_at would be a new operation right? I don't think I have covered that, only replacing indexes.

[dbu]

indeed, so i would call it array_replace_at instead of array_set to make it more explicit.