translate ${name} inside a repeat to ../relative/path/to/node

[MartijnR]

One day this will have to happen... so I'm putting this here. It relates to a long-standing javarosa bug: https://bitbucket.org/javarosa/javarosa/wiki/XFormDeviations#!model-item-property-xpaths-are-always-evaluated-in-the-context-of-the-nodeset. Seems like pyxforms could take the lead in starting to do this correctly! It will avoid the 'position-injection' hack that is currently required for evaluating absolute paths inside repeats, improve performance significantly for some forms, and reduce code ugliness.

type             | name           | constraint
----------------------------------------------------------
begin_repeat     | rep            |  
text             | A              |
text             | B              | ${A} = 'something'
end_repeat       |                |

${a} in this case should become "../A" instead of "/data/rep/A" because the absolute "/data/rep/A" in pure un-hacked XPath always returns the value of the A node in the first repeat (even if the context is the 100th repeat).

[nathanathan]

I'm wondering if there is ever a reason to generate absolute paths. Switching to relative paths like this would definitely simplify the code.

[MartijnR]

Glad to hear that! It would simplify enketo too as the workaround is messy and buggy.

There should be no reason to ever want absolute paths when referring to a node inside a repeat from another in the same repeat. I believe it's just something that happened long time ago in the early JavaRosa libs and was never corrected (so there are many forms out there that have these incorrect XPaths).

[nathanathan]

Now that I've thought about this a bit more, I've realized it's necessary to add some notion of scope to the pyxform so that the correct relative paths can be computed from ${variable}s to reference values nested at different levels. So, never-mind what I said about it simplifying things in pyxform, although I think variable scope might be nice to have.

[MartijnR]

Right, I can see that it gets more tricky with nested groups/repeats. I hope you'll come up with a way to do this, as it would be great to push this forward.