"@" context-binding operator causes the following step to evaluate to undefined

[steve-redka]

Version: jsonata 2.2.2

Input:

 {"a": {"b": 1}}

JSONata:

/* correct, returns 1 */
a.b
/* broken, returns undefined */
a@$x.b

$x isn't even referenced in the second expression - merely attaching @$x to the a step causes the subsequent .b step to return undefined instead of navigating normally. Reproduces whether a is an object or an array, and whether the following step is a bare field name (.b) or wrapped in parens (.(b)).

Possibly related to #744 (array flattening changes when @ or # is used) - this looks like a more severe case of the same underlying issue, since here the value is lost entirely rather than reshaped.

[Turings98apprentice]

This behavior is intentional. The @ symbol attaches the current context (a in our case) to the variable ($x), then jumps back up to the parent context ($$). The reason a@$x.b returns undefined is because b wasn't found in the parent (global) scope ($$).

You can use the library example in the docs to get a better feel for how it's supposed to work.

Here's a link to the relevant section in the docs that explains the rationale: https://docs.jsonata.org/path-operators#-context-variable-binding

[steve-redka]

This line is ambiguous, then: "this context will be carried forward to be the context of the next stage in the path." Read on its own, that says the bound value (loans) keeps being the context for what follows. What actually happens is the reverse: the next stage evaluates against the context that existed before the bound step (library), which is what lets .books resolve as loans's sibling rather than its child. That's clear once you trace through the worked example, but the topic sentence describes the opposite mechanism, so a reader who tries a@$x.b reasonably expects .b to still see a's value.

I think this would be a better wording: "the context for the next step reverts to the context the bound step was itself evaluated against (not the bound value) which is what allows the next step to reach a sibling of the bound array."