Should FixedPMCArray autovivify nested arrays on set_*keyed()?

[whiteknight]

(From RT #46675)

Here is an example of code:

VTABLE void set_pmc_keyed(PMC *key, PMC *value) {
        const INTVAL k = VTABLE_get_integer(INTERP, key);
        PMC   *nextkey = key_next(INTERP, key);

        if (!nextkey) {
            SELF.set_pmc_keyed_int(k, value);
        }
        else {
            PMC *box = SELF.get_pmc_keyed_int(k);

            /* RT #46675: autovivify an Array and insert it in SELF */
            if (!box)
                box = pmc_new(INTERP, SELF.type());

            VTABLE_set_pmc_keyed(INTERP, box, nextkey, value);
        }
    }

It's clear from the code above that if the element at SELF[k] doesn't exist, that a new array is created, but that new array is never inserted into SELF. So the question is this: What is the expected autovivification behavior of FixedPMCArray (and, by extension, the other Fixed*Array types)?

[whiteknight]

patch to implement nested array autovivification on fixedpmcarray

[whiteknight]

I've attached a patch above that fixes the autovivification behavior to do what I think was intended originally. If people agree, we can apply that patch and add test for the autovivification behavior.

[jkeenan]

whiteknight,

Reviewing old tickets today, I came across this one and tried to apply your patch to trunk at r49555. It did not apply cleanly.

Are the issues you raised in this ticket still relevant? If so, could you please pull a new patch and re-submit?

Thank you very much.

kid51

[bacek]

Auto-vivification of nested aggregates is deprecated in #1561. Rejecting ticket.

-- Bacek