Allow negative (placeholder) IDs in calls to get_metadata

[westonruter]

See #38072 for full background. Nav menu items are created in the customizer with negative IDs to serve as placeholders. This allows such nav menu items to be referenced before writing to the database. This works for previewing nav_menu_item posts, but it turns out to not work for postmeta for such posts. The reason for this is get_metadata uses absint as opposed to intval and so the placeholder nature of the object ID is dropped. Along with this change the get_metadata function can short-circuit for such negative IDs after the get_{$meta_type}_metadata filters apply since no such objects would exist in in the database which uses UNSIGNED INTEGER for ID fields. As such, placeholder meta would only be supplied via filters.

[dd32]

Nav menu items are created in the customizer with negative IDs to serve as placeholders.

Along with this change the get_metadata function can short-circuit for such negative IDs after the get_{$meta_type}_metadata filters apply since no such objects would exist in in the database which uses UNSIGNED INTEGER for ID fields

Can I just say this feels like a massive hack that should never have been done in the first place, and probably shouldn't be extended to other parts of the codebase? I'd actually prefer anything that *does* allow something like this to be removed, or in the event of a negative ID being passed, return falseing immediately before filters.

If there was a need to inject "placeholder" items into the structures, they shouldn't be calling/running get_*($neg_id) but instead being supplied by some kind of a pre_* filter on the object getters..

[westonruter]

And #38072 is about removing the use of such negative IDs as placeholders.

Nevertheless, the get_{$meta_type}_metadata filter is actually like pre filters elsewhere, as it short-circuits the filter returns a value. It is similar to pre_option_{$option} in this regard.

[peterwilsoncc]

Related #37746. 37746.diff:ticket:37746​ replaces absint() with intval() and returns false for negative IDs.

I agree with @dd32 that this behaviour ought not be extended.

[jorbin]

I'm with @peterwilsoncc and @dd32 on this.