hex_compact() (ISEA Z7 backend) emits duplicate cell IDs for mixed-resolution/overlapping input

[gcol33]

hex_compact() (ISEA Z7 backend) emits duplicate cell IDs for mixed-resolution/overlapping input

Defect

R/hex_compact.R:53-78. The compaction loop computes is_child_of_full <- compactable & parents %in% full_parents to remove children once their parent is reconstructed, but if the parent cell itself is already present in ids (alongside all 7 of its children), the parent is never flagged as "a child of a full parent" (its own parents value differs from itself), so it survives untouched while the newly-synthesized parent is also appended.

Reproduction

hex_compact(c("010", "0100","0101","0102","0103","0104","0105","0106"), g)  # aperture 7
# c("010", "010")  -- a duplicate

Impact

The H3 backend delegates to the vendored compactCells and doesn't have this problem — it's specific to the pure-R ISEA path. Downstream code that assumes hex_compact() output has unique cell IDs (any per-cell join, unique()-free indexing, etc.) will silently double-count that cell.

Fix

Before appending newly-synthesized parents, also drop any input cells that are themselves already a "full parent" being reconstructed (i.e., dedupe the final output against cells already present verbatim in ids), or simply unique() the final result and treat that as a signal to review the loop's edge case rather than papering over it.

[gcol33]

Fixed in commit 17011c7. The compaction loop now also drops an input cell that is itself one of the newly-synthesized full_parents (covers the case where the input already contains a parent alongside all 7 children), plus a follow-up cleanup pass for the adjacent case where a coarser ancestor and one of its own descendants both survive mixed-resolution input. Verified against the exact reproduction plus normal/incomplete/nested scenarios — no duplicates in any case.