Instancing all 1605 glyphs of Handjet at its axis midpoints spends
about 625M units, exhausting the old glyf allowance and dropping the
outline tables. CFF2 instancing, tuple splitting and retained contour
points also have less headroom than intended.
Increase those allowances using the measured monster-font workloads.
Separate the layout-table allowance from glyf so layout subsetting
keeps its existing ceiling. The affected measured margins are 12x
or greater.
Keep an API case above the old glyf allowance. Exercise exhaustion on a
small existing Handjet fixture through the real glyf subset path with a
controlled budget: the full cost succeeds, one unit less fails. This
avoids the original enlarged test timing out under Valgrind. Enlarge
the adversarial CFF and retained-point cases to retain their coverage.
Tested with the full configured Meson suite. The original audit also
covered full and sparse subsets, static and partial instances, shifted
axis defaults, single-axis instances, and IUP optimization.
Assisted-by: Codex (OpenAI)