This is the correct answer. If you are really doing what the article says (allocating all memory at compile time), why is malloc() being linked in at all?

What we do at day job is to have a 'whitelist' of all malloc callers which is checked by the CI.

Sometimes you want to allow dynamic allocation at application startup, but never thereafter. Free happens implicitly when the system reboots.

This lets you use dynamically sized buffers, depending on the particular configuration present at startup. Very handy to not need to build many different configs and maintain separate binary images (and keep track so that you never update hardware with the wrong image). The disadvantage is that you can't just remove malloc() in such a design.