I was too fast. You fixed the rustdoc while I was reviewing 😉

I think this is incorrect. Maybe it was already incorrect before, but the heap pages should be the maximum.

This is indeed how it worked originally, I just changed the name of the field to actually say what it does.

We can consider changing it, and I agree with you that changing this would make sense (which is what I wrote in this comment), but I think that'd be better for another PR as this just restores the original behavior from before the #10480 landed.

So, if polkadot it setting these for parachains it is 1gb and nothing in the wasm file should allow to exceed this value. Currently the wasm file could specify whatever initial value it wants and it would get this plus the heap pages we are requesting by polkadot?

For normal runtimes this isn't such a big problem, as they are not provided by the user.

Yes, the blob can request whatever value it wants, and it will get that amount of heap pages plus the extra_heap_pages. But there's still an absolute limit set by the Config::max_memory_size. If it goes over that then the instantiation will fail, so in practice this is - while weird - not a problem. (For PVFs the limit is set here, so with the current limit a PVF can get at most 2080 pages allocated, which is ~130MB of memory.)