Vladimir Davydov

> Should this iterator walk only over a continuous range of tuples? Or skip unsuitable tuples? > > E.g imagine data: > > ``` > 1. [1, 2] > 2....

> I guess someone's app might break after this change (even if it's correct ideologically). Hello @Yukigaru, AFAICS floating point numbers are decoded with loss of precision while integer numbers...

> The arguments I heard why we can't reuse `IPROTO_SYNC` In the old code version, it took me a while to realize that a relay field called `sync` and encoded...

> _"The syncs might be used by somebody"_. This isn't true, because they were never sent in the acks. This is how the code looks (before this patch) which sends...

Agreed to rework the CE part a little after the EE part review.

I fixed the `unit/key_def` test issue (amended the corresponding commit). It turns out on that Fedora 35 `memcmp` returned a different (albeit same-signed) result. diff ```diff diff --git a/test/unit/key_def.c b/test/unit/key_def.c...

I disabled Lua JIT in the test for aarch64 and filed a ticket https://github.com/tarantool/tarantool/issues/7739 to enable it back.

Reopened by #7643.

Tarantool Cartrdige maintains a [list](https://github.com/tarantool/cartridge/blob/d8d48a17e9bf4505941f7f5785fd4b6a9d809918/cartridge/hotreload.lua#L156-L170) of system fibers that shouldn't be cancelled during hot reload. If we marked these fibers as system, we could remove it.

Cancelling a system fiber may result in an undefined behavior (e.g. crash) - that's why we must prohibit it. Cancelling a user fiber is fine from Tarantool's point of view...