Fix usage of live in opcodes and no longer preserve registers by default

[pguyot]

Many opcodes have a live parameter which is the number of registers that should
be preserved by GC operations. Take this into account and conversely stop
preserving registers by default.

This affects significantly the safety of memory_ensure_free when writing nifs.
Indeed, arguments (which come from registers) are no longer preserved.
However, no platform-specific nif was found that needed to be fixed. If
required, calls to memory_ensure_free or memory_ensure_free_opt should be
replaced with calls to memory_ensure_free_with_roots.

These changes are made under both the "Apache 2.0" and the "GNU Lesser General
Public License 2.1 or later" license terms (dual license).

SPDX-License-Identifier: Apache-2.0 OR LGPL-2.1-or-later

[fadushin]

This affects significantly the safety of memory_ensure_free when writing nifs. Indeed, arguments (which come from registers) are no longer preserved. However, no platform-specific nif was found that needed to be fixed. If required, calls to memory_ensure_free or memory_ensure_free_opt should be replaced with calls to memory_ensure_free_with_roots.

We probably have a host of “downstream” nifs in the atomvm project that may need attention as the result of this change. Do-able, but it may have an impact on other projects.

Ones I can think of:

• atomvm_adc
• atomvm_ssd1306
• atomvm_neopixel
• atomvm_dht

others?

[pguyot]

As mentioned, no nif in AtomVM needed to be updated.

I did check the nifs you mention and they do not need an update either:

• atomvm_adc always calls memory_ensure_free to return an error tuple and these calls are not followed by accesses to non immediate terms that would be invalidated
• atomvm_ssd1306 doesn't call memory_ensure_free at all
• atomvm_neopixel is like atomvm_adc
• atomvm_dht does call memory_ensure_free for something else than an error tuple, but doesn't access any non-immediate term after this allocation.