This code must remain "obviously correct" as much as possible. Systematically doing bound check before accessing an element is absolutely part of that process.

iRebase on top of D3618 and fix comments.

This makes me a bit worried, to be honest. Isn't it possible to move the nulldummy check near the bound check? This will avoid the spooky action at a distance effect (which is probably why this redundant check survived here in the first place).

Accepting, but this probably needs to be shuffled around based on change in the parent's diff.

These are tests we inherited from core. I don't see a good reason to reorder it all.

It's actually a good idea to build with -Wdelete-non-virtual-dtor if the compiler supports it.

How come this is changing images we inherited from core ? Second, is this lossy ? If it is, it should not be done on source material.

The comment in the interpreter should probably be added somewhere in sigencoding.cpp

If whatever is calling this dockerfile has these parameters, then wouldn't that be better to simply download the file and copy it within the docker image ?

Rebase on top of D3594

Why C style error handling is garbage illustrated. I'll do a proper review later on.

Remove unecessary header include

Remove unecessary include

I read the rationale and did not understand any of it. It seems like Doxyfile is generated by the build system now, but that doesn't explain why it has to be done on the machine hosting the doc. If deploying artifacts on host is something that is challenging, then it seems like this problem should be talked, because this is pretty much the ONLY thing that deployment does.