There was a similar rush of articles like this a few years ago.

tl;dr: C defined language semantics, and leaves some behavior undefined. Each system that C is ported to has the ability to define the behavior however it wants.

This blows the mind of PL folks every decade or so.

It’s cool that we have portable methods and formal language semantics for stuff like memory fences and atomics now, but that sort of thing worked fine in C back in 1970 (or else unix would not have worked). You just needed to read the target machine’s manual when porting stuff.

The modern version is arguably better, but also arguably worse. Does anyone else remember when the JVM got this stuff wrong, making safe multithreaded code impossible, and then later had to break compatibility with the language spec?

You could claim that we can’t trust hardware folks to get instruction semantics right (this is demonstrably true), but duplicating and slightly modifying the specs in your language spec doesn’t actually fix the underlying hardware bugs.

Yeah, getting old… I’ll go find a cloud to yell at.