[keldaris]

Luckily, little of it matters if you simply write C for your actual target platforms, whatever they may be. C thankfully discourages the very notion of "general purpose" code, so unless you're writing a compiler, I've never really understood why some C programmers actually care about the standard as such.

In reality, if you're writing C in 2025, you have a finite set of specific target platforms and a finite set of compilers you care about. Those are what matter. Whether my code is robust with respect to some 80s hardware that did weird things with integers, I have no idea and really couldn't care less.

[msla]

> I've never really understood why some C programmers actually care about the standard as such.

Because I want the next version of the compiler to agree with me about what my code means.

The standard is an agreement: If you write code which conforms to it, the compiler will agree with you about what it means and not, say, optimize your important conditionals away because some "Can't Happen" optimization was triggered and the "dead" code got removed. This gets rather important as compilers get better about optimization.

[uecker]

True, we are currently eliminating a lot of UB from the future C standard to avoid compilers breaking more code.

Still, while I acknowledge that this is a real issue, in practice I find my C code from 30 years ago still working.

It is also a bit the fault of users. Why favor so many user the most aggressive optimizing compilers? Every user filing bugs or complaining about aggressive optimizing breaking code in the bug tracker, very user asking for better warnings, would help us a lot pushing back on this. But if users prefer compiler A over compiler B when you a 1% improvement in some irrelevant benchmark, it is difficult to argue that this is not exactly what they want.

[bobmcnamara]

Sadly, at least in the embedded space some of us still deal with platforms where the proprietary core vendor's compiler routinely beats open source compiler cycle counts by a factor of 1.5 to 3.

The big weak region seems to be in-order machines with smaller numbers of general purpose registers.

GCC at least seems to do its basic block planning entirely before register allocation with no feedback between phases.

[keldaris]

In practice, you're going to test the next version of the compiler anyway if you want to be sure your code actually works. Agreements or not, compilers have bugs on a regular basis. From the point of view of a programmer, it doesn't matter if your code broke because you missed some fine point in the standard or because the compiler got it wrong, either way you're going to want to fix it or work around it.

In my experience, if you don't try to be excessively clever and just write straightforward C code, these issues almost never arise. Instead of wasting my time on the standard, I'd rather spend it validating the compilers I support and making sure my code works in the real world, not the one inhabited by the abstract machine of ISO C.

[lelanthran]

> In practice, you're going to test the next version of the compiler anyway

> In my experience, if you don't try to be excessively clever and just write straightforward C code, these issues almost never arise.

I think these two sentiments are what gets missed by many programmers who didn't actually spend the last 25+ years writing software in plain C.

I lose count of the number of times I see in comments (both here and elsewhere) how it should be almost criminal to write anything life-critical in C because it is guaranteed to fail.

The reality is that, for decades now, life-critical software has been written in C - millions and millions of lines of code controlling millions and millions of devices that are sitting in millions and millions of machines that kill people in many failure modes.

The software defect rate resulting in deaths is so low that when it happens it makes the news (See Toyota's unintended acceleration lawsuit).

That's because, regardless of what the programmers think their code does, or what a compiler upgrade does to it, such code undergoes rigorous testing and, IME, is often written to be as straightforward as possible in the large majority of cases (mostly because the direct access to the hardware makes reasoning about the software a little easier).