It is only good for CLI applications and daemons.
EDIT: For the downvoters what are the POSIX APIs for:
- Graphics (besides the clunky CDE and OpenGL isn't POSIX)
- Thread/Task pools
- Bluetooth, NFC, GPS, fingerprint readers, ....
- Application containers
Currently it mostly covers hardware from PDP-11 systems still, anything else via /dev, ioctl, whatever, is vendor specific.
Last time I checked that weren't any portable API for USB for example.
So what was left out of ANSI C, is what became POSIX. Making any OS with ANSI C + POSIX compliant compatible with the original UNIX systems.
But unlike other programming languages, on C's case the evolution in terms of standard libraries stopped there.
I guess portable interfaces may just end up needing to be defined over non-portable interfaces, and C may end up not being practical for anything that needs to be portable.
Higher level languages aren't bad, and it would probably be a good thing if a language like Rust became the de-facto for low-level. My feeling is that C is still the simplest to compile though, which I think is an advantage.
But that is a talk for other day.
Aliasing is a problem, but I don't think it's strictly inherent to C.
Nowadays yes, you are forgetting C has 40 years of research into compiler optimization.
We could easily beat the register allocator in the 90's with Assembly coding, specially in x86.
There is a quote from a famous woman in compiler research from the mid-80's, where she claims C semantics have thrown the optimization research back to the dinosaur age.
Sadly I cannot find it, as I don't remember the exact words.
> Aliasing is a problem, but I don't think it's strictly inherent to C.
Most languages have stricter semantics for aliasing and do not decay arrays into pointers.
Also they tend to have better type information, which can help the optimizer.
As an easy example, multiprocessing has become more important between now and then. Why are things like processor affinities still not defined?
In the case of processor affinity, it looks to me like no-one has ever suggested it. Participating in the Austin Group is pretty open - why don't you put a proposal forward? (Maybe affinity can be added to struct sched_param somehow?)
Not really. I know someone who's spent ages trying to get O_CREATE added as an alias for O_CREAT.
http://spiegelmock.com/2015/01/23/fixing-a-typo-the-hard-way...