Re: Why couldn't an operating system be written in ada

[Optikos <optikos@verizon.net>]

On Saturday, April 6, 2019 at 9:55:21 PM UTC-4, Nick Roberts wrote:
> On 10/01/2019 23:38, cenci.br@gmail.com wrote:
> > Em sábado, 13 de julho de 1996 04:00:00 UTC-3, Mark  McKinney  escreveu:
> >> It has been claimed that the capability to interface with other languages
> >> is a great asset to ada. Sometimes interfacing can be a tremendous
> >> liability. Besides the OS could perform most of work that the language
> >> runtime does. So why not build an OS in ADA?
> > 
> > Is this discussion still active ?
> 
> I have a little bit of a plan for this, but I must admit it is very 
> ambitious. The biggest problem I face is getting help, as well as my own 
> lack of skill in running these projects.
> 
> My current plan goes along these lines:
> 
> I am building a new Ada compiler that will use a new technology for 
> targeting machine architectures. The following is a simplification, but 
> it gives the gist of it.
> 
> 1. The compiler builds Ada source text into an intermediary code, called 
> 'PXC' (a kind of abbreviation for 'extended pseudo-code'). These PXC 
> files are distributed to the target machines. Each PXC file is a module, 
> and a complete program comprises many such modules (perhaps hundreds of 
> them, or even thousands).
> 
> 2. Each target machine itself has a program called the Realizor, which 
> generates a real executable file, comprising actual machine code for the 
> architecture of its own processor(s), from the complete set of PXC 
> modules that go to make up the program.
> 
> One of the big advantages of doing it this way is that global 
> optimisations, including inlining of cross-module calls, can be 
> performed by the Realizor. In addition, the Realizor can implement a 
> strategy of capturing profiling data (execution statistics) for the 
> program, and periodically re-realising it to yield still better 
> optimisations.
> 
> 3. Because of the way the Realizor works, all privilege level-changing 
> calls (e.g. into the kernel) can be removed. Security is, nevertheless, 
> fully maintained.
> 
> (I also intend the technology to remove all use of hardware paging; this 
> will be replaced by a more intelligent software-based segment-swapping 
> mechanism.)
> 
> 4. Because of 3 above, it is possible for the kernel and all the device 
> drivers and other low-level software to be PXC modules, mixed in by the 
> Realizor along with all the other PXC modules. This means that many 
> kernel calls can be inlined, effectively eliminating all call overhead 
> on them.
> 
> I believe these technologies would enable a dedicated piece of software, 
> running 24/7, to run significantly faster.
> 
> When you consider how important performance is on today's vast farms of 
> servers, that could be a pretty compelling reason for people to switch 
> from traditional operating systems.
> 
> All I have to do is to do it :-)
> 
> 
> 
> I'm hoping it could be a way of getting Ada back into the mainstream. We 
> all know in this forum that building future software in Ada, especially 
> server software, would be a big step towards making it more secure and 
> reliable.
> 
> PS: There's a lot more about this scheme that I would like to say, but I 
> don't want to turn this post into a dissertation, and I do want to get 
> to bed soon. If you have any burning questions, please ask.

Why can't LLVM be your PXC?  LLVM already is your PXC and already built out your Realizors for a plethora of ISAs.