Writing Linux Modules in Ada (2016)(nihamkin.com) |
Writing Linux Modules in Ada (2016)(nihamkin.com) |
Later I learnt that VHDL is related to ADA. I've been itching to try it for a long while. Might give it a try this week. Currently I am in my Rust phase and loving it (thanks cargo). My Haskell fever is gone though.
"SQL/PSM is derived, seemingly directly, from Oracle's PL/SQL. Oracle developed PL/SQL and released it in 1991, basing the language on the US Department of Defense's Ada programming language."
I'd love to get a gig in ada.
I think that an ada-like language could make a real resurgence in embedded programming. Ada gets all the things about bare-metal right that C got wrong. However, it's held up by legacy tooling, clunky syntax, and obtuse compiler errors. Adacore has gone a long way towards alleviating those issues over the last few years, with alire and the ada_language_server. Time will see where this language takes us.
Is there any chance you'd expand on that? I'm curious to know your thoughts.
I'm curious, never had a look at Ada, can you elaborate?
This included SMTP, IMAP, POP3 daemons, and various other tools. I was going to write them all in Ada. I had a basic SMTP daemon that would accept mail and deliver it. But then, one of the various Outlook worms generated a ton of revenue at the expense of sleep, hygiene, and dignity, and I decided to get out of the Email business ASAP.
Still wish I would have kept working with Ada. I really liked it, and one could write tight code with it.
I'm going to be giving it another go here in a bit, I got a small little personal project where I'm going to make a serious attempt to curse at it for a bit.
With be the popularity of Rust, it makes me kind of wonder why Ada isn't more popular. I should give the language a go.
Adacore in the last few years have been investing heavily in modernizing the tool chain, but now it seems they are also investing in Rust.
Adacore is working with Ferrous Systems on Ferrocene (https://ferrous-systems.com/ferrocene/), a Rust toolchain for use on safety critical applications:
https://blog.adacore.com/announcing-publication-of-the-draft...
https://ferrous-systems.com/blog/ferrous-systems-adacore-joi...
I am hoping that Ferrocene's work will help the drive the standardization of Rust over the next few years:
It looks like Ada is having a resurgence due to Rust. I see Ada being pushed alongside Rust in a lot of places that emphasize safety.
https://learn.adacore.com/courses/intro-to-embedded-sys-prog...
Here the bit fiddling in Ada data structures is explained https://learn.adacore.com/courses/intro-to-embedded-sys-prog... Quite cool
One thing I still haven't wrapped my head around is how "dynamic" memory allocation and cleanup works in Ada. It doesn't seem as important to mention that early in any documentation anywhere. And, maybe it's the C/C++ programmer in me, but that strikes me as a bit odd. Or, perhaps I just can't see past the tip of my nose and it's there.
I kind of need to know how dynamic memory works in any programming language before I plan to invest deeply in learning it. And it needs to work on my hardware.
Ada’s dynamic memory principles are definitely unique. For heap allocation its based around memory pools, at least in GNAT. For the most part it’s RTTI but you can do manual new/free style too (though discouraged).
Ada uses a secondary stack as well for variable-length function returns, so in practice you don’t need to do heap allocation very much.
There are also equivalents of some STL containers like vector that can handle heap allocations for you safely.
I think there's a good growing consciousness of the fairly terrifying unsafety of C/C++, and the relative success of Rust is some evidence of that, at least.
Many moons ago I bought an Ada 95 manual, and learned a bit of the language with intent to fiddle with it but never finished. I like the idea but not sure I'd be wanting to give up various... modern conveniences... I get from Rust in order to work in that world.
I am trying to write show control software in SPARK2014 at the moment. Show control are critical since it is used to power lifts and stage machinery as well as performer flying systems where safety and high-integrity software is critical. I like Rust, but I feel it is not quite there yet especially in terms of the number of real-world systems in this niche. I also find SPARK2014 easier to write and read. I have been programming since 1978, and although I gravitate towards terse, functional languages like Haskell, APL/BQN/J, I experience a lot of friction whenever I dive back into Rust. SPARK2014 is very verbose and Pascal-like, but this is tedium vs. confusion or confidence in what I am writing. I know AdaCore is working with Ferrous Systems to bring Rust more up to the features of Ada/SPARK2014, but for now I needed to make a pragmatic choice based on real-world usage and ease of use and understanding.
And, speculating here, with the encroachment of AI into programming/software engineering, I assume that it's convenient to use languages that are declarative (e.g. Haskell) and/or designed for verification/formal methods (e.g. Ada/SPARK) to integrate AIs of various kinds.
Interfacing with C APIs / libraries is really easy and portable across Ada implementations:
https://learn.adacore.com/courses/intro-to-ada/chapters/inte...
http://www.ada-auth.org/standards/22rm/html/RM-B-3.html
The quality of the optimizer depends on the Ada implementation.
GNAT, the free software Ada implementation, uses the GCC backend so it is pretty good:
GNAT also has LLVM Backend: https://github.com/AdaCore/gnat-llvm It's stable and Adacore plans to ship it in GNAT Pro 24, i.e. next release. Note that it's the same front-end, so you get best of both worlds basically.
It was a multi pass, 5-10 stage process (or more. I want to say 13 but time plays tricks) Very costly language to compile, in those days. (Vax 11/780 running Unix 32V, a precursor to BSD and Ultrix by some years)
The story was it emitted an error/warning code along the lines of "Congratulations you have used the most abstruse feature of the ADA language" -which the approval people made them take out before it got certified.
Wirth had a sabbatical residency in York around the time of the Ada language selection process, his choices didn't make it through the strawman/steelman process, I think they resurfaced in Modula-II. It was a pascal teaching department like many others in the UK of the time, so it made sense for him to spend time there. Modula-II is said to be a systems programming language too.
Ada was very hard to teach. The ideas of asynchronous, and exception handling didn't sit very well on young minds. Maybe now they're well enough understood to teach in Rust. At the time, the absence of a rationale around "why" was very strong. York had a miniature 2-lift engine model which it used as a proving ground for Ada programs and undergraduate projects. Lift sequencing is a bit of a black art in itself but if you put that optimality of "which lift, which direction, which floor" to one side, the mixture of real-time controls and sensors were probably a good fit. (lift == elevator for the other side of the Atlantic)
I remember some concern in the department the only logical endpoint for Ada was to code military flight control/weapons/radar systems, and people felt uncomfortable about the implicit participation in the UK War economy. This was during the time of the Greenham common protests against US nuclear forces on UK soil.
During the Alvey 5th Generation funding debacle ("Catch up with Japan at all costs") there was another round of this using GEC400 computers, again very directly related to Uk MOD needs for weapons control systems and what I think became the Nimrod airborne radar. Probably signals processing is a very good fit for Ada. (I didn't work on that project, or the compiler)
People said that the consistency of mapping data structures to devices, chip signal lines, real things, and the abstractions around that in types worked well in Ada. I found it horrendously complicated to understand. People might say C is a hack but the literal directness of C structs on a PDP11 or Vax to the underlying architecture worked pretty well to me. I guess the problem is that C was always too close to Assembler for some people. Bliss/32 was the systems programming language of choice in Digital, and I think continued to be used to write VMS, although I read now it was almost entirely written in DEC Macro assembler.
It’s safer than C, but I’m not quite sure where recent specs line up against C++.
People keep repeating this nonsense without updating themselves beforehand.
EDIT: To simply education on Ada,
Yes there was an optional GC, no one ever implemented it, so in Ada2012 got removed from the standard.
Almost everything can be allocated on the stack, so a strategy is to catch exceptions of not enough stack space and retry the same function with a smaller size for the data structure.
Ada95 introduced controlled types, which is basically Ada's version of RAII, no need to call Unchecked_Deallocation outside implementation details. Hardly any different from Rust code that uses unsafe underneath.
Ada/SPARK, now part of regular Ada specification, provides theorem proving capabilities, and contracts, allowing another safety level still not available in Rust.
Additionally Ada Core is contributing improving lifetime rules for access types, to have a kind of borrow checker light, when needed.
Finally, there are still 7 Ada vendors in business, with 40 years of experience deploying Ada into safety critical scenarios.
Edit: there's also reference counting and controlled types of course. And the secondary stack makes many uses of heap allocation go away.
I went down a bit of a rabbit hole recently looking to see if there was a LLVM way to do this. It looked like it was being worked, but I'm not sure it's the best way to get started with Mac OS and Ada on Apple Silicon.
I'll check out homebrew and macports too... again, just in case!
Thanks again!
Text : Constant String := Read_Chapter( Book );
Additionally, nesting DECLARE blocks and subprograms allows a fairly fine-tuned memory-usage/cleanup using the stack. The above example could, for example, be part of an outer DECLARE block, which has an inner DECLARE, perhaps with "Paragraphs : Constant String_Vector := Get_Paragraphs( Text );" in its declarative region and "For Paragraph of Paragraphs loop" in its body... as soon as the block is exited, the stack is popped, reclaiming the used memory. This, in turn, means that the need for heap allocation is greatly reduced.Here's an excellent presentation on Ada's memory management: https://archive.fosdem.org/2016/schedule/event/ada_memory/
It’s interesting to hear that they had a longer history of teaching Ada with real-time sensors and controls.
There was attempt to have it in Ada 2022 Standard, but it came late. Committee decided it's best not to rush and let compiler vendors implement how they think is best and go from there.
>It’s safer than C, but I’m not quite sure where recent specs line up against C++.
Ada allows to return objects of variable size, so it's not that common that you actually need to explicitly allocate stuff. The general guideline for dynamic memory allocation is:
1) Use Second stack (this is how Ada allows to return object of variable size) 2) If cant use containers 3) If cant use controlled types (RAII) 4) Only then use New.
https://github.com/simonjwright/building-gcc-macos-arm-eabi
Bonus point is that you get the compiler for ARM bare metal
begin
return "Forty Two";
end Get_Answer;Yes indeed, heap allocated pointers are a PITA but we don't use so many of them. In 16 years as an kinda embedded dev on multiple MLOC codebases, I can count on my fingers the times I had to reach to pointers (and unchecked_deallocation) and didn't have a safer alternative (usually Controlled scoped pointer types, but also containers, indefinite types, and if really stuck reference-counted things if really you must murky the scope - but I'd flag the last one as 'need to show why you can't just copy the data around' at code review...).
Not saying it's not missing but I wish we spent more energy on automated proof of absence of runtime errors, large-scale whole-system static analysis, and even more stringent runtime or compilation checks. And better tooling/language support around network, threading, and distributed processing, improved test and coverage tools.
The secondary stack would better be used for 'functions returning objects, of which you don't know the size, at call time'.
function Any_Number (A : Natural) return String is
begin
if A = 42 return "A = Quarante Deux";
end if;
return "A =" & A'Image;
end Any_Number;
Or for those allergic to begin/end function Any_Number (A : Natural) return String is (if A = 42 then "A = Quarante Deux" else "A =" & A'Image);https://learn.adacore.com/courses/advanced-ada/parts/data_ty...
http://www.ada-auth.org/standards/22rm/html/RM-13-1.html
Wouldn't it be nice in C to be able to define how a struct is laid out in the machine representation? In Ada, you can and it is part of the standard, so it is portable:
https://learn.adacore.com/courses/advanced-ada/parts/data_ty...
// Ada
for Day use (Mon => 2#00000001#,
Tue => 2#00000010#,
Wed => 2#00000100#,
Thu => 2#00001000#,
Fri => 2#00010000#,
Sat => 2#00100000#,
Sun => 2#01000000#);
// C++
enum Day {
Mon = 0b00000001,
Tue = 0b00000010,
Wed = 0b00000100,
Thu = 0b00001000,
Fri = 0b00010000,
Sat = 0b00100000,
Sun = 0b01000000,
};
As for the record representation, my understanding is that it is equivalent to having a normalized __attribute__((__packed__)), where smart compiler padding is disabled and you can arbitrarily decide the memory layout of your struct? procedure Enable_USB_Clock is
begin
Registers.PMC_Periph.PMC_SCER.USBCLK := 1;
end Enable_USB_Clock;
Example taken from here: https://learn.adacore.com/courses/Ada_For_The_Embedded_C_Dev...General lack of dumb C stuff like switch fallthrough, null terminated strings (Arrays in Ada are passed with fat pointers), undefined behavior, no need for memcpy, no preprocessor bullshit etc.
Ada is more like C++ in functionality so it has Generics, Tasks (Threads), Exceptions, Packages, Strong types, Design by contract etc (All much much saner than C++ stuff)
Despite all of the features it's very embedded friendly. Language allows you to disable features you dont wont with Restrictions Pragma (Very long list of restrictions you can apply: https://docs.adacore.com/gnat_rm-docs/html/gnat_rm/gnat_rm/s...). You can also fairly easily change runtimes: https://docs.adacore.com/gnat_ugx-docs/html/gnat_ugx/gnat_ug....
> Ada is better than C
> lack of dumb C stuff
Back to your comment, strong types and generics look super nice for embedded. Not sure I would like fat pointers, exceptions and threads in my embedded code though.
Sorry about that, English is not my first language so i might sound rude sometimes. I actually dont hate C, but if you used it you know its flaws. I think stuff i mentioned about C is objectively bad, hence why i called them dumb stuff.
>Not sure I would like fat pointers, exceptions and threads in my embedded code though
Fat pointers are just pointer + size of an object. You have to pass array size anyway, so its just convenient. Obviously, if you need to just pass a pointer there are ways to do that.
As for exceptions and threads: As i said, you can disable or modify them just by using Restrictions pragma. Its a language defined thing. For example, Ada 2022 defines two profiles for safety-critical hard real-time computing:
https://en.wikipedia.org/wiki/Ravenscar_profile http://www.ada-auth.org/standards/22rm/html/rm-d-13.html
This is basically like C++ destructors, but the problem is there are no move semantics in Ada, so you can't implement something like unique_ptr.
It's hardly comparable with Rust.
It isn't on par, Ada has no lifetime management whatsoever. It doesn't even provide C++ style smart pointers out of the box. It is possible to implement something like shared_ptr, AdaCore even has a tutorial[1], but as the other commenter pointed out, the language doesn't provide the primitives necessary for unique_ptr.
SPARK does have something like this. In fact, in SPARK, every pointer assignment transfers the ownership. But SPARK and Ada aren't synonymous. SPARK is a formal verifier built on top of Ada. Like most such tools, it's very constraining and time-consuming. It's not something that every (or even most) Ada projects use.
Nevertheless, Ada is a perfectly good language, and it's probably safer than C++. It has some really cool features. I'm really fond of in/inout/out references (cppfront stole this), named function parameters, secondary stack, fine-grained control over records (struct) layout, etc.
However, I'm not so fond of the extreme verbosity. It isn't just because of Algol-like keywords, Pascal is less verbose than Ada. Even basic stuff like instantiating generics is very noisy:
procedure do_something is
package Integer_Vectors is new
Ada.Containers.Vectors
(Index_Type => Natural,
Element_Type => Integer);
vec : Integer_Vectors.Vector;
begin
return;
end;
In C++ that would be: void
do_something(int arg) {
std::vector<int> vec;
}
[1] - https://www.adacore.com/gems/gem-97-reference-counting-in-ad...No offense taken, not a native here either ;-)
> if you used it you know its flaws
Indeed, I've been using C for almost 20 years now. I won't say it's without flaw for sure, it has its quirks, but overall I do think it's quite okay for the job.
> Fat pointers are just pointer + size of an object
Yeah I know what fat pointers are, I even resort to handcraft some form in C for some neat performance hackery on x64.
But the thing is, we're talking embedded here. Most ucontrollers I use have 8b address space, no MMU or any form of virtual memory, separated instruction/data bus (MVHA). That kind of thing don't play well with funky fat pointers.
Sure if your definition of embedded is "64b ARM" all is good, but I guess we're on a spectrum.
Exceptions are pretty much non existing as well, since that would require some form of runtime, which you often just cannot afford on a small chip (if not just form the sheer size of it).
Threading is a no go as well. To get threads, or any form of multitasking really, you have to rely on an operating system, which by definition is a bit weird to have on an embedded IC.
You can definitely use Ada on 8bit microcontrollers, for example on 8bit AVR: https://docs.adacore.com/gnat_ugx-docs/html/gnat_ugx/gnat_ug...
You can make runtime as small as you want.
As for exceptions and threads: I mentioned them because thats what comes to mind first, there are more benefits obviously.
We (embedded company) don't hire specifically with Ada experience (it's welcome but not mandatory) we just train people and then focus on shipping stuff.
I’ve heard this exact argument about VHDL versus Verilog, with the former being explicitly based on Ada’s syntax and the latter being explicitly based on C’s. (Turns out though that VHDL is also strictly better than at least traditional Verilog, as it requires separate interface specifications which lead to improved modularity.)
If anyone were to actually try to create “Ada: The Next Generation” I’d encourage you to just go all the way to S-expressions. Focus on the completeness and correctness of _the system_ and stop worrying about superficial complaints—or brush them off with a suggestion that they can use any syntax they want and just translate it to the standard one via tree-walking.
I think of it a bit like rat's nest wiring vs. nice neat labelled cable looms.
I'll use this chance to say that I'm a fan of Ada's declarative blocks, which you could say are part of its syntax. There's a lot to like about Ada, and I'd encourage anyone interested in bare-metal programming to give it a try. Even if you don't intend to use it long-term, there's a lot of good language design ideas in Ada that can be learned from.
To address a sibling comment alleging that I'd rather it look like C, that's not necessarily true. I know this is much more controversial, but I'm actually more of a fan of Python's syntax.
I always felt that repeating myself to the compiler (down to `procedure subprogram is begin`…`end subprogram`) or otherwise being verbose was something of a feature in the vein of defense in depth than the syntax not aging well. It never seemed to me like it was meant to impose clarity on the code, just that you were meant to file your subprograms and types in triplicate. It's useful as a backstop against trying to do things quickly instead of doing them deliberately.
At least, Ada's use of verbosity stands in stark contrast to COBOL, which does have the explicit aim to make things clear.
I'm mostly of the same mind as you in re Python's (and Haskell's and F#'s) whitespace-oriented ways of doing things, at least until the real world creeps in. I've had problems with early block termination because of mixed tabs and spaces that weren't readily apparent, and the language's toolchain wasn't particularly helpful in diagnosing them, leading me to lean on external tools. (COBOL at least has the heritage of being column-oriented, so indentation being significant is less problematic there.)