Re: Adapting an Ada compiler to generate 8051 code (Again?! ;-)

["Randy Brukardt" <randy@rrsoftware.com>]

"Niklas Holsti" <niklas.holsti@tidorum.invalid> wrote in message 
news:icm5tmF7ak3U1@mid.individual.net...
> On 2021-04-01 2:14, Randy Brukardt wrote:
>> "Niklas Holsti" <niklas.holsti@tidorum.invalid> wrote in message
>> news:icgdhjF3j1eU1@mid.individual.net...
>> ...
>>> I have often wished that there would be Ada compilers for more
>>> microcontrollers, but I understand why there aren't. An Ada-to-C 
>>> compiler
>>> seems the most promising route.
>>
>> Send $$$. ;-) This was a project that was ideally suited for the
>> Janus/Ada compiler suite, but we never were able to find a customer
>> for it. The problem is always that the first customer has to pay a
>> substantial part of the development; later customers don't have to
>> pay that freight. (Back in the "waiver" days we considered doing it
>> for the "fun" of making DoD-types have to find better excuses to
>> avoid Ada than a compiler not existing for it, but the likely ROI
>> wasn't there to convince the angel investors to go along with the
>> idea.)
>
> Yes, things like that were among the reasons I had in mind for the lack of 
> such Ada compilers. Plus the fact that the pain of writing a small C 
> program is much less than the pain of writing a large C program.
>
> If you think that targeting Janus/Ada to small microcontrollers is 
> practical, I have two questions for you:
>
> Can you describe, in general, the Janus/Ada internal program 
> representation that, say, an 8051-back-end would get as input?

It's called "JCode" (named long before Java existed, at least publically), 
and it is based on the ideas of p-code, tailored for what we needed for 8086 
and similar processor output. There are other inputs (mostly in the form of 
a partial JRL, which carries the unit dependencies and the like), but most 
of those are related to the binder.

Decent code generation for it is quite easy, since it is a rather low-level 
form. It's relatively easy to write a passable code generator and then 
incrementally improve it with pattern matching. One could probably build 
much fancier schemes on top of it; we thought about that periodically, but 
experiments showed that the differences wouldn't be worth the work. (Robert 
Dewar famously said that "most optimizations are disappointing", definitely 
a true fact from my experience.)

> Would you be open to external/community participation in creating new back 
> ends for Janus/Ada, either under NDA or working from a public definition 
> of the internal representation, and either work-for-free or work for some 
> share of future sales?

Yes, certainly. Contact me directly at the usual e-mail.

                          Randy.