Re: disruptors of & inventory of Ada compilers and latest their era of ISO8652 compliance

["Dan'l Miller" <optikos@verizon.net>]

On Thursday, May 17, 2018 at 10:56:02 AM UTC-5, Luke A. Guest wrote:
> Dan'l Miller <@verizon.net> wrote:
> Darwin, the BSDs, and Linux.  I wonder why the design of an LLVM IR backend
> for GCC wasn't merely just another machine-code downstream from GIMPLE,
> e.g., as (hardware-processor machine-code ISAs are) overviewed at
> https://kristerw.blogspot.com/2017/08/writing-gcc-backend_4.html
> 
> See previous discussions about taking gcc’s ir and the ramifications of
> doing so, it makes the resultant code gplv3. It’s in the gcc faq and
> licence.

1) DragonEgg is licensed as GPLv2 (as GPLv2 was the version at the time of its inception), so there is clearly no goal there in DragonEgg of making a nonGPL LLVM backend for GCC.
2) As implied by all hardware-processor machine-code ISAs in GCC, an LLVM IR as just analogous to a machine-code-ISA backend on GCC would of course be GPLv3 nowadays, just as any hardware-processor machine-code ISA would be in GCC.

> > On this note, does anyone know how much GNAT's frontend
> 
> GNAT IR -> generic -> gimple, but I think they got rid of the generic step
> but I’m not sure.

Wait, GNAT has its own full-fledged IR stage prior to deriving either GENERIC or GIMPLE from that GNAT IR?  By IR here, I of course mean capturing •all• the semantic behavior of the program as per the _ARM_, not merely an AST of the recognized syntax.  Perhaps you are speaking of a blend:  that GNAT might encode 100% of the semantics adorned on the AST itself prior to deriving either GENERIC or GIMPLE •mechanically• without adding any more _ARM_-defined Adaness.  If such a GNAT IR exists that would capture all Ada-ness prior to •mechanically• deriving either GENERIC or GIMPLE (i.e., mechanically = with no knowledge of the _ARM_), then that might be the better place to shunt off to LLVM bitcode IR*, skipping both GENERIC and GIMPLE.

* a GPLv3ed LLVM-backended GNAT compiler, of course, analogous to the old llvm-gcc and llvm-g++ in the early days of LLVM

Conversely, if GNAT is still imbuing fresh _ARM_-defined Adaness into either GENERIC or GIMPLE that wasn't fully represented upstream prior to GENERIC/GIMPLE, then these imbuings would be skipped, nullifying any fleeting interest in a shunt off to LLVM bitcode IR prior to GENERIC/GIMPLE.

None of this arrives at a non-GPLed open-source Ada compiler though.