I think it is worth mentioning that this was Robert Dewar's second or third Ada implementation (depending on how you want to count them). Robert was involved with the original Ada/Ed implementation from the late 1970s/early 1980s at NYU (targeting VAX minicomputers), and the re-implementation of that into a C-based implementation that ran on MS-DOS. Ada/Ed was the first validated Ada implementation, and it used SETL as its implementation language which Robert was actively involved in through the 1970s (so he had a long track record of language implementations).
What we are missing about the Ada story is the long line of languages and tools that inspired the Green design that lead to Ada, that is the language LIS (Pascal/Simula inspired) and its toolchain that Jean Ichbiah and associates used as inspiration for the Ada design.
For programming skill, I hold Dewar in similar regard to Arthur Whitney and djb. I still enjoy using Spitbol for writing lexers. MINIMAL, the portable assembly language used in Spitbol, is far more interesting than today's popular programming languages, IMHO.
Hrmph, MINIMAL is a sort of abstract assembly language that uses common features of a bunch of different machines, so you can target it to different architectures pretty easily. That kind of thing has been done many times, including in Fig Forth, GNU Lightning, Picolisp, and so on. Nothing wrong with it but it's not all that exciting. Spitbol itself on the other hand was remarkable. Its implementation is worth studying.
This reminds me of a recent podcast episode from Hidden Brain about the power of subtraction. Removing things or taking things away is a path to improvement that we rarely consider, to our own detriment.
I'd be interested to know how other type-safe language compilers like GHC (Glasgow Haskell Compiler) handle this problem. I believe that .hi files (object files that GHC generates from .hs Haskell source files, like C compilers generate .o's) have type info in them. But I don't know what GHC does to ensure global type consistency. Maybe it has its own equivalent of a link phase, that has to bring in all the types from the .hi's? Is that the equivalent of Ada's originally conceived program library? It never really occurred to me to look into this.
I hadn't heard of the example of rms' technical insight there, which is interesting. Not knowing enough about Ada, I wonder how the specification and implementation compare with Fortran 90 modules.
What we are missing about the Ada story is the long line of languages and tools that inspired the Green design that lead to Ada, that is the language LIS (Pascal/Simula inspired) and its toolchain that Jean Ichbiah and associates used as inspiration for the Ada design.