@Kartik Agaram (branching my thread from #CC2JRGVLK) Can you offer examples of "existing languages [being] highly constrained in their degrees of freedom"? I usually find human language alone inadequate to discuss these matters, so I need to see specific examples. But yeah, we need to see programming language design itself as a multi-layered thing; indeed I see my task as helping to architect an entire ecosystem of languages. Decomposing a compiler and its standard libraries into many parts is insufficient, because you're still viewing it as one language (really two languages, the language being compiled and the implementation language - the implementation language shouldn't be that important, but without interoperability, it is). It's wasteful to design a new standard library for every new language, wasteful not to share common parts between languages, and in addition I think a big reason interoperability is hard is because we haven't learned how to share these things. (Interestingly, I hold to a "waste not, want not" philosophy in everyday life, as in PL design.) I created LES and Loyc trees in that spirit. Because they are so general, they have disadvantages as syntax trees - they are not specialized for your specific circumstances and aren't suitable in all situations.* Yet because it was designed for all programming languages, it avoids common impedance-mismatches - for example e.g. most languages disagree about what "identifiers" and "literals" are, but LESv3 says "anything goes" so it is naturally compatible with everything. I think every compiler could find a use for these without the usual impedence mismatches you'd usually suffer when crossing language boundaries. Regardless of the syntax or semantics of your language, you can use LES as a text format to express internal compiler data structures, or accept it as an input format so users can combine your language with preprocessors like LeMP. (* Note that you could, though, generate wrapper types to wrap Loyc trees in a language-specific manner, or implement an interface like ILNode so that although your syntax trees aren't Loyc trees, they can be consumed as if they were.)

How does your particular decomposition of the compiler compare to other approaches like Roslyn? https://github.com/dotnet/roslyn/wiki/Roslyn%20Overview

This thread needs a link to your Loyc project: http://loyc.net Also one to your intro thread: https://futureofcoding.slack.com/archives/CC2JRGVLK/p1550368121006600