|
|
|
|
|
|
by munificent
1105 days ago
|
|
|
Almost all real-world projects that are language-like or compiler-like will need a parser. A much smaller fraction of them will need register allocation, instruction selection, optimization, code generation, etc. For every big, deep, native code compiler, there are a hundred template languages, config files, report generators, etc. all of which are real programs providing real value for actual people. Emphasizing parsing provides the most value for the greatest number of people. The folks that do end up needing more back end depth will still have the resources available to learn it. |
|
|
Everybody and their dog thinks it necessary to inflict some new sub-par language on us when in about 99.9% of cases they should just either have stuck to s-expressions or some suitable subset of a popular programming or existing config language with a relatively sane syntax (blaze/bazel did that right, cmake did that very very wrong).
When was the last time you looked at some config file and thought, wow I'm so glad they didn't use toml or python or whatever, but instead made up some completely new syntax nothing in the world apart from this tool itself can parse and that I can't programaticaly manipulate?
When was the last time you thought, wow I am so glad that someone invented a new templating language that creates some new injection vulnerabilities, because no one apart from the lisp people ever seem to have worked out that if you want to interpolate into something tree shaped, you should have a tree-based interpolation syntax? Because although sexps and quasiquote solve this very nicely and concisely everyone else still seems to love string-bashing plus some ad-hoc "escaping" system for this. And one reason for this is of course precisely the enormous abundance of idiotic config languages that can't be easily manipulated as anything than opaque strings.
[Edit: if you do create a new config file language, pretty please provide some means to directly query and losslessly manipulate it; for the lossless part you will either need to have first class comments unambiguously attached to a particular syntactic construct and agreed upon deterministic formatting or IDE-style complexity, the first one is probably a better idea]