[PaulHoule]

Each PEG generator promises a revolution but only burns a car.

I was disappointed with how they do operator precedence; they use the usual trick to make a PEG do operator precedence which looks cool when you apply it to two levels of precedence but if tried to implement C or Python in it it gets unwieldy. Most of your AST winds up being nodes that exist just to force precedence in your grammar, working with the AST is a mess.

For all the horrors of the Bell C compilers, having an explicit numeric precedence for operators was a feature in yacc that newer parser gens often don't have.

I worked out the math and it is totally possible to add a stage that adds the nodes to a PEG to make numeric precedence work and also delete the fake nodes from the parsed AST. Unparsing I'm not so sure of, since if someone wrote

   int a = (b + c);

how badly you want to keep the parens is up to you; a system like that MUST have an unparse-parse identity in terms of 'value of the expression', but for sw eng automation you want to keep the text of the source code as stable as you can.