jonathanvdc
commented
5 years ago Using % for trivia sounds like a great idea! % is more succinct and it's also helpful for determining whether an attribute is "essential" or not. That's especially useful for compilers that accept Loyc trees as input. For example, ecsc currently reports an error whenever it encounters a special name it doesn't understand unless that name starts with #trivia_, in which case the node with the special name is just ignored. That's just quirky at best. Only ignoring nodes with names that start with % seems more elegant somehow. (Even though it is conceptually the same as the current solution.)
With regard to %MLComment vs %/**/, the latter is obviously more terse, but maybe having six leading non-alphanumeric characters in %/**/("This is a comment.") is a little bit over the top.
qwertie
Originally Loyc trees had a convention that there was a single character
#to mark all "special" identifiers. In this system, trivia (non-semantic information such as comments) were attached to things as attributes and required to have aNamethat starts with#trivia_, e.g.#trivia_SLCommentfor a single-line comment.Later I decided to switch to a different special prefix for operators, an apostrophe (why an apostrophe? To minimize visual noise - I liked that it was a small character. Also, I didn't want to use a punctuation mark that already connoted a specific operator). I didn't want to slow down the check that distinguishes "normal" and "special" names so rather than check if
name[0] == '\'' || name[0] == '#',LNode.IsSpecialNamechecks ifname[0] <= '\''. Thus any prefix below ASCII 40 is reserved for special names.Now I'm thinking that
#trivia_is a bit clunky - why not define a single-character prefix for trivia? At first I was thinking that the space character would be a good prefix, but then I remembered that#trivia_does have one virtue, it greps well: code referring to#trivia_is easy to find. With that in mind I think the best prefix is%, because%is a fairly rare character in most code, especially if that code is compiler-related. What do you think @jonathanvdc?The next question then is whether to keep an alphanumeric representation of trivia concepts, like
%SLCommentfor "single-line comment", or whether to go with a compact symbolic representation like%//- with the understanding that if one is parsing a language where comments are denoted# like thisor(* like this *)it would still be recommended to use%//and%/**/to represent those comments in the Loyc tree.