d-plaindoux / celma Goto Github PK

Since the Celma parser produces an AST we are able to traverse it in order to optimize the produced parser. This can be done by computing the first(s) element in order to have a more predictive parser removing try and backtracks.

rust doc should be written.

First of all, thanks for the work. Especially the macro, really making char to char consuming more clear (meaningful).

While I was reading the README, one thing I did not quite get. What do those alters mean?

I guess ^ means "not any", ! means "not all". But I have no idea what does # mean.

For example, in this parsec rule let STRING:{String} = ('"' c=#((("\"" -> {'\"'})|^'"')*) '"') -> {mk_string(c)}, I assuming that it consumes the whole string of "something". If I want to stop when next char is double quote, but leave it in the stream?

Parsing with Not is not possible at all. The compilation fails.

First of all, I do not know if this is Celma issue, or just how proc macro works.

When I make parsec_rules, I write '\"' in code. Then use cargo expand. The expanded code contains '\"' as expected.

But when I write "\"", the expanded code is "\\\"".

Celma supports generic streams. The Celma parser metalanguage should be generalized and not only reduced to input like a stream of chars.

The lookahead should be implemented in

• the core library
• as an extension of the macro-based language.

For example, bash variable substitution. The input is written as "This is ${prog_name} speaking". In my opinion, it should be parsed into something like [PLAIN("This is "), VAR("prog_name"), PLAIN(" speaking")]. But when I write the plain parser, I could not figure out how to "consume any chars until string '${'".

The current implementation requires the same type for the negated parser and the result.