Wayne Conrad's Blog

home

Fortran, take IV

01 Jun 2014

I’ve failed to make a Fortran interpreter, written in Ruby, three times now.

Take 1 used the Treetop parser. It’s a great parser, but I could not figure out how to generate a parse tree. So I went looking for another PEG for Ruby, and found Parslet. So, on to:

Take 2, using the Parslet parser. This parser seems much nicer than Treetop, but again, the generation of the parse tree was too difficult. I just wasn’t “getting it.” I decided to take a break from Fortran and write a Basic interpreter, in order to learn Parslet using a simpler language. Having finished that, I went back to Fortran with renewed confidence. So, on to:

Take 3, against using the Parslet parser. At which I failed, again, when it came to generating the parse tree. I just don’t seem to be able to wrap my head around the mechanics of building a parse tree in Parslet. Among other things, left associative operators are a pain in Parslet (and, if I had to guess, in any PEG). I pretty much gave up on Fortran after take 3.

Then, I decided to write tutorial on how recursive descent compilers can be constructed in Ruby. A recursive descent compiler is insanely easy to write. There are a few reasons they’re not used much for serious work; the most important being that a recursive descent compiler is only efficent and easy when used to parse a context free grammar. Fortran, for example, is not context free.

Partway through that tutorial I realized that it would take only a few lines of code to allow arbitrary backtracking. With backtracking, a recursive descent compiler can parse languages such as Fortran. So, with the tutorial still unpublished, I’m on to:

Take 4, using a hand-rolled recursive descent compiler with backtracking.

Finally, I’m getting some traction. For the first time, I’ve got an interpreter that can execute some simple Fortran statements, such as:

      A = (1.0 + 2) / 3 ** (-4)
      PRINT *, 'A =', A

The interpreter has:

It’s not a lot, and much of it is not spec compliant, or is incomplete, but it’s a good start. The framework for the type system has been laid down, and it look much cleaner than it did in basic101. The lesson from basic101 is to keep a very clear distinction between Ruby types such as Integer, Float and String, and language types such as INTEGER, REAL, and CHARACTER. The two type systems should only be allowed to interact at clearly defined boundaries. I sort of stumbled into that in basic101, so it’s a bit messy there, but it’s clean here.

It remains to be seen whether the recursive descent compiler with backtracking is going to fly–Wikipedia warns that there could be trouble ahead, such as possibly requiring exponential time. But, forging ahead…

comments powered by Disqus