Hey all, Thanks to everyone for replying. I was severly caught up with work for the last two days. The link Larry gave for the lookahead states seems very nice :). But I am not sure if Ill be looking to calculate the lookahead states myself, or let BNFC do the job? In either case, I think merely printing the line number where the error occured should do in the worse case. It will be exciting to have nice error messages though. I am starting to reliase the advantage of pattern matching being present, like Chris had said. I mean it definitely beats maintaining and checking a flag, the way you would do it in C :)

Definitely not, just go for it. In the IPL course @ UU we implemented Thanks, Chris :).

1. Parse input file into an abstract syntax tree representation. 2. Perform type checking on your syntax tree. 3. Transform the syntax tree using re-write rules and optimisations. 4. Pretty print the syntax tree in order to output code of your target language.

That seems like a very nice scheme to follow. I had a similar method in mind. Step 3 is what I am really worried about. How easy/difficult will it be in a pure func language, to "transform" the sytnax tree. I have to take a deeper look at BNFC. But from an initial look, it seems way too powerful for me to use? At least as powerful as yacc. And that with Haskell, should give me a very good toolset-advantage? @Chris : The ipl course website, seems very helpful, especially some of the lectures on abstract syntax. -- Pranesh Srinivasan, Third Year Student, Computer Science & Engineering, Indian Institute of Technology - Madras. http://spranesh.googlepages.com http://www.cse.iitm.ac.in/~spranesh/

On 10/23/08 03:23, Joel Björnson wrote:

On Thu, Oct 23, 2008 at 12:18 AM, Pranesh Srinivasan wrote:

[snip]

...

That seems like a very nice scheme to follow. I had a similar method in mind. Step 3 is what I am really worried about. How easy/difficult will it be in a pure func language, to "transform" the sytnax tree.

I think the pureness is actually an advantage here, since the transformation is a function from one syntax tree to another. If however, you feel the need of keeping a state and you don't wanna pass around this explicitly, you may consider using a state monad.

...

I have to take a deeper look at BNFC. But from an initial look, it seems way too powerful for me to use? At least as powerful as yacc. And that with Haskell, should give me a very good toolset-advantage?

If the project is about writing parsers then using BNFC would probably be considered cheating. But for practical purposes I think it's a great tool that would save you a lot of work.

- Joel

HI Joel. Does BNFC calculate the lookahead sets for a grammar? I'm attempting to do this with haskell, but I would hate to "reinvent the wheel". I would think if haskell can tranform a syntax tree, then haskell could transform a grammar tree to calculate the lookahead sets. My purpose it not so much create a compiler as illustrate how it's done. I found the Dragon's book description hard to follow and thought redoing it using functors (to transform the tree into another, with different operators) then simplify the transformed trees (one tree for each lookahead set) then find the fixpoint of those trees (really forest: 1 tree for each production) would be a good way to demonstrate the method. -Larry