[Jscheme-user] (J)Scheme parsing with (or without?) jsint.InputPort

[Michael R H. <bu...@zc...>]

I've been spending some time with my JScheme plugin for Eclipse again (I
just added matching parenthesis highlighting to the version in CVS), and
I've come to a point at which I have to really parse the input buffer to
add the next features on my TODO:
http://cvs.zclipse.org/viewcvs.py/zclipse/org.zclipse.jscheme/TODO?view=markup

For details on the plugin: http://www.zclipse.org/projects/jscheme

I've toyed with rolling my own parser (which I've started twice: once
ad-hoc, and once from the r4rs spec), writing a SableCC grammar for
JScheme, and using the jsint.InputPort.read() function.

None of these has struck me as quite perfect. 

Rolling my own gives me the most control over error reporting and
recovery, and I would have full access to all parts of the input
(including comments), but it's a lot of work that has already been done
in JScheme itself, and it would likely be less sharable with other
projects that might have similar needs.

Writing a SableCC grammar is less work, more sharable in some ways
(possibly directly contributable to the jscheme project itself), and
generates a very pleasant AST for traversal, but isn't so good at error
reporting or recovery during the parse (at least in the current release
of SableCC).

Using InputPort.read() is great because I would be using precisely the
same parsing mechanism as JScheme itself, and it's already been written.
However, to my mind, API isn't quite amenable for general use. Since it
returns Object, I have to use Java's RTTI features to examine and
process the parsed data (or I have to write parts of my code in JScheme
to process the input) -- I'd prefer a syntax tree that I implements the
Visitor design pattern for traversal so I don't have to do any
downcasting. My understanding is also that because it's meant for
execution, the InputPort throws away comments and will print errors to
System.err. Basically, in my understanding, it's not meant to be a Java 
API for parsing JScheme. 

It might not be too bad to cause the InputPort.read() to return an AST,
instead of Object, possibly adding comments to the syntax tree. It would
probably be harder to push error handling into the API level.

Does anyone have any other recommendations for this? Once I have an AST
that I can work with, it should be easy to add a TreeView of the source
and a source formatter, the two major remaining features I want before I
release 1.0.0 of the plugin.

mike

-- 
Michael R Head <bu...@zc...>
ZClipse -- OpenSource Eclipse plugins