| How about incremental parser! [message #1091729] |
Wed, 21 August 2013 18:44  |
Eclipse User |
|
|
|
Hi,
I Finally finish my editor. Thanks a lot for help.
But I have a little performace problem.
In my editor, every time I type a key, my parser is called(AbstractSourceParser) and I call ANTLR Parser. That's ok, but, there are many sources with 20 thousand lines of code.
I know that has a lot of lines in source isn't a good ideia, but we have many legacy sources.
Well,
I would like to do incremental parser, like JDT.
Has DLTK support for that?
Regards,
Rodrigo
|
|
|
| Re: How about incremental parser! [message #1092717 is a reply to message #1091729] |
Fri, 23 August 2013 01:55  |
| Eclipse User |
|
|
|
Hi Rodrigo,
The parser from the editor is usually called with a small delay (0.5
second) to refresh the module structure (e.g. Outline), this operation
is called "reconcile".
As I understand, when you edit a file, only that single file should be
re-parsed.
Actually JDT has no incremental parser (or probably you mean something
different). It uses hand-written parser which is way faster than ANTLR
generated and can work in 2 passes - first extract classes and
functions, then (when needed) go deeper and parse each statement and
build full AST. For example, when refreshing java file structure
(Outline), only the 1st step is enough.
If your language grammar allows such an optimization - you can try
implementing this approach. Otherwise the only way is
profiling/optimizing the current parser/validations, and delaying some
operations if possible.
If files are always big and parser can't be optimized then the only
alternative could be disabling the reconciler at all.
Hope this helps.
Regards,
Alex
|
|
|
| Re: How about incremental parser! [message #1095203 is a reply to message #1092717] |
Mon, 26 August 2013 15:38 |
| Eclipse User |
|
|
|
Hello Alex,
Thanks again for the reply.
You are correct, I need compile only one source at a time but they are really big in some cases.
Studying this excellent article on IBM:
http://www.ibm.com/developerworks/opensource/tutorials/os-ecl-commplgin1/
I thought I'd do this:
- Determine where the function source code is being edited.
- Catch the beginning and the end of this function.
- Instantiate the ANTLR Parser, passing as input only this function.
- Create a new DLTK MODEL for this Function.
- Remove the old DLTK Model
- Add New DLTK Model
However, to do what I described above, I need to be able to do this:
- Some where to receiving the IRegion when we're typing in source.( I don't get this information in my AbstractSourceParser.
- How can I delete only one IMethod, and add other iMethod?
I'm studing also possible optimizations in ANTLR.
Regards,
Rodrigo
|
|
|
Powered by
FUDForum. Page generated in 0.03303 seconds
Home