Re: [CEDET-devel] Incremental parser problem

[David P. <Dav...@wa...>]

Hi Eric,

[...]
> Do you have a reliable reproducible case=3F

Here it is ;-)

1. Create the following simple test.java file:

class test
{
  
  void method1() {
  }
  
  
  void method3() {
  }
  
}

2. Save the file, close the buffer, and open it again to get a clean
   cache like this:

(("test" type "class"
  (("method1" function "void" nil nil nil
    ((reparse-symbol . class_member_declaration))
    #<overlay from 19 to 39 in test.java>)
   ("method3" function "void" nil nil nil
    ((reparse-symbol . class_member_declaration))
    #<overlay from 48 to 68 in test.java>))
  nil nil nil nil #<overlay from 1 to 73 in test.java>))


3. Just insert a new declaration between `method1' & `method3', like
   this:

class test
{
  
  void method1() {
  }
  
  int i;
  
  void method3() {
  }
  
}

The incremental parser should say: "Inserted tokens: (i)" and you
would be able to navigate to the new token using senator (which uses
the overlays) :-)

But M-x bovinate still shows:

(("test" type "class"
  (("method1" function "void" nil nil nil
    ((reparse-symbol . class_member_declaration))
    #<overlay from 19 to 39 in test.java>)
   ("method3" function "void" nil nil nil
    ((reparse-symbol . class_member_declaration))
    #<overlay from 57 to 77 in test.java>))
  nil nil nil nil #<overlay from 1 to 82 in test.java>))

The new token is not in the cache, it is only in the buffer :-(

A side effect is that the semantic imenu (built from the cache) will
never show the new `i' variable token.

As I wrote all the above I wonder if rebuilding the cache from tokens
in the buffer (that is from the hierarchy of 'semantic' tagged
overlays) could be possible=3F  I can see the following benefits to a
such solution:

- side effect free
- no more need for splice functions
- buffer/cache consistency

The main drawback could be a slow down of the incremental parser when
processing large token trees.

Hope it helps.
Thanks!

David