Add cache for binary search in lines-map

[Matthias Simon]

The method syntax.Node.Span uses binary search to convert a file offset into a line number.
Location heavy tools, like our language server or formatter, would benefit from line-caches. For example like so:

func (t *tree) position(pos int) Position {
    if t.cachedLineBegin <= pos && pos < t.cachedLineEnd {
        line = t.cachedLine
    }
   // ...
}

To be efficient we probably need two separate line-caches, one for n.Pos() and one for n.End(). Best would be creating benchmarks comparing random position access, accessing only begin-offsets and accessing whole spans (begin- and end-offset).

[Matthias Simon]

Added benchmarks in #646

@nitinNT: Would this maybe interest you?

[Nitin Tejuja]

yes, i will try

[Nitin Tejuja]

@5nord : i have started working on this issue and i have noted the things which i need TODO and here are things

• modify tree struct present in syntax.go by adding fields (cachedLineBegin ,cachedLineEnd,cachedLine) and all of them will be int.
• modify searchLines function in syntax.go and modify newly added fields mentioned in above step according to binary search.

am i correct in above steps?

[Matthias Simon]

Sounds good to me.

What you could do is executing the benchmarks before and after your change to see how big the improvement is.

[Nitin Tejuja]

@5nord i have tried the above approach but it seems its not improving the performance. can you little guide me?

[Matthias Simon]

No worries. It happens sometimes that an idea is not as good as we initially thought. That's why we measure first. And a negative result can still be a valuable result 🙂

Sure I can help you. I suggest you push your PR as a draft and we'll have a look.