Implementing a custom syntax highlighter based on psi changes, can't use lexer Follow

Hello, I am working on a language plugin for IntelliJ platform.  The problem that I have is that this language  has a "position sensitive" lexer. It means that the same combination of symbols can be treated differently based on the context. For instance there is a syntax for java-like generics and you can also write xml like tags.  I can't determine it in the lexer, I even reworked PsiBuilder. All was good until I tried to implement syntax highlighting. I obviously can't use SyntaxHighlighter, so I tried to implement EditorHighlighter with PsiTreeChangeListener. It does not work well. It lags. And PsiTreeChangeListener leaks. It does not seem there is a good way to dispose of PsiTreeChangeListener in EditorHighlighter. I would be very grateful if someone could give me a tip or guide me into the right direction. 

// EditorHighlighterImpl.kt

override fun setEditor(editor: HighlighterClient) {
    this.editor = editor
    val editingVsFile = FileDocumentManager.getInstance().getFile(editor.document)
    val psiFile = psiDocumentManager.getPsiFile(editor.document)
    tokens = extractTokensFromPsiFile(psiFile)

    // FIXME: Leaks
    val psiTreeChangeListener = object : PsiTreeChangeAdapter() {
        private var lastModificationStamp: Long = -1

        override fun childrenChanged(event: PsiTreeChangeEvent) {
            if (editor is Editor && editor.isDisposed) {
                // A dummy way of tring to dispose of the listener
                psiManger.removePsiTreeChangeListener(this)
                return
            }

            val psiFile = psiFile ?: return

            if (psiFile.virtualFile != editingVsFile)
                return

            val newModificationStamp = psiFile.modificationStamp
            if (newModificationStamp == lastModificationStamp) {
                return
            }

            lastModificationStamp = newModificationStamp

            ApplicationManager.getApplication().executeOnPooledThread {
                runReadAction {
                    tokens = extractTokensFromPsiFile(psiFile) // This is SegmentArrayWithData
                }
                runInEdt {
                    editor.repaint(0, psiFile.textLength)
                }
            }
        }
    }

    psiManger.addPsiTreeChangeListener(psiTreeChangeListener, highlighterDisposable)
}

override fun createIterator(startOffset: Int): HighlighterIterator {
    return PsiHighlighterIterator(
        tokens,
        currentTokenIndex = tokens.findSegmentIndex(min(startOffset, tokens.lastValidOffset)),
        document = editor?.document
    )
}

// Langauge example

class A<tag> // here <tag> is lexed as LT, ID, LT

{

}

<tag>   <!-- here <tag> is lexed as OPEN_TAG -->
some plain text
class is not a keyword, keywords are treated as text
// this is not a comment, but simple raw text
<!-- this one is a comment -->
</tag>