Lookahead not evaluated for first word after pushing with (?=\S)

[deathaxe]

Description

If a context is pushed onto the stack by (?=\S) a lookahead which points to a word (e.g.: (?=import\b)) is ignored when evaluating the next nonwhitespace character.

Note: If the (?=import\b) was replaced by import\b it was evaluated, but then breaking the syntax due to popping without pushing back into the context.

Steps to reproduce

1. Open a vanilla ST.
2. Create a new syntax and past the snippet below.
3. Create an empty file and assign it the new syntax.
4. Type several lines of import.

%YAML 1.2
---
# http://www.sublimetext.com/docs/3/syntax.html
name: Lookahead
scope: source.lookahead

contexts:
  main:
    - match: import\b
      scope: keyword.control.directive
      push:
        - meta_scope: meta.preprocessor.import
        - match: (?=\S)
          push:
            - meta_content_scope: meta.path
            - match: (?=import\b)  # not evaluated for the first word after "import"!
              pop: 2
            - match: ;
              scope: punctuation.terminator
              pop: 2
            - match: \w+\b
              scope: entity.name.class

Expected behavior

Each import keyword should be scoped as keyword.control.directive

Actual behavior

Each 2nd import keyword is scoped/highlighted as entity.name.class, because the (?=import\b) lookahread in line 15 is not evaluated and therefore the bail out does not work.

Environment

• Build: 4061
• Operating system and version: Windows 10 x64 1903

[wbond]

This is a feature that prevents infinite recursion. If a zero-width push is followed by a zero-width pop, it will result in an infinite loop.

However, with the ability to pop multiple levels, we may now have to tweak the rules to allow things that weren't possible before.

[wbond]

This has been added in build 4075. There is now more complex logic to handle situations where pushes and pops are asymmetrical, and thus can be allowed even though both are zero-width.