The fix guard
An autofix that breaks a template is worse than no autofix. Pug makes that easy: indentation is structure, and moving one line without its children changes the tree. So every fix that touches a Pug template is checked before ESLint applies it.
What is checked
For each Pug template, a fix is applied only if the template still parses afterwards and its structure is unchanged:
- the tags and their nesting;
- each tag's class and id set and attribute count (so
p(class="a b")andp.a.bare equal, but losing a class is not); - the shape of text and interpolations;
- every piece of JavaScript that Pug compiles (unquoted and
${}attribute values,&attributes, mixin arguments, code, control heads,#{}), up to its indentation.
The parser reports what Pug itself rejects (for example a tab/space mix or a repeated attribute), so a fix that would introduce it is dropped too.
How fixes are applied
- Each fix is verified on its own.
- The surviving fixes are verified again as the combined set that ESLint will apply.
- A fix that fails is dropped, and its problem is still reported, so you can fix it by hand.
This covers every rule that fixes .vue files, including vue/* rules and third-party rules. While a template has a Pug syntax error, no fix is applied inside it.
Cost
The guard parses the template again for each distinct fix, so its cost grows with the number of fixes times the template size. A normal template takes milliseconds; a fully misformatted 3000-line template can take several seconds. The cost only arises while a template has problems to fix.
Partial fix sets
The guard verifies the full set of surviving fixes. If ESLint applies only some of them (--fix-type, or a custom fix predicate in the Node.js API), that subset was not verified as a set.