Special characters should be preserved when using a placeholder in a regexp query

[hubgit]

Describe the issue

When replacing text using a regexp query, with a placeholder in the replacement, special characters such as \t and \n will be replaced, when they should be preserved.

For example, in a JS console:

> '\\test'.replace(/(\\[a-z]+)/, ':$1:')
> ':\test:'

In CodeMirror, this replacement currently results in : est:.

I think this can be fixed by moving .replace(…) so it runs after .unquote(…) in the getReplacement function, I'm just not sure whether this would have any negative effects.

https://github.com/codemirror/search/blob/7a4a57e52b4af5807d195296c9933c4c7df947f5/src/search.ts#L295-L301

Browser and platform

Chrome, macOS

Reproduction link

https://codemirror.net/try/?c=aW1wb3J0IHtiYXNpY1NldHVwLCBFZGl0b3JWaWV3fSBmcm9tICJjb2RlbWlycm9yIgoKbmV3IEVkaXRvclZpZXcoewogIGRvYzogYC8vIG9wZW4gdGhlIHNlYXJjaCBwYW5lbCwgZW5hYmxlIHRoZSByZWdleCBvcHRpb24KLy8gYW5kIHJlcGxhY2UgIihcXFxcW2Etel0rKVxcbiIgd2l0aCAiOiQxOlxcbiIKXFx0ZXN0XG5cXHRlc3RcbmAsCiAgZXh0ZW5zaW9uczogW2Jhc2ljU2V0dXBdLAogIHBhcmVudDogZG9jdW1lbnQuYm9keQp9KQo=

[hubgit]

I should add that literal is set to false for these queries, as we do still want to allow matching newlines with \n in the actual query.

[marijnh]

Good point, I agree that those calls should be swapped.