Repository navigation
[Bug]: A string ending in a backslash loses it and gains a quote during repair #247
Description
Version of the library
Tested on 0.64.2 (commit f13a96b). I read the same code on main at f0eff1e (0.64.4) but did not run it there.
Describe the bug
When a document needs repair, a string that ends in an escaped backslash is corrupted. This also happens with skip_json_loads=True. A truncated Windows path is the plain case: the value comes back with a " where the backslash was. In an array, the closing ] gets pulled into the string too.
The even-run branch in _normalize_escape_sequence excludes a following quote:
json_repair/src/json_repair/parse_string.py
Line 302 in f0eff1e
So \\" falls to the generic branch, whose while loop reads that quote as an escaped one:
json_repair/src/json_repair/parse_string.py
Lines 306 to 320 in f0eff1e
I don't know whether that exclusion is there for doubly escaped quotes, so I haven't suggested a patch.
How to reproduce
from json_repair import repair_json
repair_json('{"path": "C:\\\\"', return_objects=True)
# {'path': 'C:"'}
repair_json('["R\\\\"]', return_objects=True, skip_json_loads=True)
# ['R"]']The JSON text in the first call is {"path": "C:\\", cut off before the closing brace.
Expected behavior
The backslash stays and the quote closes the string, as json.loads reads the complete text: {'path': 'C:\\'} and ['R\\'].
Activity