export_json_tags() in ezdxf, a Python library for DXF drawing files, wrote JSON that json.loads rejected, or read back as a different string, when a string value held a backslash or a double quote. I fixed it in PR #1433, merged about three and a half hours after I opened it.

When it happens

Every MTEXT with a \P line break does it:

import json

import ezdxf
from ezdxf.document import export_json_tags

doc = ezdxf.new()
doc.modelspace().add_mtext("first\\Psecond")
json.loads(export_json_tags(doc))

The text is first\Psecond, with one backslash. json.loads raised JSONDecodeError: Invalid \escape. A double quote in a value broke the output too, in the compact and the verbose format.

How the export works

export_json_tags() runs the document through JSONTagWriter, which writes the DXF tags as [group-code, value] pairs. Every string value goes into this template. The verbose format sends numbers through it too; the compact format writes numbers and vertices without it:

JSON_STRING = '[{0}, "{1}"],\n'

Why the gap existed

JSONTagWriter has put the value between quotes as it is since it was added in March 2024. The next day the compact format renamed the template JSON_STRING. The old tests export a new drawing, or binary data that comes out as hex. None has text with a backslash or a double quote.

The fix

The template no longer adds the quotes. write_tag2 encodes the value with json.dumps(str(value), ensure_ascii=False). The diff is trimmed: it leaves out the one-line import json hunk and shows no context lines.

@@ -275 +276 @@ class JSONTagWriter(AbstractTagWriter):
-    JSON_STRING = '[{0}, "{1}"],\n'
+    JSON_STRING = "[{0}, {1}],\n"
@@ -315 +316,2 @@ class JSONTagWriter(AbstractTagWriter):
-        self._stream.write(self.JSON_STRING.format(code, value))
+        text = json.dumps(str(value), ensure_ascii=False)
+        self._stream.write(self.JSON_STRING.format(code, text))

The test

test_export_json_tags_escapes_string_values runs in both formats. It adds an MTEXT with the text first\Psecond "quoted", exports the document, parses the result with json.loads, loads it with load_json_tags() and compares the text with the input.

Before the fix both cases failed. After it all eight tests in the file passed.

What the maintainer did

@mozman merged it without a review comment. It was my first ezdxf PR.

The lesson

  • Let the encoder quote the value. A template with quotes around {1} says that no value holds a quote or a backslash.
  • Put the format’s reserved characters in the test. Parse the output with a real parser and load it back with the library’s own loader. A new drawing holds no value that needs escaping.
  • Check a library default against the output you must keep. json.dumps escapes non-ASCII text unless ensure_ascii=False, and the old template wrote it as it was. I compared old and new output for a drawing with non-ASCII text, a line and a polyline. They differed in two lines that change on every export.

If a writer in your own code wraps values in quotes with a template, reply on X.