bug#70007: [PATCH] native JSON encoder

["Mattias Engdegård" <mattias.engdegard@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 708 bytes --]

Looks like the new serialiser inherited a bug from the old code: `json-insert` in a unibyte buffer does not move point correctly. Example:

(with-temp-buffer
  (set-buffer-multibyte nil)
  (json-insert "é")
  (list (buffer-string) (point)))
=> ("\"\303\251\"" 4)

The string is correct but the position should be 5, not 4.

This made me look at the Fjson_insert logic a bit. I'm probably betraying my lack of knowledge about buffer subtleties here, but since the serialiser always produces (correct) UTF-8, shouldn't it be enough to copy the bytes, don't bother with any decoding, and perform the buffer insertion ceremonies?

Proposed patch attached. (There will also be a test, of course.)


[-- Attachment #2: json-insert.diff --]
[-- Type: application/octet-stream, Size: 2025 bytes --]

diff --git a/src/json.c b/src/json.c
index c3244ad04d2..7c62d63c6fd 100644
--- a/src/json.c
+++ b/src/json.c
@@ -656,39 +656,14 @@ DEFUN ("json-insert", Fjson_insert, Sjson_insert, 1, MANY,
   /* No need to keep allocation beyond this point.  */
   unbind_to (count, Qnil);
 
-  ptrdiff_t inserted = 0;
+  bool ub_buffer = NILP (BVAR (current_buffer, enable_multibyte_characters));
   ptrdiff_t inserted_bytes = jo.size;
+  ptrdiff_t inserted = ub_buffer ? jo.size : jo.size - jo.chars_delta;
 
-  /* If required, decode the stuff we've read into the gap.  */
-  struct coding_system coding;
-  /* JSON strings are UTF-8 encoded strings.  */
-  setup_coding_system (Qutf_8_unix, &coding);
-  coding.dst_multibyte = !NILP (BVAR (current_buffer,
-				      enable_multibyte_characters));
-  if (CODING_MAY_REQUIRE_DECODING (&coding))
-    {
-      /* Now we have all the new bytes at the beginning of the gap,
-	 but `decode_coding_gap` needs them at the end of the gap, so
-	 we need to move them.  */
-      memmove (GAP_END_ADDR - inserted_bytes, GPT_ADDR, inserted_bytes);
-      decode_coding_gap (&coding, inserted_bytes);
-      inserted = coding.produced_char;
-    }
-  else
-    {
-      /* Make the inserted text part of the buffer, as unibyte text.  */
-      eassert (NILP (BVAR (current_buffer, enable_multibyte_characters)));
-      insert_from_gap_1 (inserted_bytes, inserted_bytes, false);
-
-      /* The target buffer is unibyte, so we don't need to decode.  */
-      invalidate_buffer_caches (current_buffer,
-				PT, PT + inserted_bytes);
-      adjust_after_insert (PT, PT_BYTE,
-			   PT + inserted_bytes,
-			   PT_BYTE + inserted_bytes,
-			   inserted_bytes);
-      inserted = inserted_bytes;
-    }
+  insert_from_gap_1 (inserted, inserted_bytes, false);
+  invalidate_buffer_caches (current_buffer, PT, PT + inserted);
+  adjust_after_insert (PT, PT_BYTE, PT + inserted, PT_BYTE + inserted_bytes,
+		       inserted);
 
   /* Call after-change hooks.  */
   signal_after_change (PT, 0, inserted);