Stop dropping a leading "99" pair from Code128 barcodes (#251)

barcode/codex.py

@@ -267,7 +267,14 @@ def _convert_or_buffer(self, char: str) -> int | None:
         raise RuntimeError(f"Character {char} could not be converted in charset C.")
 
     def _try_to_optimize(self, encoded: list[int]) -> list[int]:
-        if encoded[1] in code128.TO:
+        # A START_C followed by an immediate charset switch can be folded into a
+        # single START code. In charset C, however, the code number 99 is the
+        # data pair "99" (not the TO_C switch marker), so it must never be folded
+        # away -- doing so silently dropped a leading "99" from the barcode.
+        if encoded[0] == code128.START_CODES["C"] and encoded[1] in (
+            code128.C["TO_A"],
+            code128.C["TO_B"],
+        ):
             encoded[:2] = [code128.TO[encoded[1]]]
         return encoded
 

docs/changelog.rst

@@ -1,6 +1,12 @@
 Changelog
 ---------
 
+Unreleased
+~~~~~~~~~~
+* Fixed a leading ``99`` digit pair being silently dropped from Code128
+  barcodes: in charset C the value ``99`` collided with the charset-C start
+  marker and was mistaken for a redundant switch code. (#251)
+
 v0.16.2
 ~~~~~~~
 * Add support for Python 3.13.

tests/test_builds.py

@@ -15,3 +15,38 @@ def test_ean8_builds_with_longer_bars() -> None:
     ean = get_barcode("ean8", "40267708", options={"guardbar": True})
     bc = ean.build()
     assert ref == bc[0]
+
+
+# START_C symbol as produced by ``_build`` before any charset switch.
+_START_C = 105
+
+
+def _decode_code128_c(encoded: list[int]) -> str:
+    """Decode a pure charset-C Code128 symbol list back to its digit string."""
+    assert encoded[0] == _START_C
+    return "".join(f"{n:02d}" for n in encoded[1:])
+
+
+def test_code128_leading_99_not_stripped() -> None:
+    # Regression test for #251: in charset C the pair "99" encodes to the code
+    # number 99, which collided with the TO_C switch marker. ``_try_to_optimize``
+    # therefore folded it away, silently dropping the leading "99" from the
+    # barcode (e.g. "9912345678" was encoded as "12345678").
+    bc = get_barcode("code128", "9912345678")
+    assert bc.encoded == [_START_C, 99, 12, 34, 56, 78]
+    assert _decode_code128_c(bc.encoded) == "9912345678"
+
+
+def test_code128_other_leading_pairs_unaffected() -> None:
+    # Pairs whose code number does not collide with a switch code always worked
+    # and must keep working.
+    for code, first_pair in (("0012345678", 0), ("1212345678", 12)):
+        bc = get_barcode("code128", code)
+        assert bc.encoded[:2] == [_START_C, first_pair]
+        assert _decode_code128_c(bc.encoded) == code
+
+
+def test_code128_start_charset_folding_preserved() -> None:
+    # The START-code folding optimisation (START_C + an immediate charset switch
+    # collapsed into a single START code) must still apply for genuine switches.
+    assert get_barcode("code128", "Wikipedia").encoded[0] == 104  # START_B