Fix Code128 build() producing different output on repeated calls

[apoorvdarshan]

Fixes #143

Root cause

Code128._build mutates self._charset and self._digit_buffer while it walks the code (switching charsets via _new_charset, buffering digit pairs in charset C). Both attributes are initialised only once, in __init__. When a build ends in charset A/B — or with a digit still buffered — those attributes are left in a "dirty" state. The next call to build() / encoded starts from that stale state instead of the intended _charset = "C" / empty buffer, so it emits a different (and incorrect) bit string.

Reproduction

from barcode import Code128

inst = Code128("12A")
for _ in range(3):
    print(inst.build()[0])

Observed (before the fix) — the first call is correct, later calls diverge:

11010011100101100111001011110111010100011000100100011001100011101011
11010010000100111001101100111001010100011000110001011101100011101011
11010010000100111001101100111001010100011000110001011101100011101011

The divergence is the start code: the first build begins in charset C (START_C = 105); after it the instance is left in charset B, so later builds begin with START_B = 104. A freshly constructed Code128("12A") always reproduces the first (correct) line, confirming the later output is the buggy one.

After the fix every call returns the same, correct value:

11010011100101100111001011110111010100011000100100011001100011101011
11010011100101100111001011110111010100011000100100011001100011101011
11010011100101100111001011110111010100011000100100011001100011101011

The original report's all-digit example (Code128("1170773")) never left charset C, so it happened to be stable on current main; inputs that switch charset (e.g. "12A", "123ABC456") expose the bug.

Fix

Reset self._charset = "C" and self._digit_buffer = "" at the start of _build(), so each build begins from the same initial state regardless of what the previous one left behind. This is the smallest localized change and touches only the entry of _build.

Tests

Added tests/test_code128.py with three regression tests asserting that repeated build() / encoded calls on one instance are deterministic and that a reused instance matches a freshly constructed one. These tests fail on unmodified main and pass with the fix. The full existing suite (52 tests, including the GS1-128 charset-C build tests) continues to pass, and ruff check / ruff format --check are clean.

Disclosure: prepared with AI assistance; reviewed and verified locally.