[dexhunter]

TL;DR

• This PR makes parse_asm in src/tiny8/assembler.py faster. Every assemble() call and every compile_c -> assemble test uses it.
• It runs the displacement regex only on lines that contain a +. It also builds the X/Y/Z register map once per call.
• Measured parse time for the test-suite listings (median of 9 paired timings): 58.1 ms -> 39.1 ms, 32.7% faster.
• pytest tests passes (342 passed). ruff check and ruff format --check are clean.

What & Why

Before, parse_asm called re.sub on every instruction line. Most lines have no +, so the regex can never match there. The code also built the {"X": 26, "Y": 28, "Z": 30} dict again for each X/Y/Z operand.

Now the code checks "+" in line first and uses the compiled pattern only when it is true. It builds the register map once, before the loop. The pattern needs a literal +, so skipping lines without one gives the same result. p.upper() in reg_map is true for exactly the same operands as p.upper() in ("X", "Y", "Z"). The program, labels, pc_to_line and errors stay identical.

Results

Rows 1-3: best of 15 timing repeats, median of 3 alternating rounds. Row 4: median of 9 paired before/after timings. Same machine and Python 3.12:

Situation	Before	After	Change
Compiled C program (compile_c output, 154 lines)	0.268 ms	0.169 ms	36.8% faster
All examples/*.asm programs	136.0 ms	83.9 ms	38.3% faster
Listing where every line has a Y + n / Z+n displacement (400 lines)	0.988 ms	0.862 ms	12.7% faster (regex still runs; no slowdown)
C test-suite listings + test snippets (109 listings, median of 9 paired timings)	58.1 ms	39.1 ms	32.7% faster

Why this is safe and easy to maintain

The change is 9 lines in one function and uses plain Python. The regex is still the same, and lines without a + skip it because it cannot match them. tests/test_assembler.py, tests/test_compiler.py, tests/test_control_flow.py and tests/test_cpu_edge_cases.py cover labels, comments, displacement operands and bad operands.

Validation

• pytest tests: 342 passed on this branch (342 passed on main).

• ruff check src/tiny8/assembler.py and ruff format --check src/tiny8/assembler.py: clean.

• Before/after timing: the suite listings above took 58.1 ms -> 39.1 ms (32.7% faster, 1.49x) in the median of 3 separate measurements. The 3 measurements gave 0.679, 0.671 and 0.673 of the main time. An automated output check compared programs, labels, pc_to_line, source_lines and error messages with main, and they were identical.

• Quick reproduction:

  python -m timeit -s "import glob; from tiny8 import assemble; srcs = [open(f).read() for f in glob.glob('examples/*.asm')]" "[assemble(s) for s in srcs]"

AI assistance

Code candidates were generated by Codex (gpt-6.1-sol) and tracked with Weco Observe. This PR uses the smallest readable candidate. The tests, timings and an automated output-equality check above were run on it.

Weco share link: https://dashboard.weco.ai/share/gLEKWXkUReYHkIjz2BXzPLKDXaEHdYK2

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!