Files
doslang-mirror/tools/tests/test_host_syntax.py
T
coolguy 32ad50b14a dev: run the host gate as 16-bit, and drop what it found dead
The gate was using a system-wide Open Watcom that only ships the 32-bit
compiler. The project downloads its own toolchain, and .dosboxx/watcom/binnt
has the Windows-hosted 16-bit wcl.exe -- the same compiler and the same target
as the DOS build. Run that instead, with build-dos.bat's exact command.

The difference is not academic. Compiling 16-bit immediately reported three
functions the unification had orphaned: emit_type_helpers, which the M7
program emitter replaced, and emit_drop_helpers and emit_drop_fields, which it
was the only caller of. The 32-bit check had been clean.

Those warnings were going to the DOS screen, where the runner cannot see them:
COMMAND.COM redirects handle 1 only, and Watcom writes diagnostics to handle 2.
So the suite was green while the build was not quiet. Nothing in the runner
would ever have said so.

Found while chasing W210/W107 reported from a DOS screen, which this does not
yet explain -- those are not among what the compiler build emits now.

M1-M7: 214 passed.
2026-08-17 02:22:44 +09:00

116 lines
4.5 KiB
Python

"""Compile the compiler's own sources on the host, as a syntax gate.
This is not verification. AGENTS.md is explicit that a host compiler's result
does not count, and it still does not: the DOS build and the milestone suite
decide whether anything works. What this buys is the turnaround. A missing
declaration or a signature that disagrees with its definition used to surface
only after a DOSBox-X boot and a full compiler build; here it surfaces in about
a second, with the line number.
It runs the pinned toolchain's Windows-hosted 16-bit driver with the exact
command ``fec/build-dos.bat`` uses, so the diagnostics match what the DOS build
sees. A 32-bit compile is not equivalent: it misses warnings that only the
16-bit large model reports, which is how a dead function survived the M7
unification with a clean 32-bit check.
What it still cannot see is the DOS environment itself -- memory limits, the
command line length, the filesystem. The DOS build and the milestone suite
remain the gate.
Skipped when the toolchain has not been downloaded, so the suite runs anywhere.
"""
from __future__ import annotations
import os
import re
import shutil
import subprocess
from pathlib import Path
import pytest
from ferrolang_vm.paths import ROOT
SRC = ROOT / "fec" / "src"
# Mirrors the compile order in fec/build-dos.bat.
SOURCES = ("arena", "diag", "lexer", "ast", "parser", "types", "m7", "own",
"check", "lower", "emit_c", "driver")
def _watcom() -> Path | None:
"""Prefer the toolchain the runner pins over anything installed system-wide."""
root = os.environ.get("WATCOM")
candidates = [ROOT / ".dosboxx" / "watcom"]
if root:
candidates.append(Path(root))
for base in candidates:
if (base / "binnt" / "wcl.exe").is_file():
return base
return None
@pytest.fixture(scope="session")
def watcom() -> Path:
base = _watcom()
if base is None:
pytest.skip("run `ferro-dos setup` to download the pinned toolchain")
return base
@pytest.fixture(scope="session")
def objdir(tmp_path_factory: pytest.TempPathFactory) -> Path:
return tmp_path_factory.mktemp("wcc")
@pytest.mark.parametrize("name", SOURCES)
def test_source_compiles_clean(name: str, watcom: Path, objdir: Path) -> None:
source = SRC / f"{name}.c"
if not source.is_file():
pytest.fail(f"{source} is missing but build-dos.bat compiles it")
env = dict(os.environ)
env["WATCOM"] = os.fspath(watcom)
env["INCLUDE"] = os.fspath(watcom / "h")
env["PATH"] = os.pathsep.join(
[os.fspath(watcom / "binnt"), env.get("PATH", "")])
# The same command build-dos.bat runs, minus the object name.
completed = subprocess.run(
[os.fspath(watcom / "binnt" / "wcl.exe"), "-q", "-za", "-wx",
"-bt=dos", "-ml", "-k32768", "-c", f"-i={SRC}", os.fspath(source)],
cwd=objdir, capture_output=True, text=True, env=env, timeout=120,
)
output = (completed.stdout + completed.stderr).strip()
# -wx keeps warnings meaningful, so treat any diagnostic as a failure. The
# DOS build prints them to a screen nobody reads, which is how they
# accumulate unnoticed.
assert completed.returncode == 0 and not output, (
f"{name}.c does not compile clean\n{output}"
)
def test_build_scripts_agree_on_sources() -> None:
"""The two build files and this test must name the same translation units.
They drifted apart while the M7 wrapper existed, which is how a source could
stop being compiled without anyone noticing.
"""
batch = (ROOT / "fec" / "build-dos.bat").read_text(encoding="utf-8", errors="replace")
makefile = (ROOT / "fec" / "Makefile").read_text(encoding="utf-8", errors="replace")
in_batch = set(re.findall(r"src\\(\w+)\.c", batch))
srcline = next(line for line in makefile.splitlines() if line.startswith("SRC ="))
in_make = set(re.findall(r"src/(\w+)\.c", srcline))
assert in_batch == set(SOURCES), f"build-dos.bat compiles {sorted(in_batch)}"
assert in_make == set(SOURCES), f"Makefile compiles {sorted(in_make)}"
def test_no_source_is_orphaned() -> None:
"""Every .c under fec/src must be compiled by something.
check_m7.c and emitcm7.c hid check.c and emit_c.c from the build by
including them textually; nothing flagged that they had stopped being
translation units of their own.
"""
on_disk = {p.stem for p in SRC.glob("*.c")}
assert on_disk == set(SOURCES), (
f"fec/src has {sorted(on_disk - set(SOURCES))} that no build step compiles"
)