mirror of
https://github.com/uutils/coreutils
synced 2024-11-14 00:47:11 +00:00
288 lines
8.3 KiB
Python
288 lines
8.3 KiB
Python
#!/usr/bin/env python3
|
|
import multiprocessing
|
|
import subprocess
|
|
import argparse
|
|
import csv
|
|
import sys
|
|
from collections import defaultdict
|
|
from pathlib import Path
|
|
|
|
# third party dependencies
|
|
from tqdm import tqdm
|
|
|
|
# spell-checker:ignore (libs) tqdm imap ; (shell/mac) xcrun ; (vars) nargs
|
|
|
|
BINS_PATH = Path("../src/uu")
|
|
CACHE_PATH = Path("compiles_table.csv")
|
|
TARGETS = [
|
|
# Linux - GNU
|
|
"aarch64-unknown-linux-gnu",
|
|
"i686-unknown-linux-gnu",
|
|
"powerpc64-unknown-linux-gnu",
|
|
"riscv64gc-unknown-linux-gnu",
|
|
"x86_64-unknown-linux-gnu",
|
|
# Windows
|
|
"aarch64-pc-windows-msvc",
|
|
"i686-pc-windows-gnu",
|
|
"i686-pc-windows-msvc",
|
|
"x86_64-pc-windows-gnu",
|
|
"x86_64-pc-windows-msvc",
|
|
# Apple
|
|
"aarch64-apple-darwin",
|
|
"x86_64-apple-darwin",
|
|
"aarch64-apple-ios",
|
|
"x86_64-apple-ios",
|
|
# BSDs
|
|
"x86_64-unknown-freebsd",
|
|
"x86_64-unknown-netbsd",
|
|
# Android
|
|
"aarch64-linux-android",
|
|
"x86_64-linux-android",
|
|
# Solaris
|
|
"x86_64-sun-solaris",
|
|
# Illumos
|
|
"x86_64-unknown-illumos",
|
|
# WASM
|
|
"wasm32-wasi",
|
|
# Redox
|
|
"x86_64-unknown-redox",
|
|
# Fuchsia
|
|
"aarch64-fuchsia",
|
|
"x86_64-fuchsia",
|
|
]
|
|
|
|
|
|
class Target(str):
|
|
def __new__(cls, content):
|
|
obj = super().__new__(cls, content)
|
|
obj.arch, obj.platform, obj.os = Target.parse(content)
|
|
return obj
|
|
|
|
@staticmethod
|
|
def parse(s):
|
|
elem = s.split("-")
|
|
if len(elem) == 2:
|
|
arch, platform, os = elem[0], "n/a", elem[1]
|
|
else:
|
|
arch, platform, os = elem[0], elem[1], "-".join(elem[2:])
|
|
if os == "ios":
|
|
os = "apple IOS"
|
|
if os == "darwin":
|
|
os = "apple MacOS"
|
|
return (arch, platform, os)
|
|
|
|
@staticmethod
|
|
def get_heading():
|
|
return ["OS", "ARCH"]
|
|
|
|
def get_row_heading(self):
|
|
return [self.os, self.arch]
|
|
|
|
def requires_nightly(self):
|
|
return "redox" in self
|
|
|
|
# Perform the 'it-compiles' check
|
|
def check(self, binary):
|
|
if self.requires_nightly():
|
|
args = [
|
|
"cargo",
|
|
"+nightly",
|
|
"check",
|
|
"-p",
|
|
f"uu_{binary}",
|
|
"--bin",
|
|
binary,
|
|
f"--target={self}",
|
|
]
|
|
else:
|
|
args = [
|
|
"cargo",
|
|
"check",
|
|
"-p",
|
|
f"uu_{binary}",
|
|
"--bin",
|
|
binary,
|
|
f"--target={self}",
|
|
]
|
|
|
|
res = subprocess.run(args, capture_output=True)
|
|
return res.returncode
|
|
|
|
# Validate that the dependencies for running this target are met
|
|
def is_installed(self):
|
|
# check IOS sdk is installed, raise exception otherwise
|
|
if "ios" in self:
|
|
res = subprocess.run(["which", "xcrun"], capture_output=True)
|
|
if len(res.stdout) == 0:
|
|
raise Exception(
|
|
"Error: IOS sdk does not seem to be installed. Please do that manually"
|
|
)
|
|
if not self.requires_nightly():
|
|
# check std toolchains are installed
|
|
toolchains = subprocess.run(
|
|
["rustup", "target", "list"], capture_output=True
|
|
)
|
|
toolchains = toolchains.stdout.decode("utf-8").split("\n")
|
|
if "installed" not in next(filter(lambda x: self in x, toolchains)):
|
|
raise Exception(
|
|
f"Error: the {self} target is not installed. Please do that manually"
|
|
)
|
|
else:
|
|
# check nightly toolchains are installed
|
|
toolchains = subprocess.run(
|
|
["rustup", "+nightly", "target", "list"], capture_output=True
|
|
)
|
|
toolchains = toolchains.stdout.decode("utf-8").split("\n")
|
|
if "installed" not in next(filter(lambda x: self in x, toolchains)):
|
|
raise Exception(
|
|
f"Error: the {self} nightly target is not installed. Please do that manually"
|
|
)
|
|
return True
|
|
|
|
|
|
def install_targets():
|
|
cmd = ["rustup", "target", "add"] + TARGETS
|
|
print(" ".join(cmd))
|
|
ret = subprocess.run(cmd)
|
|
assert ret.returncode == 0
|
|
|
|
|
|
def get_all_bins():
|
|
bins = map(lambda x: x.name, BINS_PATH.iterdir())
|
|
return sorted(list(bins))
|
|
|
|
|
|
def get_targets(selection):
|
|
if "all" in selection:
|
|
return list(map(Target, TARGETS))
|
|
else:
|
|
# preserve the same order as in TARGETS
|
|
return list(map(Target, filter(lambda x: x in selection, TARGETS)))
|
|
|
|
|
|
def test_helper(tup):
|
|
bin, target = tup
|
|
retcode = target.check(bin)
|
|
return (target, bin, retcode)
|
|
|
|
|
|
def test_all_targets(targets, bins):
|
|
pool = multiprocessing.Pool()
|
|
inputs = [(b, t) for b in bins for t in targets]
|
|
|
|
outputs = list(tqdm(pool.imap(test_helper, inputs), total=len(inputs)))
|
|
|
|
table = defaultdict(dict)
|
|
for (t, b, r) in outputs:
|
|
table[t][b] = r
|
|
return table
|
|
|
|
|
|
def save_csv(file, table):
|
|
targets = get_targets(table.keys()) # preserve order in CSV
|
|
bins = list(list(table.values())[0].keys())
|
|
with open(file, "w") as csvfile:
|
|
header = ["target"] + bins
|
|
writer = csv.DictWriter(csvfile, fieldnames=header)
|
|
writer.writeheader()
|
|
for t in targets:
|
|
d = {"target": t}
|
|
d.update(table[t])
|
|
writer.writerow(d)
|
|
|
|
|
|
def load_csv(file):
|
|
table = {}
|
|
cols = []
|
|
rows = []
|
|
with open(file, "r") as csvfile:
|
|
reader = csv.DictReader(csvfile)
|
|
cols = list(filter(lambda x: x != "target", reader.fieldnames))
|
|
for row in reader:
|
|
t = Target(row["target"])
|
|
rows += [t]
|
|
del row["target"]
|
|
table[t] = dict([k, int(v)] for k, v in row.items())
|
|
return (table, rows, cols)
|
|
|
|
|
|
def merge_tables(old, new):
|
|
from copy import deepcopy
|
|
|
|
tmp = deepcopy(old)
|
|
tmp.update(deepcopy(new))
|
|
return tmp
|
|
|
|
|
|
def render_md(fd, table, headings: str, row_headings: Target):
|
|
def print_row(lst, lens=[]):
|
|
lens = lens + [0] * (len(lst) - len(lens))
|
|
for e, lmd in zip(lst, lens):
|
|
fmt = "|{}" if lmd == 0 else "|{:>%s}" % len(header[0])
|
|
fd.write(fmt.format(e))
|
|
fd.write("|\n")
|
|
|
|
def cell_render(target, bin):
|
|
return "y" if table[target][bin] == 0 else " "
|
|
|
|
# add some 'hard' padding to specific columns
|
|
lens = [
|
|
max(map(lambda x: len(x.os), row_headings)) + 2,
|
|
max(map(lambda x: len(x.arch), row_headings)) + 2,
|
|
]
|
|
header = Target.get_heading()
|
|
header[0] = ("{:#^%d}" % lens[0]).format(header[0])
|
|
header[1] = ("{:#^%d}" % lens[1]).format(header[1])
|
|
|
|
header += headings
|
|
print_row(header)
|
|
lines = list(map(lambda x: "-" * len(x), header))
|
|
print_row(lines)
|
|
|
|
for t in row_headings:
|
|
row = list(map(lambda b: cell_render(t, b), headings))
|
|
row = t.get_row_heading() + row
|
|
print_row(row)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
# create the top-level parser
|
|
parser = argparse.ArgumentParser(prog="compiles_table.py")
|
|
subparsers = parser.add_subparsers(
|
|
help="sub-command to execute", required=True, dest="cmd"
|
|
)
|
|
# create the parser for the "check" command
|
|
parser_a = subparsers.add_parser(
|
|
"check", help="run cargo check on specified targets and update csv cache"
|
|
)
|
|
parser_a.add_argument(
|
|
"targets",
|
|
metavar="TARGET",
|
|
type=str,
|
|
nargs="+",
|
|
choices=["all"] + TARGETS,
|
|
help="target-triple to check, as shown by 'rustup target list'",
|
|
)
|
|
# create the parser for the "render" command
|
|
parser_b = subparsers.add_parser("render", help="print a markdown table to stdout")
|
|
parser_b.add_argument(
|
|
"--equidistant",
|
|
action="store_true",
|
|
help="NOT IMPLEMENTED: render each column with an equal width (in plaintext)",
|
|
)
|
|
args = parser.parse_args()
|
|
|
|
if args.cmd == "render":
|
|
table, targets, bins = load_csv(CACHE_PATH)
|
|
render_md(sys.stdout, table, bins, targets)
|
|
|
|
if args.cmd == "check":
|
|
targets = get_targets(args.targets)
|
|
bins = get_all_bins()
|
|
|
|
assert all(map(Target.is_installed, targets))
|
|
table = test_all_targets(targets, bins)
|
|
|
|
prev_table, _, _ = load_csv(CACHE_PATH)
|
|
new_table = merge_tables(prev_table, table)
|
|
save_csv(CACHE_PATH, new_table)
|