makehuman/.github/workflows/ci.yml at master · SeedeXR/makehuman · GitHub

View runs

History

408 lines (367 loc) · 18.9 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

# SPDX-License-Identifier: AGPL-3.0-or-later

# Gates from memory/instruction.md section 10 and memory/test.md.

on:

push:

branches: [master]

pull_request:

workflow_dispatch:

concurrency:

group: ${{ github.workflow }}-${{ github.ref }}

cancel-in-progress: true

jobs:

format:

runs-on: macos-15

steps:

- uses: actions/checkout@v4

- name: Check formatting

run: |

CF=$(xcrun -f clang-format)

"$CF" --version

find src include tests benchmarks $ -name '*.cpp' -o -name '*.h' $ -print0 \

| xargs -0 "$CF" --style=file --dry-run --Werror

build-test:

runs-on: macos-15

strategy:

fail-fast: false

matrix:

preset: [macos-arm64-debug, macos-arm64-release, macos-arm64-asan, macos-arm64-tsan]

steps:

- uses: actions/checkout@v4

# Qt is needed or MH_HAVE_RENDER stays FALSE and the whole render/ui/app

# tree is silently skipped -- CI was green for weeks without ever

# compiling a line of it.

# `draco` because `tests/golden/test_draco.cpp` has never run anywhere

# but a developer machine: `find_package(draco QUIET)` simply failed

# here, `MH_HAVE_DRACO` stayed undefined, and 13 test cases quietly did

# not exist. It is a bottled formula, so the install is a download.

# Deliberately in the MATRIX job, so draco is covered by asan and tsan

# too. That was MEASURED rather than assumed before being turned on:

# locally `mh_tests "[draco]"` passes 96 assertions in 13 cases under

# BOTH sanitisers, with zero ThreadSanitizer warnings. The libktx note

# in memory/todo.md had advised keeping such libraries out of the

# sanitiser builds; that advice was worth checking rather than obeying.

- name: Install dependencies

run: brew install ninja assimp qt draco

- name: Configure

run: >-

cmake --preset ${{ matrix.preset }}

-DCMAKE_PREFIX_PATH="$(brew --prefix qt)"

-DMH_REQUIRE_RENDER=ON

# -Werror is on for our targets, so any new warning fails here.

- name: Build

run: cmake --build --preset ${{ matrix.preset }} -j

- name: Test

run: ctest --preset ${{ matrix.preset }} --output-on-failure

# The macos-15 runner ships Xcode 16.4, whose libc++ has NO floating-point

# from_chars/to_chars -- the float call resolves to the deleted bool overload

# and every build here failed while local builds on Apple clang 21 passed.

# foundation/Chars.h falls back to a C-locale path when the library lacks

# them. Newer runner images will make the fallback the UNUSED branch, so this

# job pins it open: without it the fallback rots silently until the next

# machine that needs it.

charconv-fallback:

runs-on: macos-15

steps:

- uses: actions/checkout@v4

- name: Install dependencies

run: brew install ninja assimp

- name: Configure with the fallback forced on

run: >

cmake -S . -B build/fallback -G Ninja

-DCMAKE_BUILD_TYPE=Debug

-DMH_HAVE_FP_CHARCONV=OFF

- name: Build

run: cmake --build build/fallback -j

- name: Test

run: ctest --test-dir build/fallback --output-on-failure

ktx2:

runs-on: macos-15

steps:

- uses: actions/checkout@v4

# basis_universal supplies the `basisu` CLI, which is the EXTERNAL

# decoder the KTX2 gates read our encoder's output back with.

# `qt` as well: `mh_ktx2_encode` links Qt6::Gui for QImage, so without it

# the encoder gates are skipped rather than run and this job proves only

# the licence boundary. The whole point of the job is to run them.

- name: Install dependencies

run: brew install ninja assimp basis_universal qt

# The only configuration in which libktx is built at all. MH_WITH_KTX2

# defaults OFF, so without this job every KTX2 gate -- including the

# licence one -- runs on a contributor's laptop and nowhere else. This

# repository has already shipped that mistake once (see the commit

# "The gate that only worked on my screen").

- name: Configure with libktx on

run: >

cmake -S . -B build/ktx2 -G Ninja

-DCMAKE_BUILD_TYPE=Debug

-DMH_WITH_KTX2=ON

- name: Build

run: cmake --build build/ktx2 -j

# The FULL suite, not just -R ktx. The KTX2 gates are the point, but

# linking a new library into mh_io can break things that have nothing

# to do with textures, and the charconv job sets the same precedent.

- name: Test

run: ctest --test-dir build/ktx2 --output-on-failure

benchmark:

runs-on: macos-15

steps:

- uses: actions/checkout@v4

- run: brew install ninja assimp

- run: cmake --preset macos-arm64-release

- run: cmake --build --preset macos-arm64-release -j

- name: Run benchmarks

run: ./build/macos-arm64-release/benchmarks/mh_bench | tee bench.txt

- uses: actions/upload-artifact@v4

with:

path: bench.txt

licence:

runs-on: macos-15

steps:

- uses: actions/checkout@v4

# Every source file must declare an SPDX identifier, and it must be one

# of the two the project uses. See LICENSING.md section 4.

- name: SPDX headers present and valid

run: |

fail=0

while IFS= read -r -d '' f; do

line=$(head -3 "$f" | grep -o 'SPDX-License-Identifier: .*' || true)

if [ -z "$line" ]; then

echo "::error file=$f::missing SPDX-License-Identifier"; fail=1; continue

id=${line#SPDX-License-Identifier: }

case "$id" in

AGPL-3.0-or-later|Apache-2.0) ;;

# BSD-3-Clause is admitted for ports of third-party BSD code that

# MakeHuman vendored -- currently only the transformations.py

# port (C. Gohlke). Those files must retain the upstream notice.

# See LICENSING.md 4.1.

BSD-3-Clause) ;;

*) echo "::error file=$f::unexpected licence '$id'"; fail=1 ;;

esac

done < <(find src include tests benchmarks $ -name '*.cpp' -o -name '*.h' $ -print0)

exit $fail

# A dependency that is not recorded in LICENSING.md section 5.1 fails CI

# (memory/instruction.md section 7), and one section 5.2 forbids fails

# with a different message, because the fix is a different one.

# This was twelve lines of inline shell that grepped the WHOLE file for

# each name. That held only while LICENSING.md never named a library it

# had turned down. The moment it recorded four refusals BY NAME, adding

# `find_package(FFTW)` passed -- matching the very row that forbids it.

# Measured before the rewrite, not theorised.

# Now a script, for the same reason the other audits are: it can be run

# and mutation-tested locally, and it is a ctest as well.

- name: Dependencies are recorded in LICENSING.md

run: python3 tools/audit_dependencies.py

# The Python-era tree is kept as a reference oracle only. If the C++

# build ever reaches into it, the port has a hidden dependency on code

# that is meant to be deleted at the end.

# Comments are stripped first. Every module's CMakeLists explains WHICH

# legacy/ file it ports -- that is provenance we want to keep, and it is

# not a build dependency. This gate matched that prose three separate

# times before the stripping was made uniform; a gate that greps for a

# forbidden string will always match the document forbidding it.

- name: Legacy tree is not part of the build

run: |

fail=0

while IFS= read -r -d '' f; do

if sed 's/#.*//' "$f" | grep -qn "legacy/"; then

echo "::error file=$f::the build references legacy/"; fail=1

done < <(find CMakeLists.txt cmake src include tests benchmarks \

$ -name CMakeLists.txt -o -name '*.cmake' $ -print0 2>/dev/null)

exit $fail

# A BSD-3-Clause file must carry the upstream copyright notice, or we

# are redistributing it in breach of the one condition BSD imposes.

- name: BSD-licensed files retain their upstream notice

run: |

fail=0

while IFS= read -r -d '' f; do

# Parse the SPDX FIELD and compare it. Grepping the file for the

# string "BSD-3-Clause" matches any prose mentioning it -- SceneIO.h

# says assimp is BSD-3-Clause in its second comment line and was

# flagged despite being Apache-2.0. Fourth time a gate here has

# matched a document instead of a violation; see

# memory/instruction.md.

line=$(head -3 "$f" | grep -o 'SPDX-License-Identifier: .*' || true)

id=${line#SPDX-License-Identifier: }

if [ "$id" = "BSD-3-Clause" ]; then

grep -q 'Copyright (c)' "$f" || {

echo "::error file=$f::BSD-3-Clause file without a copyright notice"

fail=1

}

done < <(find src include $ -name '*.cpp' -o -name '*.h' $ -print0)

exit $fail

# The licence boundary, enforced rather than documented.

# LICENSING.md 4: an Apache-2.0 module may be depended ON by an AGPL one,

# never the reverse. mh_io violated that for months -- it was stamped

# Apache-2.0 while PUBLIC-linking mh::core (AGPL) and calling

# RenderMesh::build, a port of module3d.py. Its Apache stamp bought

# nobody anything: linking io pulled in AGPL regardless.

# The SPDX gate above cannot catch this. It checks that each file

# DECLARES a licence, not that the dependency direction is legal.

- name: Apache-2.0 modules do not depend on AGPL ones

run: |

fail=0

# Every Apache-2.0 module, not a subset. `ui` was Apache-2.0 and

# ungated for months -- the rule was being enforced by review, which

# is exactly what this gate exists to replace. Keep this list equal to

# the set of modules whose CMakeLists declares Apache-2.0.

for permissive in io foundation render ui; do

# Strip comments first: this file EXPLAINS why it must not link

# mh::core, and a naive grep flags that explanation. Same trap as

# the FBX gate below.

if sed 's/#.*//' "src/$permissive/CMakeLists.txt" \

| grep -qE 'mh::core|mh_core'; then

echo "::error::src/$permissive links mh::core (AGPL); it is Apache-2.0"

fail=1

[ -d "src/$permissive" ] || continue

hits=$(grep -rln 'makehuman/core/' "src/$permissive" "include/makehuman/$permissive" \

2>/dev/null || true)

if [ -n "$hits" ]; then

echo "::error::Apache-2.0 module includes an AGPL header:"; echo "$hits"

fail=1

done

exit $fail

# <charconv> is confined to mh_foundation. Floating-point from_chars and

# to_chars do not exist in every libc++ we build on (the macos-15

# runner's Xcode 16.4 lacks them), so a direct call compiles locally and

# fails in CI. That reached CI TWICE: once because five call sites each

# had their own parser, and again because a sixth used a different helper

# name and was missed by a hand audit.

# "Don't call std::from_chars on a float" cannot be checked. This can.

- name: charconv is confined to foundation

run: |

# Match the INCLUDE, not the word: several files explain in prose why

# they must not use <charconv> directly.

hits=$(grep -rlnE '^[[:space:]]*#[[:space:]]*include[[:space:]]*<charconv>' \

src include --include='*.cpp' --include='*.h' \

| grep -v '/foundation/' || true)

if [ -n "$hits" ]; then

echo "::error::<charconv> outside mh_foundation:"; echo "$hits"

echo "Use mh::foundation::parseFloat / parseInteger / format* instead."

exit 1

exit 0

- name: Forbidden dependencies absent

run: |

# LICENSING.md section 5.2. The FBX SDK's licence is incompatible with

# AGPL redistribution; MetaHuman content is EULA-restricted.

# Match actual USE -- an include, a link, or a CMake target -- not

# the explanatory comments in SceneIO.h and LICENSING.md that say why

# the Autodesk SDK is not used.

if grep -rnE '#include *[<"]fbxsdk|-lfbxsdk|target_link_libraries.*fbxsdk|find_package *\( *FBX' \

--include=CMakeLists.txt --include='*.cmake' --include='*.cpp' \

--include='*.h' src include tests benchmarks cmake CMakeLists.txt ; then

echo "::error::a forbidden dependency is actually linked"; exit 1

exit 0

inventories:

# Pure Python -- no need for a macOS runner. Mostly stdlib; the eye-asset

# check decodes a PNG and does array maths, so this job installs Pillow and

# NumPy for that one step. They are the SAME libraries the tool uses locally

# from .venv-mh, deliberately: a second implementation of the iris mask --

# a stdlib PNG decoder written to keep the job dependency-free -- could

# disagree with the one that generates the assets, and then the gate would

# be checking the wrong thing.

runs-on: ubuntu-latest

steps:

- uses: actions/checkout@v4

# An interpreter pip can install into. The runner's system Python can be

# marked externally-managed (PEP 668), which turns `pip install` into a

# hard error -- and this job now needs two wheels.

- uses: actions/setup-python@v5

with:

python-version: '3.12'

# memory/taskviews.md carries counts taken from the reference. A stale

# number there misdirects the roadmap, so it is re-derived, not trusted.

- name: Task-view inventory matches the reference

run: python3 tools/audit_taskviews.py

# The Mixamo mapping claims every Mixamo parent relationship survives onto

# the MakeHuman skeleton. Checked, not trusted: a wrong-but-plausible

# pairing is exactly the mistake a hand-written bone table makes.

- name: Mixamo bone mapping is consistent

run: python3 tools/mixamo_mapping.py

# LICENSING.md answers every licence question this project gets asked, and

# its rows describe whole directories -- the easiest thing in the repo to

# be quietly wrong about. It was: "GLSL shaders | AGPL-3.0" did not hold

# for data/shaders/glsl/xray_*, which is MeshLab under GPL-2.0-or-later.

- name: Bundled licences are all recorded

run: python3 tools/audit_licences.py

# A .mhmat names a shader and this port ignores the name -- the viewport

# picks its model from settings. Harmless while every stem in data/ is

# accounted for; the moment an asset asks for one we do not have, it

# renders as something else with nothing to explain it.

- name: Shipped materials only ask for shading we have

run: python3 tools/audit_shading.py

# MH_DATA_DIR and friends are absolute paths into the tree that COMPILED

# the binary. A bundled copy carries its own assets, so any code reading

# the macro directly ignores them -- which is how the DMG shipped an app

# that only ran on the build machine.

- name: Runtime asset paths are resolved, not compiled in

run: python3 tools/audit_runtime_paths.py

# foundation, core, rig and io stay free of Qt, so the headless

# parameters -> mesh path can be used without a GUI stack (M10).

- name: The headless modules do not depend on Qt

run: python3 tools/audit_headless.py

# The six eye colours are GENERATED from brown_eye.png by recolouring the

# iris, and the tool rests on a measured assumption: every pixel with

# saturation > 0.35 is inside one of the two iris discs, so the sclera's

# blood vessels are untouched by construction. A future retouch of the

# source could break that silently, and the recoloured assets could drift

# from the generator that documents them. Both are checked here.

# It was not in CI at all until 2026-09-07, for want of these two wheels.

- name: Install Pillow and NumPy for the eye-asset check

run: python3 -m pip install --quiet pillow numpy

- name: Generated eye colours match their generator

run: python3 tools/make_eyes.py --check

# The helper-cage proxies (teeth, tongue) are derived from

# `data/3dobjs/base.obj`'s own `helper-*` cages, and their matcaps from

# `skinmat_caucasian.png`. Any of those sources could be retouched and the

# committed proxies would go on describing the old geometry in silence, so

# the derivation is re-run here and compared against what is committed.

- name: Generated helper-cage proxies match their generator

run: python3 tools/make_helper_proxies.py --check

# ONE product version, in /VERSION. CMake reads it, the generated

# Version.h carries it, and the two files that cannot read it -- the

# SonarQube properties among them -- are checked against it here. The

# repository had three independent declarations, one of them dead and

# contradicting the other two.

- name: Product version declarations agree

run: python3 tools/audit_version.py

# The superset skeleton is generated, not hand-edited. If someone edits it

# directly, or changes the generator without regenerating, the committed

# file and the tool disagree -- and the tool is the one the docs describe.

- name: Superset skeleton is not stale

run: python3 tools/build_mixamo_superset.py --check

# The retarget table is generated from the same proven mapping and checked

# the same way. --check re-derives it and fails on any drift; the run also

# re-validates ancestry against the 179-bone rig, so a plausible-looking

# hand edit (a fingertip moved onto the wrong finger's chain) fails here

# rather than showing up as a twisted hand in an animation.

- name: Mixamo retarget table is not stale

run: python3 tools/mixamo_mapping.py --check

# body-poseunits.json was authored against a richer, differently-named

# skeleton than MakeHuman ships: only 29 of its 61 poses resolve fully and

# 8 resolve to nothing. Gated so a future pose-unit consumer cannot be

# built on it in ignorance, and so the numbers move only upward.

- name: Body pose-unit resolvability has not regressed

run: python3 tools/audit_poseunits.py

makehuman/.github/workflows/ci.yml at master · SeedeXR/makehuman · GitHub

Expand file tree

Breadcrumbs

ci.yml

Latest commit

History

Breadcrumbs

ci.yml

File metadata and controls