Skip to content

Community setup - #5

Merged
phantoslayer21 merged 6 commits into
mainfrom
community-setup
Nov 13, 2026
Merged

phantoslayer21 merged 6 commits into
mainfrom
community-setup

Conversation

@phantoslayer21

@phantoslayer21 phantoslayer21 commented Nov 13, 2026 •

Copy link
Copy Markdown
Owner

Sets WrongGPU up as a community project: CI on every pull request, a written contribution flow, a roadmap, and the
game-compatibility page back in the repository.

What's in it

  • CI (.github/workflows/ci.yml, Windows runner, actions pinned to released versions; no GPU jobs). The
    scripts it runs (tools/ci/) can also be run locally. Every multi-line step stops at its first failing command.
    • Licence and line endings: REUSE lint (reuse==5.0.2) and tools/ci/check_line_endings.py (files stored
      exactly as committed: .gitattributes is * -text, no CR LF).
    • CPU build and tests: converter, shared settings, FSR front end (CPU tests), launcher core (wg-core,
      launcher-core, coexist-core, source size, wording), the Ray Reconstruction plugin built with --no-tests and its
      CPU test executables (tools/ci/run_proxy_core_tests.py), and the CPU Python test set
      (tools/ci/run_python_tests.py: every clean/**/test_*.py except the modules that need a GPU device).
  • CONTRIBUTING.md for pull requests: how a pull request gets merged (CI, then the maintainer's clean-room scan,
    review by someone who did not write it, a GPU test on an RX 9000-series card when the change touches kernels, the
    runtime, the front end or the Ray Reconstruction plugin, then the owner merges), the clean-room statement, sign-off
    (DCO 1.1), and a rule for numbers in documents.
  • Pull-request template with the clean-room statement and the sign-off as checkboxes; CODEOWNERS
    (* @phantoslayer21).
  • ROADMAP.md: next release, Ray Reconstruction speed, RDNA 3, Vulkan / Linux. Plans, no dates.
  • COMPATIBILITY.md (from the repository's history, updated for 0.6.0) with a "How to add your game" section; the
    game compatibility issue form (label game-compatibility) and docs/TESTING-OTHER-GAMES.md back.
    Timings appear only where a measurement record exists (Borderlands 4).
  • RELEASING.md: how official releases are built and tested on real hardware, published only by the owner, tagged
    on a public commit, with the package's SHA-256 on the release page (the zip) and in SHA256SUMS.txt inside the zip
    (every file), and every figure traced to its measurement record.
  • README: the project's status as a community project, links to the welcome discussion (WrongGPU is now open source and a community project: start here #1) and the fresh-build
    issue (Build WrongGPU from a fresh clone and report where the guide breaks #2), and a Contributing section.
  • clean/rr_f/proxy/tools/build.py: a relative --out is made absolute, because the compiles run inside the
    output folder. Without it the first cl call failed with C1083.

Tested

  • CI (both checks) builds the plugin with a relative --out and passes.
  • A full BUILDING.md run from a fresh copy of the source passes every step with that fix.

Not tested

  • No GPU or game test was run. The change to build.py is path handling only.

Settings still to set (proposals; nothing was changed)

  • Branch protection on main: require the two CI checks once this pull request has shown them green (their exact
    names are in the first passing run).
  • Optionally require a review from a code owner on main.
  • Turn on private vulnerability reporting (SECURITY.md and CODE_OF_CONDUCT.md point to it).

Signed-off-by: phantoslayer21 156617979+phantoslayer21@users.noreply.github.com

- CONTRIBUTING.md for pull requests: how a pull request gets merged, what CI
  checks, what a maintainer checks by hand (clean-room scan, review, GPU test).
- Pull-request template with the clean-room statement as a checkbox; CODEOWNERS.
- ROADMAP.md, COMPATIBILITY.md (0.6.0, how to add your game), the game
  compatibility issue form and docs/TESTING-OTHER-GAMES.md.
- README: community status and a Contributing section.
- tools/ci: line-ending check, the Ray Reconstruction plugin's CPU tests and
  the CPU Python test set, for the CI workflow and for local runs.

Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
…dows runner

Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
…tep at its first failing command

The plugin's build script defaults VSDEVCMD to Visual Studio 18 Community; the runner has
Visual Studio 2022, so the step now sets VSDEVCMD from VSINSTALLDIR. Every multi-line step
also sets $PSNativeCommandUseErrorActionPreference, so a failing command fails the step at
once instead of only the step's last command counting.

Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
build.py runs its compiles inside the output folder, so a relative --out (as BUILDING.md
gives it: --out out\rr-proxy) was resolved twice and the compiler could not write its
object files. The folder is now made absolute right after the arguments are read.

Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
- CONTRIBUTING: CI's REUSE lint does not fail on a missing SPDX line (the
  `**` rule in REUSE.toml covers every file); say so, and say a missing
  header is caught in review.
- README, COMPATIBILITY: Borderlands 4 Model M is "tested with 0.4.5", not a
  0.6.0 result. The "about 1.6 ms" figure cites
  clean/installer/tests/logs/bl4-rr-only.log; the Ray Reconstruction mode and
  the WrongGPU version of that run are not recorded in the excerpt.
- RELEASING: the release page lists the zip's SHA-256 and SHA256SUMS.txt in
  the zip lists every file's, as the 0.6.0 notes say.
- ROADMAP: Hogwarts Legacy: state the observed fact; the cause is not known.
- README, PR template: the same list for when a GPU test is required as in
  CONTRIBUTING. The template links CONTRIBUTING.md absolutely and says the
  clean-room statement is made once, in the first pull request.
- tools/ci/run_python_tests.py: keep the output of a timed-out test module
  (TimeoutExpired carries bytes).
- ci.yml: cancel superseded runs on pull requests only, so every main commit
  keeps its result.
- tools/ci/run_proxy_core_tests.py: the docstring says that build.py
  --no-tests still builds and runs the WARP-based probe and hook tests.

Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
@phantoslayer21
phantoslayer21 merged commit 33a6bc0 into main Nov 13, 2026
2 checks passed
@phantoslayer21
phantoslayer21 deleted the community-setup branch November 13, 2026 15:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant