[mooki000]

A build from a clone stops within seconds when --work is the clone's own work/ folder, where the development guide
puts everything made from the dump and CONTRIBUTING.md has developers put the play profile:
shutil.SameFileError: '<clone>/work/image/chunks.txt' and '<clone>/work/image/chunks.txt' are the same file.

Fixes #2331.

What changed

• nx2/make_game.py copy_writable: when the destination already exists and is the source itself
  (os.path.samefile: the same path, or a hard or symbolic link to it), nothing is copied. Before every lift make_game
  copies work/image/chunks.txt from the checkout into the build's work folder with this function, and with --work
  the checkout's own work/ both are one file. The four profile files copied just before it were already safe: they are
  copied only when they differ (filecmp.cmp).
• tests/test_copy_writable.py: also copies a file onto itself and onto a hard link of itself, and checks its bytes
  stay.

An earlier pull request, #1577, carried a guard for the same copy (comparing the two paths) beside an unrelated switch;
its author closed it the same day to finish testing. This one is the guard alone, with a test.

How it was tested

Linux (Arch, Python 3.12.16 and 3.14), no GPU involved:

• The test fails on main with shutil.SameFileError and passes with the change.
• make_game.py --romfs ... --exefs ... --work <clone>/work, the clone's work/ holding the v2.1.0 play profile (hot
  list, link order, PGO profile, image/chunks.txt), on a real dump, stopped at the lift's start (the lift's command
  replaced by a stop, so nothing was translated or compiled): main stops at work/image/chunks.txt with
  SameFileError after reading the game files; with the change it reaches @@stage lift in 4 s, and chunks.txt is
  unchanged (same SHA-256 before and after).
• python -I -B tests/run_all.py on main 49f75dc with this change (2026-11-11, load 6 to over 100): 360 passed, 4 failed,
  10 skipped; tests/test_copy_writable.py passes. Three of the failures are this machine's and fail on main as well:
  tests/test_make_game_cache.py (it fails wherever work/ holds the play profile), tests/test_wayland_libdecor_hint.py
  (Clang 23) and tests/test_updater_asset_digest.py (only a .NET 10 SDK here); each has a pull request of its own. The
  fourth, nx2/test_depth_sample_types.py, lost its compiler to a SIGKILL while the load was over 100; run alone
  afterwards it passes (13 s).

Not tested

• Windows: the same code path (fix(builder): handle the checkout's own work folder and an optional damaged-file scan #1577's author met it on Windows with --work E:\Hookshot\Hookshot\work); not run here.
• A whole build with --work <clone>/work through the lift, the compile and the link: stopped at the lift's start.

Assisted-by: Claude Opus 5.5
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com

[mooki000]

Rebased onto main b29967bf so the workflows run here: they had waited for an approval on the old head, and main's game-free tests are green again since #2380. The patch is unchanged (git range-diff: the commit is equal), and its test passes on the new base.

[Joey-Magnus]

Looks good. copy_writable is only reached for work/image/chunks.txt (nx2/make_game.py:1698), the four profile files (:1694, already skipped by filecmp.cmp when they're the same file) and the window icon (:1887, a different folder). So returning early when os.path.samefile(src, dst) (:1170-1171) is the whole fix for #2331. Comparing by file identity rather than by path, as #1577 did, also covers a symlinked or hard-linked work/ and differences in case on Windows. The early return happens before the chmod, so a read-only profile file the user owns doesn't have its mode changed either.

The extended tests/test_copy_writable.py fails on main with SameFileError and checks that the bytes survive. os.link there works on the Windows runners' NTFS temp folder, and the Windows shard that runs this test passed.

On the red checks: neither comes from this PR. Python source checks (windows-2022, part 1/2) failed only in tests/test_native_astc_texture_count.py (the fixture's scanf deprecation warning under -Werror), which #2397 has since fixed on main. appimage stopped in its System packages step on network timeouts. A rerun, or a rebase onto current main, should turn both green.

[mooki000]

Thanks for checking the file-identity cases and the Windows result. This is now merged as d6dc45bf.