[PucaVaz]

What and why

After a full Ocarina of Time build, mds_stores stayed at 100–156% CPU for more than 40 minutes with nothing else running, and mdfind found the build folder's .metal sources (measurement on #1355). Spotlight indexes each file the shader step writes, and reads the text of the .metal sources: b-<chain>/ holds a file per shader variant until the step packs them into one, 195,157 for Ocarina of Time. On macOS 27 only a folder whose name ends in .noindex keeps Spotlight out; a .metadata_never_index file inside does not, as @lunks found in #1355, which names the temporary folder tmp.noindex.

The shader folder is now shader.noindex (SHADER). A shader/ from before moves there before any step reads it: a rename keeps every file and its times, and no step's key holds the folder's path, so the split, the libraries and the compile cache are reused and nothing is translated again. The first build after the move packs the Mac app again, since pack_app.py's key names its inputs' paths. When both names exist (an older checkout made shader/ again), shader/ stays and a [NOTE] line names it. bench.py, regress.py and bug_report.py read either name.

This is independent of #1355: it is based on main and does not touch the temporary folder. @lunks will rebase #1355 on whichever lands first; a trial merge of the two branches (git merge-tree) has no conflict.

Which folders Spotlight spends its time on

MacBook Air M4, 24 GB, macOS 27.2. APFS clones of the subfolders of that Ocarina of Time build folder went into a folder Spotlight indexes (~/nxnative-spotlight-test, never the real build folder), one at a time. The figure is the CPU of mds, mds_stores and mdworker_shared above their idle level, sampled each second until the minute's mean was back at idle. mdfind found every entry of the shader, pso/, logs/ and extract/ clones afterwards. The two shader samples are the files whose names start with 0, a sixteenth.

Folder	Files	Spotlight CPU	Whole folder
shader/cache/b-<chain>/, a sixteenth	11,973	152 s, up to 772%	195,157 files, about 2,500 s
the same files deleted	11,973	under 1 s
shader/cache/cas/ (before cas.db), a sixteenth	19,839	210 s, up to 540%	311,903 files, about 3,300 s
pso/	1,162	33 s
patched_v131072/, profile/, tools/, font/, app/	45	4 s
logs/ (one log of 277 MB)	15	3 s
extract/ (a 26 GB RomFS.img)	7	2 s

About 5,800 s for the shader files of a build folder from before the one-file libraries and cache: an hour or more at 100–156%, which covers the more than 40 minutes seen. On main the libraries and cache are one file each at rest, but a shader build still writes b-<chain>/, about 2,500 s of indexing for Ocarina of Time. Every other folder together took 42 s, so they keep their names; logs/ and profile/ are also the paths docs/frontends.md gives front ends. Importers (mdimport -t): .metal, .tsv and .log go through the RichText importer, which reads their text; .json, .metallib, .mtl4a and RomFS.img get none. A game benchmark ran in another session during most of the measurement.

Checked

• Tests: CI's tool-test loop, every dm/tools/tests/test_*.py, dm/ios/tests/test_*.py and dm/gfx/tests/test_*.py with DM_GAME=010015100B514000 and DMSH_TEST_XLATE set: 154 files, all pass but test_tree_tar.py's test_packs_a_quiet_folder_and_keeps_its_attributes, which fails the same on main on this Mac (the 14 dm/gfx/tests files that compile Metal ran with Xcode 27.2's DEVELOPER_DIR, as the selected Xcode beta has no Metal toolchain). New: test_nsp_chain.py Spotlight (the move keeps contents and times, the new name wins over a leftover, a link is left alone, no work / 'shader' left in the script, bench.py reads the same name), test_nsp_work_dir.py test_shader_folder_from_before_moves_first (main moves it before any step), test_bench.py test_shader_folder; test_bug_report.py, test_regress.py and test_nsp_mac_app.py follow the name. lint_python.py with the pinned ruff 0.17.2 and pyright 1.1.414: no new finding.
• The real folder: the shader package (516 MB) and keys of that Ocarina of Time build, cloned: its pkg.outputs.key matches split_output_key before and after move_to_noindex.
• Games run: none. No build was run with this change: a full Ocarina of Time shader build takes about two hours here, so the reuse after the move is shown by the keys, not by a build.
• Devices and OS: MacBook Air M4, 24 GB, macOS 27.2.
• No chain, host or shader tool change. regress.py only finds the shader folder by either name now (its tests pass); regress.py check was not run.

Front ends

Front-ends: interface: the build folder's shader folder is shader.noindex/ instead of shader/ (cache/, ios/, vk/, gl/, pkg/ and the keys inside it, and the Mac run command's DM_MSL_DIR); a shader/ from before moves there at the next build.

Docs

• The finding is in its docs/reference/<subsystem>.md entry (Spotlight indexes the files a shader build writes), and python3 dm/tools/docs_symptoms.py was run (the symptom index). docs/development.md, the bug report form and the two game notes that named shader/ follow the name.
• A chain tool changed: not applicable
• No game files, keys or data extracted from a game in the change or the attachments

Drafted with Claude; reviewed by me.

🤖 Generated with Claude Code

[PucaVaz]

Closing: since 9c2fb0c a build no longer keeps the per-shader files this measured (195,157 in b-<chain>/, which kept mds_stores at 100–156% CPU for over 40 minutes after a full Ocarina of Time build); what's left is short-lived and unmeasured here.