构建：从打包数据中剔除 .gitignore 等版本控制残留并加校验

collect_data_files("orisys") 会把 SDK 包内受 git 跟踪的 src/orisys/.gitignore 一并拷进 _internal\orisys；现在 spec 统一过滤版本控制与缓存文件名，build.ps1 第 6 步对我们自己发布的三个包硬失败、其它位置仅告警。验证：清理重建 exit 0，dist 内 .git* / __pycache__ / 缓存文件 0 命中，Open3D 检查通过，冒烟测试打印 "the application started" 与 "the packaged 3D scene loaded"，输出 488.7 MB。
2026-10-10 16:17:44 +08:00
parent f94b51e2de
commit c4663ef769
3 changed files with 89 additions and 11 deletions
@@ -11,7 +11,7 @@
    1. make sure that provisioning happened (and run `setup.ps1` if it plainly has not),
    2. verify the environment can import what the frozen app imports,
    3. build,
-    4. verify the artefact — config, excludes, and a real start-up of the produced EXE.
+    4. verify the artefact — config, excludes, residue, and a real start-up of the produced EXE.

  Step 2 and step 4 exist because the interesting failures are silent ones: a stale venv
  still builds (the spec's pathex finds the modules) yet loses the theme data; an excluded
@@ -177,7 +177,36 @@ if ($o3dHits) {
    Write-Host "Open3D check: none found under dist."
 }

-# ----------------------------------------------------------------- 6. smoke test
+# -------------------------------------------- 6. no source-control or cache residue
+# PyInstaller copies whatever data a package declares, and `collect_data_files("orisys")`
+# copies every non-Python file the SDK package holds — including the tracked
+# `src/orisys/.gitignore`, which is how `_internal\orisys\.gitignore` got shipped. A file
+# like that is not a crash, so nothing else in this script would notice it; this check is
+# the only thing that keeps it from coming back. It is a hard failure inside the bundles we
+# own, and a warning elsewhere, where the file may belong to a third-party wheel.
+$residueNames = '^(\.git.*|\.gitkeep|__pycache__|\.pytest_cache|\.mypy_cache|\.ruff_cache|\.scratch|\.tox|\.idea|\.vscode)$'
+$ourPackages = @("_internal\orisys", "_internal\orisys_app_shared", "_internal\orisys_studio")
+$residue = @(Get-ChildItem $OutDir -Recurse -Force -ErrorAction SilentlyContinue |
+    Where-Object { $_.Name -match $residueNames })
+$ownedResidue = @()
+foreach ($item in $residue) {
+    $relative = $item.FullName.Substring($OutDir.Length + 1)
+    foreach ($package in $ourPackages) {
+        if ($relative -like "$package\*") { $ownedResidue += $item; break }
+    }
+}
+if ($ownedResidue.Count -gt 0) {
+    Write-Host "Residue inside the bundles we own:"
+    $ownedResidue | ForEach-Object { Write-Host ("  " + $_.FullName.Substring($OutDir.Length + 1)) }
+    throw "the bundle ships $($ownedResidue.Count) source-control/cache file(s); drop them in build\studio\orisys-studio.spec and rebuild"
+} elseif ($residue.Count -gt 0) {
+    Write-Warning "Residue outside our packages ($($residue.Count) item(s)), left alone:"
+    $residue | ForEach-Object { Write-Warning ("  " + $_.FullName.Substring($OutDir.Length + 1)) }
+} else {
+    Write-Host "Residue check: no .git* / __pycache__ / cache files under dist."
+}
+
+# ----------------------------------------------------------------- 7. smoke test
 # Start the EXE the way a user would and require the application's own log to say it came up.
 # This is the step that would have caught every silent failure found in this build.
 $smokeFailed = $false
@@ -276,7 +305,7 @@ if (-not $SkipSmokeTest -and (Test-Path $Exe)) {
    }
 }

-# ------------------------------------------------------------------- 7. summary
+# ------------------------------------------------------------------- 8. summary
 if (Test-Path $OutDir) {
    $bytes = (Get-ChildItem $OutDir -Recurse -File -ErrorAction SilentlyContinue |
        Measure-Object -Property Length -Sum).Sum
@@ -71,6 +71,46 @@ def _branding_datas() -> list[tuple[str, str]]:
    return datas


+#: Source-control and tool-cache names that must never ship. ``collect_data_files`` copies
+#: *every* non-Python file a package contains, and the SDK keeps a tracked
+#: ``src/orisys/.gitignore`` — which is how ``_internal\orisys\.gitignore`` reached the
+#: bundle. The list is deliberately narrow: a real asset (``theme/*.json``, cv2's own
+#: ``config.py``) must survive, so only VCS/cache names are dropped.
+_RESIDUE_NAMES = frozenset(
+    {
+        ".git",
+        ".gitignore",
+        ".gitattributes",
+        ".gitmodules",
+        ".gitkeep",
+        "__pycache__",
+        ".pytest_cache",
+        ".mypy_cache",
+        ".ruff_cache",
+        ".scratch",
+        ".tox",
+        ".idea",
+        ".vscode",
+    }
+)
+_RESIDUE_SUFFIXES = (".pyc", ".pyo", ".orig", ".rej", "~")
+
+
+def _without_residue(datas: list[tuple[str, str]]) -> list[tuple[str, str]]:
+    """Drop VCS/cache residue from collected package data, naming what was dropped."""
+    kept: list[tuple[str, str]] = []
+    dropped: list[str] = []
+    for source, dest in datas:
+        parts = Path(source).parts
+        if any(part in _RESIDUE_NAMES for part in parts) or source.endswith(_RESIDUE_SUFFIXES):
+            dropped.append(source)
+            continue
+        kept.append((source, dest))
+    for item in dropped:
+        print(f"[spec] dropped residue from package data: {item}")
+    return kept
+
+
 block_cipher = None

 # Config is copied beside the EXE by build.ps1 (writable / editable).
@@ -92,15 +132,20 @@ hiddenimports = [
 datas += collect_data_files("orisys_app_shared", includes=["theme/*.json"])
 try:
    datas += collect_data_files("orisys")
-except Exception:
-    pass
+except Exception as exc:
+    # Tolerated so a spec run by hand still works, but never silently: a package whose data
+    # cannot be collected loses assets, and that is invisible in the built EXE.
+    print(f"[spec] WARNING: could not collect orisys package data: {exc}")
 try:
    cv2_datas, cv2_binaries, cv2_hidden = collect_all("cv2")
    datas += cv2_datas
    binaries += cv2_binaries
    hiddenimports += list(cv2_hidden)
-except Exception:
-    pass
+except Exception as exc:
+    print(f"[spec] WARNING: could not collect cv2: {exc}")
+
+# Applied last, so it covers every contributor above (including a third-party wheel).
+datas = _without_residue(datas)

 icon_path = BRANDING_DIR / "logo.ico"
 icon = str(icon_path) if icon_path.is_file() else None