Problem
analyze records archives.name as the bare file name of the bundle, not its path relative to the scanned directory. SerializedFileParser.Parse passes the file's own directory as the root, so Path.GetRelativePath(root, file) always collapses to the file name. The schema requires archive names to be unique, so two bundles with the same file name in different folders are treated as a duplicate build and the second one is skipped:
Skipping dlc\foo\main: Duplicate archive name 'main'. Each analyzed archive must have a unique name; only a single build can be analyzed at a time.
This blocks analyzing real BuildPipeline.BuildAssetBundles output where bundle names are paths. Bundle names like dlc/<item>/main are legal and common (the AssetBundle name is a path, and the build writes the folder structure to disk). In a bundle customer build almost every item folder contained a bundle named all, so analyze over any folder with more than one item fails nearly every bundle:
Finalizing database. Successfully processed files: 279, Failed files: 16309, ...
The current work-around is to copy the bundles into a flat folder with the relative path encoded in the file name (dlc__foo__main), which is slow and awkward for multi-GB builds.
Why this is fixable
References between bundles never use the archive name. A PPtr resolves through the SerializedFile's external reference table, which names the target by its internal CAB-<hash> path. analyze already keys references on the SerializedFile name, so the archive name is only a label for the user. Recording the path relative to the scanned root (for example dlc/limitedtimeoffers/offer2/all) keeps names unique whenever the build itself is unique, and matches the name Unity gives the bundle in the AssetBundleManifest, which makes joining against manifest data easier.
Proposed change
- Pass the scanned root through to
SerializedFileParser so archives.name becomes the path relative to that root, with / separators. A file passed directly on the command line keeps its file name, as today.
- Keep the
Duplicate archive name message for genuine duplicates (two builds with the same relative layout).
- Document the change in
Documentation/analyzer-schema.md (archives.name) and Documentation/command-analyze.md. The archive column in object_view changes value for nested bundles, so bump PRAGMA user_version and add a row to the schema version table.
- Add a test with two bundles of the same file name in different sub-folders of one scanned directory, asserting both archives are recorded with distinct path-based names and that references between them still resolve.
Related: #51 introduced the duplicate-name handling.
Problem
analyzerecordsarchives.nameas the bare file name of the bundle, not its path relative to the scanned directory.SerializedFileParser.Parsepasses the file's own directory as the root, soPath.GetRelativePath(root, file)always collapses to the file name. The schema requires archive names to be unique, so two bundles with the same file name in different folders are treated as a duplicate build and the second one is skipped:This blocks analyzing real
BuildPipeline.BuildAssetBundlesoutput where bundle names are paths. Bundle names likedlc/<item>/mainare legal and common (the AssetBundle name is a path, and the build writes the folder structure to disk). In a bundle customer build almost every item folder contained a bundle namedall, soanalyzeover any folder with more than one item fails nearly every bundle:The current work-around is to copy the bundles into a flat folder with the relative path encoded in the file name (
dlc__foo__main), which is slow and awkward for multi-GB builds.Why this is fixable
References between bundles never use the archive name. A
PPtrresolves through the SerializedFile's external reference table, which names the target by its internalCAB-<hash>path.analyzealready keys references on the SerializedFile name, so the archive name is only a label for the user. Recording the path relative to the scanned root (for exampledlc/limitedtimeoffers/offer2/all) keeps names unique whenever the build itself is unique, and matches the name Unity gives the bundle in theAssetBundleManifest, which makes joining against manifest data easier.Proposed change
SerializedFileParsersoarchives.namebecomes the path relative to that root, with/separators. A file passed directly on the command line keeps its file name, as today.Duplicate archive namemessage for genuine duplicates (two builds with the same relative layout).Documentation/analyzer-schema.md(archives.name) andDocumentation/command-analyze.md. Thearchivecolumn inobject_viewchanges value for nested bundles, so bumpPRAGMA user_versionand add a row to the schema version table.Related: #51 introduced the duplicate-name handling.