feat(symbols): upload Java/Kotlin sources with R8 mappings (--include-sources)

[abelonogov-ld]

Extends --include-sources to --type android, so a retraced Android frame can show the code around it instead of only class/method/line. Stacks on #757 (Apple sources).

_sym/android/id/<symbolsID>/mapping.txt         # existing
_sym/android/id/<symbolsID>/sources.srcbundle   # new, same lane

Why Android can't reuse the Apple approach

A dSYM's DWARF names every source file it covers, so ldcli never has to search for anything. An R8 mapping records no paths at all — only (class, method, line). So Android has to scan for sources and choose a key the backend can reconstruct from a retraced class name.

That key is the package-relative path (com/example/MainActivity.kt), taken from each file's own package declaration rather than from its directory, because Kotlin does not require the two to agree. The backend then rebuilds it from the retraced class: package as the directory, top-level class as the file name.

--source-path

New flag saying where to scan, defaulting to the current directory — --path points at the mapping.txt output directory, which holds no sources.

Because the default can be a whole repo root, the scan skips build/ (R8's own generated code), .gradle/, .git/, .idea/ and node_modules/, plus files over 2 MiB and bundles over 64 MiB. Pointing --source-path at app/src/main is faster and avoids bundling test or sample code.

Safety and opt-in

Off by default, because it ships your source to LaunchDarkly. A scan that finds nothing prints a note and uploads the mapping alone, so the flag can't break an upload that would otherwise succeed.

Testing

./cmd/symbols/... passes. New tests cover package extraction across the shapes real files take (Java semicolons, Kotlin without, license headers, @file: annotations, trailing comments, default package, and import before any package), keying by declared package rather than directory layout, skipping build output and node_modules, oversize files, an empty scan yielding no bundle rather than an error, and rejecting a non-directory source path.

Made with Cursor

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit cc54a1c. Configure here.}

[cursor]

License blocks hide package below

High Severity

javaPackageOf treats any non-package line that is not skipped as proof the file has no package and returns immediately. Text inside multi-line block comments (without a leading * on each line) and lines that start with /* but also contain a package clause on the same line never reach the real package statement, so bundle keys fall back to bare filenames and no longer match what the backend builds from retraced class names.

 

^{Reviewed by Cursor Bugbot for commit cc54a1c. Configure here.}

[cursor]

Duplicate keys inflate bundle budget

Low Severity

When two scanned files produce the same bundle key, srcbundle.Builder.Add keeps the first entry but total still increases by both file sizes. Later files can be dropped because the walk thinks the bundle cap is reached even though only one copy of the duplicate key is stored.

 

^{Reviewed by Cursor Bugbot for commit cc54a1c. Configure here.}