You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
lombok_jar absolute Windows path is joined to extension workdir, breaking JDTLS startup #300
The language server resets the connection. The log shows:
initializing server jdtls, id 1: Server reset the connection
-- stderr --
Error opening zip file or JAR manifest missing : <USERPROFILE>/AppData/Local/Zed/extensions/work/java/D:/path/to/lombok-1.18.36.jar
The configured absolute path is concatenated onto the extension workdir instead of being used as-is.
Expected behavior
The configured absolute path is passed to JDTLS as the -javaagent argument without modification.
Root cause
In src/jdtls_server.rs, command() builds the agent argument with current_dir.join(lombok_jar_path). The extension runs as WASM, which uses POSIX path semantics — a Windows drive-qualified path like D:\... is not treated as absolute there, so join() appends it to the workdir. This does not happen on macOS/Linux because POSIX absolute paths start with /.
Also affected
java_debug_jar — src/debugger.rs uses the same current_dir.join(...) pattern in inject_plugin_into_options().
Workaround
Copy the jar into the extension workdir and reference it with a relative path:
extensions/work/java/lombok/lombok-1.18.36.jar
"lombok_jar": "lombok/lombok-1.18.36.jar"
Suggested fix
Only join the configured path with the workdir when it is relative; detect Windows drive-letter paths (e.g. a leading X:\ or X:/ prefix) and use them as-is.
Thanks for reporting, this issue has been in the code for some time but it never surfaced. I'll work on a fix for it covering both Lombok and the debugger.
Thanks for the quick fix! I've reviewed the diff — it matches my suggested approach: user-provided paths are now passed through unchanged, and the regression tests even cover Windows drive paths.
I can't run a full test on Windows right now, but one heads-up: relative-path configs (my old workaround) are now passed through as a relative -javaagent:lombok/..., which will only resolve if the JVM's cwd happens to be the extension workdir. Absolute paths are fine. Might be worth a note in the README so relative-path users know what to expect.
Otherwise this looks good to me — thanks again for the fast turnaround!
Activity
tartarughina commentedon Aug 13, 2026
Thanks for reporting, this issue has been in the code for some time but it never surfaced. I'll work on a fix for it covering both Lombok and the debugger.
tartarughina commentedon Aug 13, 2026
Fix in #301
tartarughina commentedon Aug 14, 2026
@cloud0106 could you test that on Windows?
cloud0106 commentedon Aug 15, 2026
Thanks for the quick fix! I've reviewed the diff — it matches my suggested approach: user-provided paths are now passed through unchanged, and the regression tests even cover Windows drive paths.
I can't run a full test on Windows right now, but one heads-up: relative-path configs (my old workaround) are now passed through as a relative
-javaagent:lombok/..., which will only resolve if the JVM's cwd happens to be the extension workdir. Absolute paths are fine. Might be worth a note in the README so relative-path users know what to expect.Otherwise this looks good to me — thanks again for the fast turnaround!
tartarughina commentedon Aug 15, 2026
Thanks for taking a look.
About the relative path. That is expected, the user takes ownership of where a component needs to be retrieved.
Nonetheless, I'll consider whether to make it explicitly clear in the README.
tartarughina commentedon Aug 17, 2026
The fix is included in v6.8.26 and that's now available from the marketplace