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
Fails to start in a JDK 21 environment, but runs correctly with JDK 25.
When switching to a manually installed version of jdtls (1.61.0), it runs successfully with both JDK 21 and JDK 25.
Zed shows cannot read LSP message headers / Server reset the connection on initialize.
The WARNING: jdk.incubator.vector line is benign, not the cause.
The proxy log is the real tell:
JDTLS process spawned (pid=57717)
HTTP server listening on 127.0.0.1:63814
JDTLS process exited: Ok(ExitStatus(unix_wait_status(3328)))
unix_wait_status(3328) = exit code 13 = jdtls’s standard “stdin closed” exit.
Reproduced outside Zed with the extension’s exact command: sending initialize immediately → handshake succeeds; keeping stdin open but delaying initialize ~4s (mimicking Zed’s normal latency) → identical unix_wait_status(3328); launching jdtls with stdin < /dev/null → exit 13 directly. So the proxy spawns jdtls but doesn’t keep its stdin open while waiting for the first message — jdtls gets EOF and exits before the handshake, and Zed loses this race on every launch.
Ruled out (verified by hand): Java 21, jdtls 1.63.0, the Gradle project, lombok (lombok_support: false doesn’t help), loopback bind, quarantine/signing, PATH/JAVA_HOME, TMPDIR, permissions.
The issue lies with the latest update 6.8.29 and JDK 21, JDK25 works as expected. While working on a proper fix the option available to address the issue is:
manually rolling back to 6.8.28 not the issue
Point the option java_home to JDK25
As @luca-666 identified, providing a different version of jdtls lower than 1.63 1.62 works as well regardless of the JDK version.
Starting from v6.8.30 the extension requires Java25 aligning with what jdtls expects. Closing the issue as now the dependency is well reported both in the error and in the README
9. Starting from v6.8.30 the extension requires Java25 aligning with what jdtls expects. Closing the issue as now the dependency is well reported both in the error and in the README
Thanks for the reply!
I have a suggestion regarding the JDK version: Could the extension automatically detect installed JDK 25+? This would save users from having to manually modify the Java path or default version.
@luca-666 that's a nice suggestion, although the extension runs on a WASM sandbox. It has permissions only for its installation directory.
Going through the installed JDK versions would require an external script that the extension retrieves during installation. Supporting such a script across multiple platforms would either reintroduce Python, which we worked previously on removing, or require different versions based on shell types, OSes, etc.
Users can already flip the flag jdk_auto_download to true, and a version of OpenJDK25 is going to be installed for them.
We could have a TS script, rather than a Python one, and leverage Zed's vended Node runtime actually. But using an external script to overcome limitations purposely set is not the right approach either, @MrSubidubi to confirm this sentiment.
Yep, can confirm this: While I acknowledge this is not optimal, we generally oppose such workarounds, since we do not want arbitrary scripts to run and probe the users environment. The extension can search in PATH or something similar if it wants (and can also use the Command API to do some probing that can also be blocked), but having external scripts is not something that want nor want to endorse.
Activity
tartarughina commentedon Oct 23, 2026
@luca-666 could you please share any log available?
quangth14 commentedon Oct 23, 2026
java-lsp-proxy lets jdtls’s stdin hit EOF before initialize → jdtls exits 13
Env: macOS 15.8 (arm64), java extension v6.8.29 (proxy bin/v6.8.29/java-lsp-proxy), jdtls 1.63.0, OpenJDK 21.0.11.
Zed shows cannot read LSP message headers / Server reset the connection on initialize.
The WARNING: jdk.incubator.vector line is benign, not the cause.
The proxy log is the real tell:
JDTLS process spawned (pid=57717)
HTTP server listening on 127.0.0.1:63814
JDTLS process exited: Ok(ExitStatus(unix_wait_status(3328)))
unix_wait_status(3328) = exit code 13 = jdtls’s standard “stdin closed” exit.
Reproduced outside Zed with the extension’s exact command: sending initialize immediately → handshake succeeds; keeping stdin open but delaying initialize ~4s (mimicking Zed’s normal latency) → identical unix_wait_status(3328); launching jdtls with stdin < /dev/null → exit 13 directly. So the proxy spawns jdtls but doesn’t keep its stdin open while waiting for the first message — jdtls gets EOF and exits before the handshake, and Zed loses this race on every launch.
Ruled out (verified by hand): Java 21, jdtls 1.63.0, the Gradle project, lombok (lombok_support: false doesn’t help), loopback bind, quarantine/signing, PATH/JAVA_HOME, TMPDIR, permissions.
tartarughina commentedon Oct 23, 2026
@quangth14 thanks for the report.
The issue lies with the latest update6.8.29and JDK 21, JDK25 works as expected.While working on a proper fix theoption available to address the issue is:manually rolling back tonot the issue6.8.28As @luca-666 identified, providing a different version of jdtls lower than
1.631.62 works as well regardless of the JDK version.tartarughina commentedon Oct 23, 2026
I did some testing removing from the equation Zed and this extension.
Starting from JDTLS 1.62, Java25 is required.
Update min Java version required from 21 to 25 (#320)
tartarughina commentedon Oct 27, 2026
Starting from v6.8.30 the extension requires Java25 aligning with what jdtls expects. Closing the issue as now the dependency is well reported both in the error and in the README
luca-666 commentedon Oct 28, 2026
Thanks for the reply!
I have a suggestion regarding the JDK version: Could the extension automatically detect installed JDK 25+? This would save users from having to manually modify the Java path or default version.
tartarughina commentedon Oct 28, 2026
@luca-666 that's a nice suggestion, although the extension runs on a WASM sandbox. It has permissions only for its installation directory.
Going through the installed JDK versions would require an external script that the extension retrieves during installation. Supporting such a script across multiple platforms would either reintroduce Python, which we worked previously on removing, or require different versions based on shell types, OSes, etc.
Users can already flip the flag
jdk_auto_downloadtotrue, and a version of OpenJDK25 is going to be installed for them.We could have a TS script, rather than a Python one, and leverage Zed's vended Node runtime actually. But using an external script to overcome limitations purposely set is not the right approach either, @MrSubidubi to confirm this sentiment.
MrSubidubi commentedon Oct 28, 2026
Yep, can confirm this: While I acknowledge this is not optimal, we generally oppose such workarounds, since we do not want arbitrary scripts to run and probe the users environment. The extension can search in PATH or something similar if it wants (and can also use the Command API to do some probing that can also be blocked), but having external scripts is not something that want nor want to endorse.