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
Hey @nikosszzz@mylvghb@kenzson, I've published #161. Before merging it could you give it a try and letting me know if the issue is fixed for you? I've tested it on my Windows machine but I'd love to get also your feedback.
@tartarughina I cloned your repository and installed the development extension, but the problem still persists.
log: 2025-12-23T10:14:48+08:00 ERROR [project::lsp_store] Failed to start language server "jdtls": from extension "Java" version 6.8.3: 文件名、目录名或卷标语法不正确。 (os error 123)
Just return Ok(path_str) directly instead of calling format_path_for_os, and it works fine. If quotes are redefined to mean that the path begins and ends with a quote, then such a path is invalid.
@mylvghb Since the fix works for me, could you tell me what your path looks like?
The original change that introduced this regression was made for a reason, as quotes were required to ensure the jdtls arguments were recognized properly.
Knowing how your path is structured would help me better understand why this behavior occurs on Windows and how we can make it work reliably for everyone.
After diving deeper into the issue and creating an ad hoc user with a space in the name, I've identified that the issues we are facing are all coming from the spawn command in the proxy. The spawn for Windows is done in shell mode as historically we were relying on jdtls.bat to launch the language server. @playdohface removed this dependency, and we no longer are using that.
If you need to read more, check out #84 and #86. Removing the shell: true from the proxy works like a charm. I'll update the PR, and if you can further verify that everything works as intended, I'll merge it.
Slightly off topic, but maybe it would be worth adding a simple Actions workflow that runs a test of the "download and start JDTLS" pipeline on each OS? Could maybe catch these kinds of issues before they happen. 🙂
That is currently not possible because the infrastructure to add such tests does not exist on the side of Zed.
I don't know Rust, but is it not possible to create like a unit test that just invokes that part of the code, without requiring all of Zed? The functions that download/launch JDTLS should not have to depend on any Zed-specific constructs.
@Komposten It actually does depend on Zed - this all runs within a WASM-environment of which Zed is the host and uses APIs that are provided by Zed. E.g. the downloading happens because the extension asks Zed to download said file. Hence, in order to test this properly, one has to have access to such environment hosted by Zed.
Activity
tartarughina commentedon Dec 22, 2025
Could you please update to the latest version 6.8.3 to confirm you are still incurring into this issue?
kenzson commentedon Dec 22, 2025
Language server jdtls:
from extension "Java" version 6.8.3: 文件名、目录名或卷标语法不正确。 (os error 123)
tartarughina commentedon Dec 22, 2025
@kenzson are you too on Windows?
@mylvghb could you please check what's not available in the java extension work directory that the extension assumes should be instead?
mylvghb commentedon Dec 22, 2025
@tartarughina The same project runs without any issues using v6.8.1.
mylvghb commentedon Dec 22, 2025
@tartarughina This issue might be caused by commit a540253.
tartarughina commentedon Dec 22, 2025
@mylvghb thanks for sharing that. Very likely the issue is due to that. Quoting may have specific edge cases where escaping the space is preferred
mylvghb commentedon Dec 22, 2025
Waiting for the fix.
nikosszzz commentedon Dec 22, 2025
I am also getting this error. Windows 11 25H2. v6.8.1 works.
tartarughina commentedon Dec 22, 2025
Hey @nikosszzz @mylvghb @kenzson, I've published #161. Before merging it could you give it a try and letting me know if the issue is fixed for you? I've tested it on my Windows machine but I'd love to get also your feedback.
mylvghb commentedon Dec 23, 2025
@tartarughina I cloned your repository and installed the development extension, but the problem still persists.
log: 2025-12-23T10:14:48+08:00 ERROR [project::lsp_store] Failed to start language server "jdtls": from extension "Java" version 6.8.3: 文件名、目录名或卷标语法不正确。 (os error 123)
mylvghb commentedon Dec 23, 2025
@tartarughina
Just return Ok(path_str) directly instead of calling format_path_for_os, and it works fine. If quotes are redefined to mean that the path begins and ends with a quote, then such a path is invalid.
tartarughina commentedon Dec 23, 2025
@mylvghb Since the fix works for me, could you tell me what your path looks like?
The original change that introduced this regression was made for a reason, as quotes were required to ensure the jdtls arguments were recognized properly.
Knowing how your path is structured would help me better understand why this behavior occurs on Windows and how we can make it work reliably for everyone.
tartarughina commentedon Dec 23, 2025
After diving deeper into the issue and creating an ad hoc user with a space in the name, I've identified that the issues we are facing are all coming from the spawn command in the proxy. The spawn for Windows is done in shell mode as historically we were relying on
jdtls.batto launch the language server. @playdohface removed this dependency, and we no longer are using that.If you need to read more, check out #84 and #86. Removing the shell: true from the proxy works like a charm. I'll update the PR, and if you can further verify that everything works as intended, I'll merge it.
mylvghb commentedon Dec 24, 2025
@tartarughina I tested the branch fix-win-path-quoting, and it works fine.
nikosszzz commentedon Dec 24, 2025
That branch also works good for me.
tartarughina commentedon Dec 24, 2025
Thanks @mylvghb @nikosszzz for testing.
I still need to cover an edge case, I think, but once tested I'll merge the PR.
Komposten commentedon Dec 25, 2025
Slightly off topic, but maybe it would be worth adding a simple Actions workflow that runs a test of the "download and start JDTLS" pipeline on each OS? Could maybe catch these kinds of issues before they happen. 🙂
MrSubidubi commentedon Dec 26, 2025
@Komposten That is currently not possible because the infrastructure to add such tests does not exist on the side of Zed.
Komposten commentedon Dec 26, 2025
I don't know Rust, but is it not possible to create like a unit test that just invokes that part of the code, without requiring all of Zed? The functions that download/launch JDTLS should not have to depend on any Zed-specific constructs.
tartarughina commentedon Jan 3, 2026
Closing this issue as the error has been fixed for a while now.
MrSubidubi commentedon Jan 3, 2026
@Komposten It actually does depend on Zed - this all runs within a WASM-environment of which Zed is the host and uses APIs that are provided by Zed. E.g. the downloading happens because the extension asks Zed to download said file. Hence, in order to test this properly, one has to have access to such environment hosted by Zed.
Komposten commentedon Jan 4, 2026
@MrSubidubi Ah, okay. That's unfortunate.☹️