cpptools crash - The language server crashed. Restarting...

[Colengms]

We're very interested in fixing crashes. Information that could help us fix crashes include:

• Crash stacks.
• Specific guidance on how to repro a crash.
• Isolated examples that reproduce a crash and can be shared with us directly.

If you're experiencing a crash, any of the above information could be extremely useful to us to investigate and fix it. Please consider opening a new issue in this repo. Or, add a comment on this issue.

I'm pinning this issue, to make this more obvious to users who might be browsing our repo due to experiencing a crash.

[H-G-Hristov]

Just to remind that on macOS: #9810

[sean-mcmanus]

@H-G-Hristov Yes, but that issue mentions a work around via running lldb -p <processId>, then continue (assume it gets into a stopped state), and then repro the crash, and use bt to get the full stack of the crashing thread (maybe bt all if bt isn't sufficient for some reason).

[sean-mcmanus]

I've reproed multiple crashes. I'm tracking it with #10636 .

We may not need this issue anymore.

[Colengms]

Reopening. The intent of this issue was to be pinned so users experiencing a crash can find the issue and easily find instructions. I'd like to keep this open and pinned for now.

[sean-mcmanus]

@Colengms Yeah, that seems fine, but user should know that we don't currently have a widespread crashing issue (as of https://github.com/microsoft/vscode-cpptools/releases/tag/v1.14.5 or later).

[OnielN14]

I also encountered similar issue.
OS: Windows 11
VS Code version : 1.81.0
C/C++ Extension version: 1.16.3

here is what I found when I check the log in developer tools :

ERR [Extension Host] (node:20304) [DEP0005] DeprecationWarning: Buffer() is deprecated due to security and usability issues. Please use the Buffer.alloc(), Buffer.allocUnsafe(), or Buffer.from() methods instead.
(Use `Code --trace-deprecation ...` to show where the warning was created)

[OnielN14]

I also encountered similar issue. OS: Windows 11 VS Code version : 1.81.0 C/C++ Extension version: 1.16.3

here is what I found when I check the log in developer tools :

ERR [Extension Host] (node:20304) [DEP0005] DeprecationWarning: Buffer() is deprecated due to security and usability issues. Please use the Buffer.alloc(), Buffer.allocUnsafe(), or Buffer.from() methods instead.
(Use `Code --trace-deprecation ...` to show where the warning was created)

I just try to manually install the extension into fresh VSCodium v1.81.0.. the extension works

[chris0306]

大佬，这个问题怎么破？

[doxxx]

Working with a large Visual Studio solution of ~70 C# projects and one C++/CMake project. Also had the same solution open in Visual Studio at the same time. Performed a Rebuild in Visual Studio and then came back to 10000+ duplicate errors in VS Code like this:

[Error - 9:24:53 AM] Sending notification cpptools/fileDeleted failed.
Error: Cannot call write after a stream was destroyed
	at new NodeError (node:internal/errors:387:5)
	at _write (node:internal/streams/writable:323:11)
	at Writable.write (node:internal/streams/writable:336:10)
	at c:\Users\gordon.tyler\.vscode\extensions\ms-vscode.cpptools-1.16.3-win32-x64\dist\main.js:77222:29
	at new Promise (<anonymous>)
	at WritableStreamWrapper.write (c:\Users\gordon.tyler\.vscode\extensions\ms-vscode.cpptools-1.16.3-win32-x64\dist\main.js:77212:16)
	at StreamMessageWriter.doWrite (c:\Users\gordon.tyler\.vscode\extensions\ms-vscode.cpptools-1.16.3-win32-x64\dist\main.js:76424:33)
	at c:\Users\gordon.tyler\.vscode\extensions\ms-vscode.cpptools-1.16.3-win32-x64\dist\main.js:76415:29
	at runNextTicks (node:internal/process/task_queues:61:5)
	at process.processImmediate (node:internal/timers:437:9)

[sean-mcmanus]

@OnielN14 @chris0306 @doxxx That info is insufficient. We more info, particular the info that's described in the description for this issue, such as a call stack of cpptools or a repro.

[sean-mcmanus]

@OnielN14 @chris0306 @doxxx Also, 1.17.1 and 1.17.2 of our extension has some crash fixes, so that may be worth trying.

[vineetsoni]

I start to have the same issue since the extension auto-updated to 1.17.3 today. I'm running AlmaLinux 8.8 with kernel 4.18.0-477.15.1.el8_8.x86_64. The VSCode version is 1.81

Not sure if this is any useful, but the output message is:

[Error - 2:14:08 PM] The language server crashed. Restarting...
[Error - 2:14:08 PM] Sending document notification textDocument/didOpen failed
Error: Attempting to use languageClient before initialized
	at DefaultClient.get languageClient [as languageClient] (/u/vinson3z/.vscode-server/extensions/ms-vscode.cpptools-1.17.3-linux-x64/dist/main.js:47283:19)
	at DefaultClient.sendDidOpen (/u/vinson3z/.vscode-server/extensions/ms-vscode.cpptools-1.17.3-linux-x64/dist/main.js:48450:20)
	at DefaultClient.takeOwnership (/u/vinson3z/.vscode-server/extensions/ms-vscode.cpptools-1.17.3-linux-x64/dist/main.js:48439:20)
	at processDelayedDidOpen (/u/vinson3z/.vscode-server/extensions/ms-vscode.cpptools-1.17.3-linux-x64/dist/main.js:53325:30)
	at runMicrotasks (<anonymous>)
	at processTicksAndRejections (node:internal/process/task_queues:96:5)
	at async Function.dispatch (/u/vinson3z/.vscode-server/extensions/ms-vscode.cpptools-1.17.3-linux-x64/dist/main.js:48476:44)

Is there any known workaround for this problem?

Fixed the problem by downgrading to 1.16.3. Downgrading from 1.17.3 to 1.17.2 didn't work.

PS: This happens only when I try to use VSCode with a very large project. For a medium-sized project, it works fine.

[bobbrow]

@vineetsoni it looks like the extension isn't recovering from a crash properly. There were some code changes in this area recently, so thank you for reporting it. I'll move your report into a new issue.

[borjamunozf]

Sadly, this is a showstopper for a full team of developers in our company. They have transitioned from QtCreator to VScode for C++ and the experience has been really bad for production grade projects.

I have tried to attach to the C++ extension, but it does not show any crash or stops anywhere.

Tested with:

• VScode 1.84.2
• Cpp 1.18.5

I also removed .vscode-server && .vscode folder, installed again extensions.

[rganascim]

I encountered an escalation of issues when the system build was located within the source code folder, particularly when compiling gRPC libraries. By relocating the build ecosystem outside of the folder, the problems had diminished.

You could try it.

*I don't have more troubles with C++ extension

[bobbrow]

@sean-mcmanus I'm supportive of a 1.18.6 release containing that patch if we can confirm that's the cause of these recent problems.

@borjamunozf would you be willing to try 1.19.1 and see if that reduces the crashes you're seeing?

[borjamunozf]

@sean-mcmanus I'm supportive of a 1.18.6 release containing that patch if we can confirm that's the cause of these recent problems.

@borjamunozf would you be willing to try 1.19.1 and see if that reduces the crashes you're seeing?

Yes, absolutely!

[sean-mcmanus]

@borjamunozf This "issue" is just a generic issue tracking crashes -- there could be many potential causes, but we need a repro or call stack of the crash. What OS are you using? What process are you attaching to?

[sean-mcmanus]

@rganascim What do you mean by "escalation of issues"? Crashes? I don't repro a crash in that scenario. Putting your build folder in the workspace folder will trigger didChange messages that will cause IntelliSense updates, so it's not recommended to put your build folder in the workspace folder (for performance reasons).

@bobbrow We haven't identified the source of 1.18.x crashes yet.

[borjamunozf]

@borjamunozf This "issue" is just a generic issue tracking crashes -- there could be many potential causes, but we need a repro or call stack of the crash. What OS are you using? What process are you attaching to?

Snippet of Developer Tools console after crash:

I cannot reproduce the crash with the Debugger. These are the processes I can see; if i attach to the first one it shows this in Debug Console:

New LWP 228881]
[New LWP 228882]
[New LWP 228883]
[New LWP 233149]
0x00007f09b7d500a9 in __stdio_read ()

If I attach to the second one in shows:

[Detaching after fork from child process 553242]
[LWP 540987 exited]
[LWP 540986 exited]
[LWP 540985 exited]
[LWP 540983 exited]
[LWP 540982 exited]
[LWP 540981 exited]
[LWP 540980 exited]
[LWP 540984 exited]
[New process 540980]

Program terminated with signal SIGKILL, Killed.
The program no longer exists.
Debugger has disconnected from the program '/home/borjamf/.vscode-server/extensions/ms-vscode.cpptools-1.18.5-linux-x64/bin/cpptools'.

We're using WSL2 - Ubuntu 22.04: 1.2.5 WSL and 2.0.9

[borjamunozf]

Update:

• Seems that when we remove or rename the file

c_cpp_properties.json, which it contains

{
    "configurations": [
        {
            "name": "Linux",
            "includePath": [
                "${default}"
            ],
            "defines": [],
            "compilerPath": "/usr/bin/g++",
            "cStandard": "c17",
            "cppStandard": "c++14",
            "intelliSenseMode": "linux-gcc-x64",
            "compileCommands": "${workspaceFolder}/build/Debug/compile_commands.json",
            "configurationProvider": "ms-vscode.cmake-tools"
        }
    ],
    "version": 4
}

it's not failing. The easiest to reproduce the failure in our case is to execute Show Call Hierarchy: with the custom c_cpp_properties it ends usually with Crash, without it it works (or it's not failing yet)

[bobbrow]

@rganascim What do you mean by "escalation of issues"? Crashes? I don't repro a crash in that scenario. Putting your build folder in the workspace folder will trigger didChange messages that will cause IntelliSense updates, so it's not recommended to put your build folder in the workspace folder (for performance reasons).

@sean-mcmanus it's extremely common to have the build folder in the workspace (e.g. CMake Tools and others do this by default). It is also very common to have a .git folder in the workspace and we don't filter those messages either. I wanted to reduce the noise with #11493, but I thought it was decided that since we filtered these in the message handler, it was ok.

[bobbrow]

@borjamunozf were you able to try 1.19.1 yet? Your screenshots suggest you are attempting to attach to 1.18.5. Do you happen to be working on anything open source that we can clone and test locally?

[bobbrow]

@borjamunozf, also I see that you opened microsoft/vscode-cmake-tools#3469. Not sure if these issues are related but I'm guessing you have a pretty big compile_commands.json file. Would you have time this week or next week to jump on a video call and walk me through the crash(es)?

[borjamunozf]

@borjamunozf were you able to try 1.19.1 yet? Your screenshots suggest you are attempting to attach to 1.18.5. Do you happen to be working on anything open source that we can clone and test locally?

Same. Attach to process, 2-3 minutes showing lot of messages

[Detaching after fork from child process 1163080]
[Detaching after fork from child process 1163081]
[Detaching after fork from child process 1163082]
[Detaching after fork from child process 1163083]
[Detaching after fork from child process 1163084]
[Detaching after fork from child process 1163085]
[Detaching after fork from child process 1163086]
[Detaching after fork from child process 1163087]

and finally the crash.

Sadly, we work on propietary code. It's all code from my company, complex and huge projects I guess.

@borjamunozf, also I see that you opened microsoft/vscode-cmake-tools#3469. Not sure if these issues are related but I'm guessing you have a pretty big compile_commands.json file. Would you have time this week or next week to jump on a video call and walk me through the crash(es)?

It's not the same project, but yes, we have big compile_commands.json for all projects I have seen in my division. In that issue we have found that disabling the loadCompileCommands at least allows to run CMake Configure & build the project, but it's only a workaround. Just as reference, for this project the compile_commands.json is 47.56MB size., ~83000 lines

In this project, as I mentioned above, removing or renaming c_cpp_properties seems to avoid the crash, but the IntelliSense misses stuff, of course. Also if comment out the "compileCommands" only it seems to not crash, or not crash so fast, but it also does not work any Show Call Hierarchy, for example.

About the video call, that would be great. However I have to check given the nature of the code. Will let you know.

[Colengms]

Hi @borjamunozf . I see a number of clues in the information you've posted so far.

In one of your images, I see that the IntelliSense process failed to be restarted. That would imply either an issue with our restart logic, or that something is immediately in a bad state when it starts back up, such as some anomolous configuration information.

The fact you were unable to detect a process crash from the debugger suggests that perhaps the process isn't actually crashing, but being terminated by another process. We've seen symptoms like this in the past with overzealous anti-malware software killing our process. To my knowledge, we do not have an influx of users reporting these symptoms on Linux, so something specific to your environment could potentially be involved.

Your C/C++ configuration refers to both use of CMake Tools as a configurationProvider, and a compile_commands.json file. Presumably, the compile_commands.json file is one being generated by CMake. This is superfluous, as use of CMake Tools as a custom configuration provider would provide configurations for all of the same files as in that compile_commands.json file. I'd suggest removing the compileCommands field from your config. (If that addresses the issue, could you help us investigate further, in case fixing this could help other compileCommands users?)

One of the most useful sources of information for investigating issues with the C/C++ extension is its debug output. Could you enable the setting "C_Cpp.loggingLevel": "Debug", monitor the C/C++ output channel leading up to the repro, and provide us with any pertinent content? Also, could you open a new issue here to track your specific repro, as this issue is more general and crashes often do not have the same cause.

I'm hoping there is something in your log output immediately before the issue that might point to what's going on here, or at least what it was in the process of doing when the problem started to occur.

Also if comment out the "compileCommands" only it seems to not crash, or not crash so fast, but it also does not work any Show Call Hierarchy, for example.

Given that this setting seems superfluous in your case, I would not expect any difference in behavior from removing it. Though, when using a compile_commands.json file, the C/C++ extension will analyze all of those command lines on startup (mainly to build a list of 'browse' paths to populate the browse database with), whereas it would only analyze those command lines as they are needed and provided for a file to configure IntelliSense for it. (Custom configuration providers provide browse paths up front, so all command lines do not need to be parsed in that scenario.) My guess would be that there is a failure parsing one of your command lines. Debug log output may refer to which one.

[sean-mcmanus]

@bobbrow fileChanged/Created/Deleted triggers an IntelliSense update unless the file or the file's folder is filtered from a files.exclude setting. So sure you can put the cmake build folder in your workspace folder, but you should also add it to the C_Cpp.files.exclude to avoid unnecessary IntelliSense updating.

[bobbrow]

@bobbrow fileChanged/Created/Deleted triggers an IntelliSense update unless the file or the file's folder is filtered from a files.exclude setting. So sure you can put the cmake build folder in your workspace folder, but you should also add it to the C_Cpp.files.exclude to avoid unnecessary IntelliSense updating.

Can we change that behavior? If it's obviously not a C or C++ file, I don't see why we would want to trigger an IntelliSense update. We have filters for non-C/C++ files (based on file extension) in the TypeScript code so it doesn't seem wrong to do the same thing in the language server.

[Zes-MinKey-Young]

How to get a call stack? I cannot attach a debugger because the crash arises before that. If I open cpptools.exe outside VS Code, I cannot reproduce the crash. The executable does nothing.

[sean-mcmanus]

@SemenAntonov Your issue is with the clangd extension and not our extension.

[godlikeue]

Please tell me which previous version did not have this issue.

1. I install a clean Ubuntu system on a new computer.
2. sudo snap install --classic code
3. install cpptools in extensions

Open a cpp file , this issue will occur!

[godlikeue]

I have tried many times and finally located the problematic version to be v1.19.2.
If I revert to v1.19.1, as well as the earlier v1.18.5, this issue will not occur, and the IntelliSense is working now!

My vscode version is 1.89.1.

[smp73]

@microsoft-github-policy-service agree company="Microsoft"

[hbina]

I am remotely developing on my RPI5, it appears that the LSP server is crashing consistently. This is the only output of cpptools.

[Error - 4:01:48 PM] The language server crashed. Restarting...
[Error - 4:01:48 PM] Server process exited with signal SIGSEGV.

[abhishekd-1]

I am remotely developing on my RPI5, it appears that the LSP server is crashing consistently. This is the only output of cpptools.

[Error - 4:01:48 PM] The language server crashed. Restarting...
[Error - 4:01:48 PM] Server process exited with signal SIGSEGV.

I'm seeing the same thing as well on a Raspberry Pi 4 with extension version 1.21.2. Downgrading the C/C++ extension to v1.20.5 fixed the problem for me.

I'm running VS Code version 1.89.1 and have attached the crash stack here as instructed in the first post.

[dkvg-77]

I have been receiving the following error message ever since the update was released 2 days ago.

[Error - 7:36:32 PM] Server initialization failed.
  Message: Pending response rejected since connection got disposed
  Code: -32097 
[Error - 7:36:32 PM] The language server crashed 5 times in the last 3 minutes. It will not be restarted.
[Error - 7:36:32 PM] cpptools client: couldn't create connection to server.
  Message: Pending response rejected since connection got disposed
  Code: -32097 
[Error - 7:36:32 PM] Server process exited with code 3221226505.

I just open up the VS code and this error is thrown.
I tried reinstalling the extensions, reinstalling VS Code, and many other things.
But there has been no solution.
Please help me.

I am using Windows.

[imanabu]

On Windows 11, the extension is now crashing on Language server in the past few days. What can I provide you to help diagnose this?

[sean-mcmanus]

@abhishekd-1 That crash is fixed with 1.21.4.

@dkvg-77 @imanabu Are you able to attach a debugger to the cpptools.exe process to get a crash call stack? https://github.com/microsoft/vscode-cpptools/wiki/Attaching-debugger-to-cpptools-or-cpptools%E2%80%90srv

[dkvg-77]

Hey @sean-mcmanus

I have been trying to attach a debugger to the cpptools.exe process following the steps in the given URL.
After clicking on the “Start Debugging” button, I get to attach a process.
But I can't find 'cpptools.exe' option in there.

And, to talk about the problem of the extension,
I am facing no issues in the 1.20.5 or below version.
I am facing issues only in the last 3 updates.

Also for an option called 'C/C++ tag parser status', it says 'initializing'.
Where as it says 'parsing complete' for the 1.20.5's version.

[imanabu]

@abhishekd-1 That crash is fixed with 1.21.4.

@dkvg-77 @imanabu Are you able to attach a debugger to the cpptools.exe process to get a crash call stack? https://github.com/microsoft/vscode-cpptools/wiki/Attaching-debugger-to-cpptools-or-cpptools%E2%80%90srv

Well, here is what happens so with my knowledge of VSC extensions, right now I am at a loss how to get this attach while the extension is starting up. I have a flutter project with Windows target and I think when that gets detected, the extension starts and then it crashes. I am on 1.21.4

[sean-mcmanus]

@dkvg-77 @imanabu It sounds like it's crashing too early to attach. Are you able to set C_Cpp.logginglevel to "Debug" and look at the C/C++ logging before the 2nd "cpptools/initialize" logging (which occurs after it crashes and re-initializes).

Does it crash with an empty workspace folder without any C/C++ files open? Does opening a file trigger the crash or adding certain files to the workspace folder? If you can attach before it crashes and then do something to trigger the crash, then you can get a crash call stack that way.