Author Topic: Code completion using LSP and clangd  (Read 12386 times)

Offline Pecan

  • Plugin developer
  • Lives here!
  • ****
  • Posts: 2345
Re: Code completion using LSP and clangd
« Reply #60 on: January 18, 2022, 08:30:01 pm »
I see there are many pch files in the folder such as:

C:\Users\[myusername]\AppData\Local\Temp\preamble-c7460b.pch

I think those files is created by clangd, and are there any way to automatically delete them when exit C::B?

EDIT:

clangd writes too much disk : CPP-19402

This discussion looks like the pch can keep in "memory". :)

Here is the patch to fix this pch file issue:

Code
 clangd_client/src/LSPclient/src/client.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/clangd_client/src/LSPclient/src/client.cpp b/clangd_client/src/LSPclient/src/client.cpp
index 88f5f8f..f6b5eb9 100644
--- a/clangd_client/src/LSPclient/src/client.cpp
+++ b/clangd_client/src/LSPclient/src/client.cpp
@@ -266,6 +266,10 @@ ProcessLanguageClient::ProcessLanguageClient(const cbProject* pProject, const ch
 
     command += " --limit-results=20";              // Limit the number of results returned by clangd. 0 means no limit (default=100)
 
+    // clangd writes too much disk : CPP-19402 https://youtrack.jetbrains.com/issue/CPP-19402
+    // "-pch-storage=memory"
+    command += " -pch-storage=memory";
+
     if (wxDirExists(clangResourceDir))
         command += " --resource-dir=" + clangResourceDir;  // Directory for system includes
 

Thanks for that.

I've added code to remove preamble-*.tmp and preamble-*.pch files when the project closes.

I'll later add an option to keep the .pch in memory. (For me, I'd rather keep them on disk and remove them at project close time).

As an aside: Windows does not allow me to remove files that are open and being used. But Linux does. Do you know of a way to tell Linux NOT to allow me to delete open files?

Offline Pecan

  • Plugin developer
  • Lives here!
  • ****
  • Posts: 2345
Re: Code completion using LSP and clangd
« Reply #61 on: January 18, 2022, 08:54:19 pm »
I see there are some code snippet like:

Code
m_MutexInputBufGuard.Lock;

m_MutexInputBufGuard.Unlock();

But in the code, we have to carefully handle the unlocking the wxMutex when return the function body, especially when there are multiply returns.

Is it possible to use the wxMutexLocker, and check the IsOK() function for checking whether it get locked or not.

There are only 2 locks in the code that can cause any trouble. The lock on the input buffer. One to write to the buffer, and one to get the next clangd response out of the buffer. And neither affect the UI thread.

I tried wxMutexLocker first before giving up on it.
I want to be able to unlock the input buffer and then do more work in the function.
When I unlocked the mutex before the function ended,  wxWidgets gave me errors about the mutex. I lost confidence that I could mix a wxMutexLocker and manual unlocks.

In fact, I removed all locks on the main UI thread and used idle time callbacks instead.

Offline ollydbg

  • Developer
  • Lives here!
  • *****
  • Posts: 5443
  • OpenCV and Robotics
    • Chinese OpenCV forum moderator
Re: Code completion using LSP and clangd
« Reply #62 on: Yesterday at 08:07:39 am »
I see there are some code snippet like:

Code
m_MutexInputBufGuard.Lock;

m_MutexInputBufGuard.Unlock();

But in the code, we have to carefully handle the unlocking the wxMutex when return the function body, especially when there are multiply returns.

Is it possible to use the wxMutexLocker, and check the IsOK() function for checking whether it get locked or not.

There are only 2 locks in the code that can cause any trouble. The lock on the input buffer. One to write to the buffer, and one to get the next clangd response out of the buffer. And neither affect the UI thread.

I tried wxMutexLocker first before giving up on it.
I want to be able to unlock the input buffer and then do more work in the function.
When I unlocked the mutex before the function ended,  wxWidgets gave me errors about the mutex. I lost confidence that I could mix a wxMutexLocker and manual unlocks.

In fact, I removed all locks on the main UI thread and used idle time callbacks instead.

Yes, there are only There are only 2 locks in the code. Sometimes, I got error message about lock failed, I'm not sure why.

BTW: since the console pipe is connected with clangd.exe, I'm not sure why the locker is needed. Since the content is from a thread to the main GUI thread by the Event, when you got the Event, you were already in the main GUI thread, and the content in the Event(wxThreadEvent) is already deep copied. So, I think the locker is not necessary, am I correct?

If some piece of memory should be reused, turn them to variables (or const variables).
If some piece of operations should be reused, turn them to functions.
If they happened together, then turn them to classes.

Offline Pecan

  • Plugin developer
  • Lives here!
  • ****
  • Posts: 2345
Re: Code completion using LSP and clangd
« Reply #63 on: Yesterday at 11:49:52 pm »

BTW: since the console pipe is connected with clangd.exe, I'm not sure why the locker is needed. Since the content is from a thread to the main GUI thread by the Event, when you got the Event, you were already in the main GUI thread, and the content in the Event(wxThreadEvent) is already deep copied. So, I think the locker is not necessary, am I correct?

There are two non main UI threads accessing the clangd input buffer. The pipe thread and the ReadJson thread. Without the lock I assume the ReadJason thread could remove data at the same time the pipe "ProcessEvent" was writing to the (UI client.h) std::string buffer.

the "locks failed" messages could be coming from attempts to lock the symbols tree (comsuming). But it's ok. If the symbols tree update thread can't get the lock. Its ok to block.

If the lock fails in :LSP_ParseDocumentSymbols (stows symbols in tree), it just requeues itself for an idle time callback. It does not block. 
« Last Edit: Today at 12:01:51 am by Pecan »

Offline ollydbg

  • Developer
  • Lives here!
  • *****
  • Posts: 5443
  • OpenCV and Robotics
    • Chinese OpenCV forum moderator
Re: Code completion using LSP and clangd
« Reply #64 on: Today at 02:50:10 am »

BTW: since the console pipe is connected with clangd.exe, I'm not sure why the locker is needed. Since the content is from a thread to the main GUI thread by the Event, when you got the Event, you were already in the main GUI thread, and the content in the Event(wxThreadEvent) is already deep copied. So, I think the locker is not necessary, am I correct?

There are two non main UI threads accessing the clangd input buffer. The pipe thread and the ReadJson thread. Without the lock I assume the ReadJason thread could remove data at the same time the pipe "ProcessEvent" was writing to the (UI client.h) std::string buffer.

the "locks failed" messages could be coming from attempts to lock the symbols tree (comsuming). But it's ok. If the symbols tree update thread can't get the lock. Its ok to block.

If the lock fails in :LSP_ParseDocumentSymbols (stows symbols in tree), it just requeues itself for an idle time callback. It does not block.

Hi, Pecan. Thanks for the explanation.

I'm surprised to find that you implemented the symbol tree. This is really a good job!

If some piece of memory should be reused, turn them to variables (or const variables).
If some piece of operations should be reused, turn them to functions.
If they happened together, then turn them to classes.