Code completion using LSP and clangd

[ollydbg]

Probably because of recent changes in C::B sdk, clangd_client does not compile when using wxWidgets 3.2.1 (probably >= 3.1.6)
Error in client.cpp, line 3672 : CodeBlocksLogEvent waits for a wxBitmapBundle*, but logbmp is a wxBitmap*

I'm building C::B rev12886, and clangd_client rev76, I don't have build issue against wx 3.2.1.

Maybe the recent change from rev12887 to rev12890 causes the build issue.

[Miguel Gimenez]

Starting with r12887, if you are compiling with wxWidgets >= 3.1.6 the last parameter of CodeBlocksLogEvent must be a pointer to wxBitmapBundle.

The change is straightforward, I will publish a patch in this thread tomorrow (I have detected an error in the original notebook HiDPI code that must be fixed beforehand).

[Miguel Gimenez]

Bug fixed in r12891.

To fix the plugin replace (in client.cpp:3664) this

Code Select Expand


        const int uiSize = Manager::Get()->GetImageSize(Manager::UIComponent::InfoPaneNotebooks);
        const int uiScaleFactor = Manager::Get()->GetUIScaleFactor(Manager::UIComponent::InfoPaneNotebooks);
        const wxString imgFile = ConfigManager::GetDataFolder()
                              + wxString::Format(_T("/resources.zip#zip:/images/%dx%d/filefind.png"),
                                                 uiSize, uiSize);
        wxBitmap* logbmp = new wxBitmap(cbLoadBitmapScaled(imgFile, wxBITMAP_TYPE_PNG, uiScaleFactor));

with this

Code Select Expand


        const int uiSize = Manager::Get()->GetImageSize(Manager::UIComponent::InfoPaneNotebooks);
        wxString prefix(ConfigManager::GetDataFolder()+"/resources.zip#zip:/images/");
#if wxCHECK_VERSION(3, 1, 6)
        const double uiScaleFactor = Manager::Get()->GetUIScaleFactor(Manager::UIComponent::InfoPaneNotebooks);
        wxBitmapBundle* logbmp = new wxBitmapBundle(cbLoadBitmapBundle(prefix, "filefind.png", wxRound(uiSize/uiScaleFactor), wxBITMAP_TYPE_PNG));
#else
        prefix << wxString::Format("%dx%d/", uiSize, uiSize);
        wxBitmap* logbmp = new wxBitmap(cbLoadBitmap(prefix+"filefind.png", wxBITMAP_TYPE_PNG));
#endif


[ollydbg]

Hi, Miguel Gimenez, Thanks for the fix.

[Pecan]

@ MaxGaspa

Until I can find the cause of the crash, you can disable Settings/Editor/GeneralSetting/Documentation popup.

This should eliminate the stall/crash. If It does not, let me know.

Thanks

[Pecan]

@ MaxGaspa
Until I can find the cause of the crash, you can disable Settings/Editor/GeneralSetting/Documentation popup.
This should eliminate the stall/crash. If It does not, let me know.
Thanks

RFC: patch for ticket 1168.
This is a Request For Comment for the attached patch.
I have posted a patch on the second page of ticket #1168:
Direct link to Patch

The attached patch (ccManagerMouseTrap220915-1.patch) resolves the following described conditions for both CodeCompletion and Clangd_client. The patch has been tested on both MSW and Linux.

1) The HtmlDocumentationPopup gets stuck showing and cannot be closed.
2) The user did a double-click to select from the AutocompPopup, the HtmlDocumentationPopup got left showing and CB is frozen (and may crash).
3) The AutocompPopup and the HtmlPopup are showing, the user unfocused CB then re-focused CB and CB is frozen and must be killed.
4) The user has unChecked Documentation popup in options and now AutocompPopups cannot be scrolled.
5) The AutocompPopup only is stuck showing and cannot be closed.

What is actually happening? (tldr version)
CB is getting stuck in function OnPopScroll statement:

Code Select Expand

else if (m_pAutocompPopup && m_pAutocompPopup->GetScreenRect().Contains(pos))
in the GetScreenRect() function because the AutocompPopup window has been hidden and the HtmlDocumentationPopup may or may not be hidden, but a wxEVT_MOUSEWHEEL connection is owned by the html popup.

It seems that wxWidgets does not like this condition.
The debugger cannot even step out of this condition.

The problem usually arises from a slower double-mouse-click on an AutocompPopup item such that the double click gets interpreted as two single-clicks.
The AutocompPopup will get closed leaving the HtmlDocumentationPopup out to dry.
Now the conditions on which ccManager depended to remove the event connection will never be met.
We than have a happy deadly embrace.

Other sequences of action also cause the deadly embrace including (occasionally) un-focusing and re-focusing CB while both popups are showing.

I didn't trace through all the conditions causing the deadly embrace since the fix to avoid connecting wxEVT_MOUSEWHEEL to HtmlPopup when AutocompPopup was hidden fixed most of the other conditions also. (ie. the first fix eliminated the possibility of the others).
I did, however, trace through all listed conditions once the fix was applied locally.

I would like to apply the patch before the next nightly.
Comments please....

[ollydbg]

1) The HtmlDocumentationPopup gets stuck showing and cannot be closed.
2) The user did a double-click to select from the AutocompPopup, the HtmlDocumentationPopup got left showing and CB is frozen (and may crash).
3) The AutocompPopup and the HtmlPopup are showing, the user unfocused CB then re-focused CB and CB is frozen and must be killed.
4) The user has unChecked Documentation popup in options and now AutocompPopups cannot be scrolled.
5) The AutocompPopup only is stuck showing and cannot be closed.

Hi, Pecan, good work! I see issue5 many times randomly in my daily work, but it is hard to reproduce, I even don't know how to reproduce this issue. The popup window shown on top of every application, and even CB is not focused, the popup window is still showing. What I have to do is kill the C::B process from the task manager.

Hope your fix will solve those issues, thanks!

[ollydbg]

About the ccManagerMouseTrap220915-1.patch file.

I'm not sure, but this variable

Code Select Expand


std::map<cbEditor*,bool> m_EdAutoCompMouseTraps;


The bool value is always "true"? Since I see the only code here is:

Code Select Expand


m_EdAutoCompMouseTraps[ed] = true;

Also, is the code Windows related? If yes, I think the "#ifdef __WXMSW__" should also be placed in the header file around this member variable declaration.

[Pecan]

About the ccManagerMouseTrap220915-1.patch file.

I'm not sure, but this variable

Code Select Expand


std::map<cbEditor*,bool> m_EdAutoCompMouseTraps;


The bool value is always "true"? Since I see the only code here is:

Code Select Expand


m_EdAutoCompMouseTraps[ed] = true;

Also, is the code Windows related? If yes, I think the "#ifdef __WXMSW__" should also be placed in the header file around this member variable declaration.

The declaration and all references to m_EdAutoCompMouseTraps are already guarded by "#ifdef __WXMSW__"

Code Select Expand

#ifdef __WXMSW__
        /** a handle to the autocomplete list window created by (wx)scintilla, needed under Windows
         * to determine its dimensions (so the scroll event can be sent to it, if relevant)
         */
        wxListView* m_pAutocompPopup;

        /**
         * List of editors holding an event connect to popup mouse scroll event
         * for AutoCompPopup and html Documentation popup
         */
        std::map<cbEditor*,bool> m_EdAutoCompMouseTraps;
#endif // __WXMSW__

The statement:

Code Select Expand

m_EdAutoCompMouseTraps[ed] = true; is only used to enter the cbEditor* into the map assuring only one entry for this editor exists. Is there a better method for inserting a entry into a map?

It is removed by statement:

Code Select Expand

m_EdAutoCompMouseTraps.erase(m_pLastEditor);

I will make a specific comment explaining that.
Thanks for reviewing.

[ollydbg]

I'm tweaking wx sample workspace, see here: codeblocks cbp projects for wx samples

While, after switch from one cbp to other cbp for many times, I got many clangd.exe process running, see the image shot below, is it correct?

Thanks.

[ollydbg]

Is there a better method for inserting a entry into a map?

I'm not fully understand the code logic, but do you really need a map? When a key is added, it's  value is always "true".

I mean a std::set<cbEditor*> is better than std::map<cbEditor*,bool> ?

[Pecan]

I'm not fully understand the code logic, but do you really need a map? When a key is added, it's  value is always "true".

I mean a std::set<cbEditor*> is better than std::map<cbEditor*,bool> ?

Thanks, I learned a long time ago that other eyes reviewing code can make it better.
Direct link to newer patch

[Miguel Gimenez]

Patch for ticket #1168 applied in r12899, thank you.

[ollydbg]

The msys2 project now has clangd 15.0 for mingw64 targets.

[ollydbg]

I have a question:

LSP_ParseSemanticTokens(), what does this function used for?

I see this code:

Code Select Expand


// ----------------------------------------------------------------------------
void Parser::OnLSP_RequestedSemanticTokensResponse(wxCommandEvent& event)  //(ph 2021/03/12)
// ----------------------------------------------------------------------------
{
    if (GetIsShuttingDown()) return;

    // This is a callback after requesting textDocument/Symbol (request done at end of OnLSP_RequestedSymbolsResponse() )
    // Currently, we allow SemanticTokens for the BuiltinActiveEditor only,

    // ----------------------------------------------------------------------------
    ///  GetClientData() contains ptr to json object
    ///  DONT free it! The return to OnLSP_Event() will free it as a unique_ptr
    // ----------------------------------------------------------------------------
    json* pJson = (json*)event.GetClientData();
    wxString idStr = event.GetString();
    wxString URI = idStr.AfterFirst(STX);
    if (URI.Contains(STX))
        URI = URI.BeforeFirst(STX); //filename

    wxString uriFilename = fileUtils.FilePathFromURI(URI);      //(ph 2021/12/21)
    cbEditor*  pEditor =  nullptr;
    cbProject* pProject = nullptr;
    EditorManager* pEdMgr = Manager::Get()->GetEditorManager();
    EditorBase* pEdBase = pEdMgr->IsOpen(uriFilename);
    if (pEdBase)
    {
        pEditor = pEdMgr->GetBuiltinActiveEditor();
        if (not pEditor or (pEditor->GetFilename() != uriFilename))
            return;
        ProjectFile* pProjectFile = pEditor->GetProjectFile();
        if (pProjectFile) pProject = pProjectFile->GetParentProject();
        if ( (not pProjectFile) or (not pProject) ) return;
        ParserBase* pParser = GetParseManager()->GetParserByProject(pProject);
        if (not pParser)
            return;
    }
    else return;

    if (not pProject) pProject = Manager::Get()->GetProjectManager()->GetActiveProject();
    ProcessLanguageClient* pClient = GetLSPClient();

    // Queue the the json data to OnLSP_ParseDocumentSymbols() event, passing it the json pointer
    // The json data will be placed in a queue to be processed during OnIdle() events. //(ph 2021/09/11)
    wxCommandEvent symEvent(wxEVT_COMMAND_MENU_SELECTED, XRCID("textDocument/semanticTokens"));
    symEvent.SetString(uriFilename);
    symEvent.SetClientData(pJson);
    LSP_ParseSemanticTokens(symEvent); //Call directly


The last line:
LSP_ParseSemanticTokens(symEvent); //Call directly

This use the our old-CC's parsing code, for example

Code Select Expand

// ----------------------------------------------------------------------------
Token* LSP_SymbolsParser::DoHandleClass(EClassType ct, int lineNumber, int lastLineNumber, int endCol)            //(ph 2021/05/15)
// ----------------------------------------------------------------------------
{
    // need to force the tokenizer _not_ skip anything
    // as we 're manually parsing class decls
    // don't forget to reset that if you add any early exit condition!
    TokenizerState oldState = m_Tokenizer.GetState();
    m_Tokenizer.SetState(tsNormal);