C / C++ / MFC

Re: Best way to check if a thread is still active c++ win32

21-Feb-12 11:35

I'm running a cosmetic loop, and I think I should check to see if the CreateProcess is still running. I've seen other post for the same thing, but could not remember the nomenclature for it.

Just looking for recommendations

I played around with checking the exit code, but I know it's not the proper way to do it

GetExitCodeProcess(pi.hProcess, (unsigned long *)&exit_status);
if (exit_status != 259)
    break;

My CreateProcess

if (CreateProcess(sz_SQLServer_Install_FileName, szParameters, NULL, NULL, FALSE, CREATE_DEFAULT_ERROR_MODE | NORMAL_PRIORITY_CLASS, 0, sz_SQLServer_Install_FolderPath, &si, &pi) ) {
        while(WAIT_TIMEOUT == WaitForSingleObject(pi.hProcess, 10)) {
            MSG oMSG;
            while(::PeekMessage(&oMSG, NULL, 0,0, PM_NOREMOVE)) {
                if(::GetMessage(&oMSG, NULL, 0,0) ) {
                    ::TranslateMessage(&oMSG);
                    ::DispatchMessage(&oMSG);
                }
                else {
                    break;
                }

Randor 21-Feb-12 14:27

Randor

21-Feb-12 14:27

Hi Jim,

What problems are you having? The call to GetExitCodeProcess looks good to me although not sure why you needed the cast on exit_status.

Also... WaitForSingleObject will return WAIT_OBJECT_0 when the process has exited.

Re: Best way to check if a thread is still active c++ win32

jkirkerx21-Feb-12 15:18

Re: Best way to check if a thread is still active c++ win32

21-Feb-12 15:18

The original problem that lead to the question was that when the program running in the process was canceled, or halted due to a requirement missing, the cosmetic loop would still be running, instead of exiting and sending the appropriate message to the message pump to go to the next stage. So I added the GetExitCode to double check that the process was still running, or if the process had been halted.

Just didn't think my quick fix was the correct way to check, and just wanted to do it right.

Randor 21-Feb-12 16:55

Randor

21-Feb-12 16:55

jkirkerx wrote:
The original problem that lead to the question was that when the program running in the process was canceled, or halted due to a requirement missing, the cosmetic loop would still be running, instead of exiting and sending the appropriate message to the message pump to go to the next stage.

I actually have something to say about this. I have noticed over the last few years that many of the younger junior programmers coming out of college are no longer using exit codes. I have noticed this in many of the recent Microsoft products and applications from various other software vendors. I believe it is probably because most of us older engineers have experience with the Unix and DOS shell and have extensively used exit codes in bash,cshrc scripts and/or DOS batch files.

Unfortunately since the installer you are using is not returning an exit code upon cancellation... you may need to think outside the box and come up with a creative way to determine if your installer has completed. I noticed that one of your variable names was associated with Microsoft SQL Server... so maybe you could check the registry keys under the tree: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Microsoft SQL Server\

Finally, I would encourage you (and anyone reading this) to think of this experience as a lesson; Exit codes might be useful to someone... rather than always return 0; consider returning a useful value that reflects the outcome of the task.

-Best Wishes,
-David Delaune

Re: Best way to check if a thread is still active c++ win32

jkirkerx21-Feb-12 17:05

Re: Best way to check if a thread is still active c++ win32

21-Feb-12 17:05

I was considering doing something like that.

Chunk is saying to dump my line of code, and trust the WaitForSingleObject.

I do know the name of the SQL Server installation process, just an idea.

Let me run some more test with and without the cancel after removing the lines, and see what happens again. They take 30 minute each for a full run. Back tomorrow.

Chuck O'Toole21-Feb-12 14:29

Re: Best way to check if a thread is still active c++ win32

21-Feb-12 14:29

There's nothing wrong with using WaitForSingleObject() on the hProcess returned from CreateProcess(). Its very purpose is to give you a waitable object that's signalled when the process teminates. Until it terminates, the GetExitCodeProcess() cannot tell you the exit code Smile | :)

Some folks might comment on the fact that your wait loop includes a message pump which implies that you are waiting during processing of a GUI element / button but to each his own. This works.

jkirkerx21-Feb-12 15:28

Re: Best way to check if a thread is still active c++ win32

21-Feb-12 15:28

Cool.

Well I'm not getting the exit code, but I do get a still active code which is OK for now.

Thanks Chuck.

Chuck O'Toole21-Feb-12 16:12

Complete code for CreateProcess and loop

21-Feb-12 16:12

That's the defined behavior of GetExitCodeProcess(). The exit code does not exist until the process exits! The defined return is "Still Active", meaning "No exit code exists", or colloquially expressed, "Yo, Dude, What Exit Code? It Ain't Exited Yet".

jkirkerx21-Feb-12 15:26

21-Feb-12 15:26

This is the complete code I wrote.

If no one has an objection to the GetExitCodeProcess in the loop, I'll stick with it because it works. I just thought it was completely wrong to use, or perhaps I'm just getting better at this.

if (CreateProcess(sz_SQLServer_Install_FileName, szParameters, NULL, NULL, FALSE, CREATE_DEFAULT_ERROR_MODE | NORMAL_PRIORITY_CLASS, 0, sz_SQLServer_Install_FolderPath, &si, &pi) ) {
        while(WAIT_TIMEOUT == WaitForSingleObject(pi.hProcess, 10)) {
            MSG oMSG;
            while(::PeekMessage(&oMSG, NULL, 0,0, PM_NOREMOVE)) {
                if(::GetMessage(&oMSG, NULL, 0,0) ) {
                    ::TranslateMessage(&oMSG);
                    ::DispatchMessage(&oMSG);
                }
                else {
                    break;
                }

                // Run the Cosmetics on the MDI Window
                if (ndx != 100) {
                    SendMessage(g_SQLServer_Install_Progress_Bar, PBM_SETPOS, (WPARAM)ndx, 0);
                    UpdateWindow(g_SQLServer_Install_MDIWindow);
                    Sleep(100);
                    ++ndx;

                }
                else {
                    if( iMO == 0 ) {
                        SetWindowText(g_SQLServer_Install_Progress_Text, L"Installation may take up to 30 minutes");
                        iMO = 1;
                    }
                    else {
                        SetWindowText(g_SQLServer_Install_Progress_Text, L"Installing Microsoft SQL Server Express 2008");
                        iMO = 0;
                    }

                    UpdateWindow(g_SQLServer_Install_MDIWindow);
                    ndx = 0;
                }

                // Check the process to see if it's still running
                GetExitCodeProcess(pi.hProcess, (unsigned long *)&exit_status);
                if (exit_status != 259)
                    break;
            }
        }

        GetExitCodeProcess(pi.hProcess, (unsigned long *)&exit_status);
        if (exit_status == 0) {
            // The program completed successfully
            SendMessage(g_SQLServer_Install_MDIWindow, WM_COMMAND, (WPARAM)g_SQLServer_Install_IDC_COMMAND, (LPARAM)g_SQLServer_Install_IDM_COMPLETE );
        }
        else {
            // The program was either canceled or failed for some reason
            SendMessage(g_SQLServer_Install_MDIWindow, WM_COMMAND, (WPARAM)g_SQLServer_Install_IDC_COMMAND, (LPARAM)g_SQLServer_Install_IDM_CANCEL );
        }

        CloseHandle(pi.hProcess);
        CloseHandle(pi.hThread);
        exitCode = 1;
        bResult = TRUE;

Chuck O'Toole21-Feb-12 16:22

21-Feb-12 16:22

Well, the call inside the loop is pretty useless but if it makes you happy...

The sole purpose, as written, is to cause the big while loop to break out if the response is anything but "still running". But the WaitForSingleObject() returning anything other than "WAIT_TIMEOUT" (that is, WAIT_OBJECT_0) also breaks out of the loop (the "else clause") which will happen as soon as the thread stops running (which sets the exit code and signals the hProcess object).

Now, I'll grant you that by having the call at the bottom of the while loop may cause you to notice the process termination a couple of microseconds earlier than if you went to the top of the loop but you counter act that miniscule gain by discarding the status you might have received then and issue another GetExitCodeProcess() immediately upon exiting the while loop.

So, it adds nothing to the code, adds processing at each iteration. But, like I said, if it makes you happy ....

Chuck O'Toole21-Feb-12 16:34

21-Feb-12 16:34

HA, I just read the documentation on GetExitCodeProcess()[^] (twice to be sure).

If I were a malicious / devious person, I'd write my process to "return 259;" as the exit status. Since GetExitCodeProcess() returns either status STILL_ACTIVE or the return code from the process. If 259 is my exit code, following Microsoft's method for testing the status would be fooled into thinking the process is still running. Does this count as "Amusing Microsoft API Tricks" Big Grin | :-D

Anyway, WaitForSingleObject() on the hProcess object is the only way to be sure.

jkirkerx21-Feb-12 16:50

21-Feb-12 16:50

Hmm.

So if I just removed my GetExitCode from the loop, I'm good to go?

It's not that I like it.

I'll remove it and run the test again a couple of more times, just to make sure.

Thanks Chuck

Randor 21-Feb-12 17:08

Randor

21-Feb-12 17:08

Jim,

I observe that you are passing a variable named sz_SQLServer_Install_FileName to CreateProcess which would obviously imply that you are installing SQL Server. Are you using a MSI installer? According to the documentation... the MsiExec.exe and InstMsi.exe MSI installers will return useful status/error codes.

Error Codes[^]

Best Wishes,
-David Delaune

jkirkerx21-Feb-12 18:18

21-Feb-12 18:18

I'm running these programs. I think Microsoft wrote the installer.

I just finished running the test, and put the line back in. Can't detect a cancel when I click the red X square during the expansion of the files to the temp dir. With the line, I can detect the click. So I guess the line is useful. I just want to cover my butt, who knows what the user will do.

So I detect the cancel, use the message pump to acknowledge the cancel, messagebox, start the installation again.

WCHAR                   *szFileName_X86         = L"SQLEXPRADV_x86_ENU.EXE";
WCHAR                   *szFileName_X64         = L"SQLEXPRADV_x64_ENU.EXE";

jkirkerx22-Feb-12 6:58