Issue starting brand new projects

mrjj

@koahnig
Ok so it's not path confusion.

Does it creates the exe?

koahnig

@mrjj
Sorry, answered too fast.
Yes, exes are always created. Had checked this before.

mrjj

@koahnig
Ok so it's not a permission issue.

Do you have mutiple version of Qt installed?

Can it have shifted path so it see other version on start?

Have you checked the kit. the debugger it points to etc?

Must be a (some sort of) config problem since it didn't happen at the win 10 upgrade.

kshegunov

@koahnig
My suspicion would be that the build and run configurations don't match, so creator is not able to start the exe. Could you check whether the run config matches the build paths?

koahnig

@kshegunov
Yes, it does does.
I have changed sessions. The first session has quite a number of projects (around 15, did not check). The new session had only 3 projects. Did the same thing creating the different projects (GUI, core and also a test project) with two different Qt versions as also used with the larger session. In the smaller session all works.

However, loading the test project created in the previous, large session I have the same issue again. Also when loading after deleting the *.pro.user.

koahnig

@mrjj @kshegunov
It gets more ridiculous.

Reopened the large session (has 20 entries BTW). Opened another test project and now all works again.

However, a bit more testing showed that it might have been the naming of the project. I used always "Setup" in the name. Before i had often removed my failures. This time I made sure that the project name was new "SetupHelping", it failed again.
When creating another project with the name "S_E_T_U_P_Helping" it works.

That is really ridiculous.

kshegunov

@koahnig
That indeed sounds ridiculous. I tried on my machine and it works fine, but I wouldn't expect problems, as I'm developing on Linux (Qt Creator 3.6.1 if you're wondering).

Any updates on Qt/Qt Creator when you changed the windows version?
What about Win10 security policies and/or antivirus software, those could be interfering with the debugger?
I think MS introduced some changes with the security in Win10, and I believe it also reflects the debugging APIs, so perhaps it's worth taking a look into that? One thing you could also try is to start a console debugging session with gdb and see if it starts okay. If it doesn't I'd search for some debugger/OS incompatibility.

mrjj

@koahnig said:
SetupHelping

Wow. That I never would have guess in a million years.
I can reproduce it.
If I make brand new project and name it SetupHelping
it cannot run.
and
"Failed to start program. Path or permissions wrong?"

Win 10, Qt 5.6. mingw

kshegunov

@mrjj

That I never would have guess in a million years.

Neither would have I.

I can reproduce it ... Win 10, Qt 5.6. mingw

What about the creator version? Can you perhaps start the application directly (not through creator/debugger)? Can you "run" it in creator without debugger? Can you start a console debugging session?

mrjj

@kshegunov said:
It cannot run from creator at all.
I can start it in in deploy folder.

koahnig

@mrjj @kshegunov
I am using a self-compiled Qt creator 3.1.2 (rather old and compiled with MinGW 4.8.2 32 bit). Qt version 5.4.1 and 5.3.1 precompiled.

I thought already that the issue has something to do with my AV-scanner bitdefender.

@mrjj Which are you using?

Possibly it could be the windows defender stuff too.

It is the TARGET name you have to change. Simply changed in .pro to "TARGET = S_e_t_u_p_Helping" or "TARGET = Set_upHelping" and it works.

kshegunov

@mrjj
That is very, very strange ...

@koahnig
To be honest I'm running out of ideas ... perhaps putting the target name into a variable (I quoted it to be sure) ...?

MY_TARGET_NAME = "SetupHelping"
TARGET = $(MY_TARGET_NAME)

koahnig

@kshegunov
You are the only one on Linux. So, possibly limited to Windows or even Win10.

@kshegunov said:

MY_TARGET_NAME = "SetupHelping"
TARGET = $(MY_TARGET_NAME)

This does not do the trick.

kshegunov

@koahnig

You are the only one on Linux. So, possibly limited to Windows or even Win10.

I'd even say probable. I don't experience any strange behavior by changing the target to the problematic one.

This does not do the trick.

Well, my "shot in dark" approach has met its demise ... sadly I have nothing further to suggest. :|

mrjj

it doesn't even like
e:\build-mysetup-Desktop_Qt_5_6_0_MinGW_32bit-Debug\debug\mysetup.exe

I mean mysetup. ?!

occult name. i say..

Maybe others with win 10 can try.

hskoglund

Hi, I remember Windows can be very sensitive to .exe filenames containing the words "setup" or "update", see discussion here

koahnig

@hskoglund
I was also my guess that it is a windows/AV scanner issue. However @mrjj could start the exe in the deploy folder. See further above.

mrjj

Hi
Just for test.
On windows 7, you can name it anything setup.
Still runs.

So its a win 10 feature. :)
Tonight I will reset my UAC and see if it actually try to warn.

hskoglund

Hi, just tested with an empty widgets app, named TestSetup :-)

Compile/building it with a Microsoft compiler, for example MSVC2013 or MSVC2015 (32-bit or 64-bit same result) will give you an .exe which you can start both from Qt Creator and from File Explorer, without anyone complaining, in any version of Windows.

However compile/build with a MinGW compiler, for example 4.9.2 32-bit, will give you an .exe file that will fail to start from Qt Creator and give you a UAC prompt when started from File Explorer in Windows 7, Windows 8 and Windows 10 (but not in Windows XP).

The reason is that the MinGW does not include a manifest XML file, the crucial part of the manifest is this part:

<trustInfo xmlns="urn:schemas-microsoft-com:asm.v3">
    <security>
      <requestedPrivileges>
        <requestedExecutionLevel level='asInvoker' uiAccess='false' />
      </requestedPrivileges>
    </security>
  </trustInfo>

(i.e. if I blank out that text in the .exe file which was compiled with a MSVC compiler, it will then also give you a UAC prompt.)

kshegunov

@hskoglund
Doesn't this mean you're bound to use the M$ compiler (if there isn't a way to generate and embed the manifest with MinGW)? If so, this'd be terrible ...