www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.debugger - -checkaction=C - Debugger out of sync?

reply frame <frame86 live.com> writes:
In Visual Studio I can step into a thread and hold on a while. By 
resuming the next instruction some invariant kicks in and throws 
an error but it doesn't hold the thread in the debugger - I just 
see the error on the output.

The debugger is still active but now running from 
`rt.lifetime.finalize_struct(...)` (at least this is shown in the 
stack window) while executing the next step of my user code in a 
already half destroyed struct.

Setting a breakpoint on the invariant assert does not work. In 
detail it's some Schrödingers assert:

If I set a breakpoint without holding the thread on, nothing 
happens and all variants are fine. If I hold the thread on with 
the breakpoint set and then resume with the next step, also no 
error appears. But if I do the same with disabled breakpoint, the 
error is thrown in the output.

It's the same thread, I checked it with GetCurrentThreadId(). 
Also the invariant is in the same thread checked, so the debugger 
is somehow out of sync while the program flow is ahead.

Then I checked that I have supplied `-checkaction=C` on the main 
app while a DLL was compiled with default settings. After 
changing the DLL, the error disappeared. Is this just a 
coincidence or can `-checkaction` cause a stack corruption or 
something?
Jan 25 2022
parent reply frame <frame86 live.com> writes:
 Rainer Schuetze
I quote this (you have sent me somehow a private email)


 If your invariant is called from the GC during finalization, it 
 might
 have to do with the debugger executing code to show structs or 
 ranges.
 Are you using the visualizer methods __debugOverView, 
 __debugExpanded,
 etc.? Are ranges displayed in your Auto/Local/Watch windows?


No, I don't.


 You can try to disable implicit code execution by the debugger 
 in
 Tools->Options->Debugging->Mago by unchecking options "Call
 struct/class..." and "Call range methods".


Yes, the option `Call range methods` causes the problem! Is that 
still experimental in VisualD 1.2.0? Please do not enable it by 
default then.
Jan 27 2022
next sibling parent frame <frame86 live.com> writes:
On Thursday, 27 January 2022 at 17:32:03 UTC, frame wrote:

  Rainer Schuetze


Forgot to say thank you =)
Jan 27 2022
prev sibling parent Rainer Schuetze <r.sagitario gmx.de> writes:
On 27/01/2022 18:32, frame wrote:

  Rainer Schuetze
 I quote this (you have sent me somehow a private email)


Sorry, I still screw this up too often since the default/button order 
has changed in Thunderbird.


 

 If your invariant is called from the GC during finalization, it might
 have to do with the debugger executing code to show structs or ranges.
 Are you using the visualizer methods __debugOverView, __debugExpanded,
 etc.? Are ranges displayed in your Auto/Local/Watch windows?


 
 No, I don't.
 

 You can try to disable implicit code execution by the debugger in
 Tools->Options->Debugging->Mago by unchecking options "Call
 struct/class..." and "Call range methods".


 
 Yes, the option `Call range methods` causes the problem! Is that still 
 experimental in VisualD 1.2.0? Please do not enable it by default then.


It's a bit dangerous because executing code automatically can mutate 
your program state.

The default has been changed quite some time ago 
(https://github.com/rainers/mago/commit/8ce1bb37818ea38e88ec8e
f2439383b592fca46), 
so you probably picked up the enabled state as an early adopter.
Jan 27 2022