Re: SIGTRAP in kill emulation on Windows

[Alain Schneble]

Eli Zaretskii <address@hidden> writes:

>> Sending SIGTRAP to a process -- whether it's an unrelated, child or the
>> calling Emacs process itself -- does not seem to have any effect as long
>> as no debugger is attached to the receiving process.
>
> Do you see the same with the debugbreak program you've built earlier?
> If not, then that program does something that your patch doesn't.

Yes, debugbreak and the proposed implementation behave identically.

> Also, were all the programs you tried built with MinGW?  If so,
> perhaps try with other programs, like Python or one of the programs
> that came with Windows.  I think the results of the call depend on the
> exception handlers installed by the program when it starts, so
> different programs and different ways of building programs might have
> different effects.

I tried it with different programs (e.g. notepad, calculator as well as
programs built with MinGW such as Emacs and find).  Yes, I think as well
that it depends on the exception handlers installed by the programs.  So
there might be programs out there that will terminate.  But I had no
luck in finding one so far...

> And finally, what happens if you signal the calling Emacs process that
> is itself being debugged?  I guess the debugger will kick in, but if
> you then continue execution, does Emacs continues to run normally?

Exactly, the control will be passed to the debugger.  I did not see any
strange behavior when resuming it ( (gdb) cont ).  After all, this will
be used by one of the use cases I have in mind, to be able to switch
back to the debugger from within the debuggee.

>> At least this is what I observe on Windows 10.  Shall I test on
>> other OS versions as well?
>
> We could ask people here to test on other versions and report their
> results.

FWIW, I tested on Windows 7 in addition to Windows 10.  Same behavior
on both systems.

>> If so, I would have to install say an Win XP first as I do not
>> have one available right now.
>
> I could test on XP here.

Thanks.  That would save me some time.

>> Unfortunately, on MSDN the remark on
>> DebugBreakProcess says it causes the receiving process to terminate in
>> most cases, see
>> https://msdn.microsoft.com/en-us/library/windows/desktop/ms679298(v=vs.85).aspx
>> 
>> But with all processes I tried, none of them was ever terminated.
>
> That's a pity; on GNU/Linux the target program does terminate.  If
> there's nothing that can be done, we will just have to document this
> quirk, at least for many/most programs Emacs users will meet on their
> systems.

Yes, and on GNU/Linux I observed that it prints a backtrace prior to
termination.  But that's kind of irrelevant here I guess.  Where would
be the best place to document this quirk (c, signal-process, info
manual)?

FWIW, there is probably something we could do about it -- query if the
process in question is attached to a debugger.  If not, we could
terminate it.

>> >From ecf00b52957bb0bf15735180a67cc9ce7027ff19 Mon Sep 17 00:00:00 2001
>> From: Alain Schneble <address@hidden>
>> Date: Thu, 25 Aug 2016 21:31:48 +0200
>> Subject: [PATCH] Support SIGTRAP in kill emulation on Windows
>
> The (latest version of) your patch looks fine to me, modulo the issues
> mentioned above.  If nothing new and no other comments come up within
> a week, let's install this, with the necessary updates for the NEWS
> entry.

Thanks, Eli.