Re: [PATCH v2 00/25] Towards great frontend GDB consoles

[Pedro Alves <palves at redhat dot com>]

On 03/21/2016 06:51 PM, Pedro Alves wrote:
> 
>> > 2- Using 'interpreter-exec mi" in the CLI console confuses GDB
>> >     To reproduce from a shell
>> >     - start GDB in CLI mode
>> >     - give a "new-ui mi" command to start an MI channel
>> >     - start running the inferior => *stopped/*running events are on the MI channel
>> >     - in the CLI send command interpreter-exec mi ""
>> >     - step or resume
>> >     => *stopped/*running events are now on the CLI channel and not the MI one
> I can reproduce this.
> 
> This may be related to the fact that MI's code still has a bunch of
> globals that I hadn't made per-UI, thinking that that would be sufficient
> for the use case at hand.  While annoying, I wouldn't expect a
> regular user / script to be issuing MI commands, right?

For the record, this was indeed related to MI globals.  Specifically,
"raw_stdout".  Fixed by making it be per-MI instead:

 https://sourceware.org/ml/gdb-patches/2016-05/msg00130.html

>> > 
>> > 3- Issues when using "set inferior-tty"
>> >     To reproduce from a shell
>> >     - start GDB in CLI mode
>> >     - give a "new-ui mi" command to start an MI channel
>> >     - use "set inferior-tty" to another tty (like eclipse does)
>> >     - in the MI channel, use -exec-run
>> >     => the shell where GDB CLI is running no longer responds to input
> Interesting.  I can reproduce this as well.  I wonder if it's 
> related to PR gdb/18077 (-exec-run results in synchronous execution),
> but then again, it doesn't reproduce without "set inferior-tty".

This was related to an issue with how gdb starts inferiors.  It
vforks, then uses ptrace (PTRACE_ME), and then execs.  Between
the vfork and the exec, the child was running with the MI channel
as current UI, and because it's a vfork, that messed up the
parent gdb's structures.  Fixed here, with a testcase that
ensures that "set inferior-tty" use cases are correctly handled:

 https://sourceware.org/ml/gdb-patches/2016-05/msg00128.html

Thanks,
Pedro Alves