Re: [PATCH 0/3 V3] Test mingw32 GDB in cygwin

[Yao Qi <yao@codesourcery.com>]

On 07/29/2013 10:03 PM, Pierre Muller wrote:
>    I didn't like it because I explained that the changes you propose are
> useful
> to run the testsuite on Windows OS, but not only if running
> under a cygwin terminal.

Pierre,
the intention of this patch series is to improve the testing mingw 
native gdb running on a cygwin through ssh.

>    I am using a msys port of dejagnu expect,
> and this needs the same changes (remove buffering and switch to binary
> mode),
> but with the new version of your patch,
> nothing would happen for me and the testsuite would still fail.
>

Your environment is different from ours.  We run dejagnu/expect on linux 
machine, and test mingw32 gdb in a cygwin on windows machine through 
remote host mechanism.

>
>
>> >Thanks to Corinna's example, we can know whether GDB is in cygwin
>> >by checking the file name of handler of stdin (or stdout).  As a
>> >result, a new option '--cygwin-tty' is avoided.  Patch 1/3 is
>> >almost rewritten in V3.
>    Yes, that's true, but the down side is that you transform
> the terminal settings only when you are on a cygwin terminal...
>   I think this is not correct:
>   I suspect that is_in_cygwin will return true on every type of cygwin
> terminal,
> in particular also when you are in interactive mode...

Yes, anything wrong here?

>    I thought that your patch should only change settings if the pipes
> from expect redirection are detected, but there are no pipe checks
> anymore...

expect redirection is not involved in our testing.  We are running 
testsuite in remote host mode, so dejagnu/expect is on a linux machine.

> Note that I am not sure of the above, as 'maybe' the pty/tty name doesn't
> match in the interactive shell case.
>    Secondly, your patch will never trigger the no buffer/binary mode changes
> on non-cygwin terminals, while I argued before that it should also be usable
> in those conditions.
>

It is expected.  We test GDB for different targets on linux host and 
windows host (in cygwin) respectively.  We don't have the environment 
you described, so I am unable to extend the patches to support your 
environment, sorry.

>    This is why I would rather like to have the settings modification
> grouped into a function (let's call it setup_handles_for_testsuite)
> which would be an empty function for all but __MINGW32__ code.
> This function could then be called by
> (gdb) set maint testsuite-mode on
> command or automatically when a cygwin shell and pipes are detected.

If you really need an option, feel free to add one, and adjust my patches.

>
>    I am sorry to bother you again, but I would really like to
> get a patch that can be used in more general framework and
> that avoids to change the standard handles when not necessary.
>
>    And, to finish, I would like to reiterate my support for this
> patch, even though I am still criticizing it.
>
> Sorry to bother you more,

Never mind.  I don't see much conflicts here.  I'd like to improve mingw 
gdb testing in cygwin via remote host, and you'd like to improve mingw 
gdb testing on windows machine.  We have something in common, such as 
unbuffer stdout/stderr, etc, but we have different criteria on when to 
turn on these changes.  In my side, they can be turned on when gdb is 
running in cygwin, while in your side, then can be turned on when a 
certain option is set.  I don't think my patches make troubles or 
regressions to what you want to achieve.  I am inclined to to get these 
patches committed (after the review), and you can do some changes for 
your purpose on top of it.

-- 
Yao (齐尧)