Re: [RFC] win32-nat.c: Handle EXCEPTION_INVALID_HANDLE as SIGSYS

[Christopher Faylor <cgf-use-the-mailinglist-please@sourceware.org>]

On Wed, Oct 24, 2007 at 09:39:58AM +0100, Pedro Alves wrote:
>Pierre Muller wrote:
>>
>>   If I cut my patch into three parts:
>> 3) Remove wrong CloseHandle calls
>
>I'd prefer if this went in first.  I'm convinced this is the correct
>thing to do everywhere.

I agree.  I thought that I noticed that these handles needed to be
closed so I'd rather that my confusion go away as soon as possible.

>(Hey, I'm happy in that allows me to remove an ugly #ifdef in
>gdbserver/win32-low.c.)
>
>( ( ( If it isn't [the correct thing to do everywhere], then we can
>follow Cygwin's lead, and
>install a top level SEH handler in gdb that handles this exception.
>It could have
>always have been done in gdb the first place, even for Cygwin.  Remember, the
>exception is only thrown iff a debugger is attached, and a debugger
>can easilly not
>pass it to the inferior.  That would mean 0 changes to win32-nat.c debug API
>usage.  Of course, any other inferior that throws that exception will have
>the same problem. ) ) )
>
>> 1) Handle EXCEPTION_INVALID_HANDLE exception as a SIGSYS target signal
>> 2) Add the stoponinvalidhandle
>
>EXCEPTION_INVALID_HANDLE is not quite a SIGSYS.  Plus, a Cygwin inferior should
>never see this, as you shouldn't be using CloseHandle in user apps there anyway.
>Maybe we can take the oportunity to implement more generic Windows exceptions
>support, not just EXCEPTION_INVALID_HANDLE.  Something similar to the "handle"
>command.

That seems like a better way to go to me.

However, even if it isn't the need to do something like:

if (something)
  DoSomething(arg)

should really make you think "Hey, I guess I'd better use a function" which
tests something and then calls DoSomething with the arg.

>> then, after the 1) is applied,
>> you will get several SIGSYS when debugging gdb with itself.
>>   I suppose that this would introduce lots of
>> new failures to the testsuite, especially in the gdb.gdb subdirectory.
>>
>
>Only on MinGW gdb, so they wouldn't qualify as "new" failures.  And only on
>tests that load gdb in gdb, or that have a `close' with a broken argument.
>
>>   I would really prefer to keep trace of
>> the old CloseHandles, in case we discover that under special
>> circumstances, or for some versions of the system,
>> these calls are needed.
>>
>
>I'd prefer we don't add such ugliness.

Ditto.

cgf