This is the mail archive of the cygwin mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: ptys - I give up


On 7/26/2013 10:10 AM, Corinna Vinschen wrote:
> On Jul 26 09:55, Daniel Colascione wrote:
>> On 7/26/2013 9:35 AM, Corinna Vinschen wrote:
>>> On Jul 26 09:21, Daniel Colascione wrote:
>>>> On 7/26/2013 8:27 AM, Christopher Faylor wrote:
>>>>> On Thu, Jul 25, 2013 at 11:44:32PM -0700, Daniel Colascione wrote:
>>>>>
>>>>>> Ugly, only half-implemented, but better: a hook-based pseudoconsole
>>>>>> system for Windows.
>>>>>
>>>>> This is what I was holding out for.  The last time it came up here,
>>>>> people seemed vehemently opposed to the idea since implementations that
>>>>> do this (like the Console app and a couple of libraries floating around)
>>>>> seem to have to poll the console looking for data and that could result
>>>>> in data loss.
>>>>>
>>>>> I have had code (not mine) sitting in my Cygwin directory for years
>>>>> which does this but I couldn't 100% convince myself that the detractors
>>>>> for this idea weren't right.  Does your code work around this?
>>>>
>>>> I'm also against screen scrapers. That's why I wrote this library: it doesn't
>>>> poll. It actually intercepts console APIs and implements console handles as
>>>> pseudohandles, just like Windows <= 7 did, forwarding console API requests to a
>>>> server. Any console program that works with Windows 7 should work with this
>>>> library.
>>>
>>> Out of curiosity, can you expand what you mean here?  I see a difference
>>> between Windows pre-7 and Windows 7 and later. 
>>
>> Conhost exists to solve the problem Raymond Chen described here:
>> http://blogs.msdn.com/b/oldnewthing/archive/2007/12/31/6909007.aspx. Conhost
>> exists so there's something running with lower privileges that can draw console
>> UI --- it's a broker. The actual console-ing still happens entirely in user mode
>> though, with the functions in kernel32 conspiring to provide the illusion that
>> the system has something called a console handle.
>>
>> In Windows 8, the implementation is completely different. There, console handles
>> are real kernel objects.
> 
> Really?  That's entirely new to me.  Do you know if there's some description
> available?  What are the implications?  Are console handles now handles to
> something like named pipes as well?  Is there a console FS comparable to
> \Device\NamedPipe\?

I don't know the details, and I'm not aware of any new facilities exposed to
applications. I'd have loved a pseudoconsole API, but we didn't get one.

You can see the difference in a debugger, though: console pseudohandles *look*
strange and don't show up in the process handle table. The difference shouldn't
really matter, though: applications shouldn't really care what kind of handles
they get as long as system APIs accept them. The problem with writes larger than
8k (I think that was the limit?) to console handles failing seems to have
disappeared, though.

>> Anyway, that pre-8 Windows used pseudo-handles for console handles is a good
>> thing now. It means that any library that tries to intercept all console
>> functions doesn't need to emulate all possible operations on console handles:
>> instead, it just needs to emulate the operations Windows actually allowed on
>> console handles, which was actually a fairly limited subset of what's allowed
>> for handles in general.
>>
>> Another way of saying it is that it's easy to fake console handles because
>> Windows, too, was faking all along.
> 
> But then again, doesn't that mean that your method stops working on W8?

Not really --- it'll work fine. The difference means that operations on console
handles that used not to work on Windows 7 might work on Windows 8. The public
API contract hasn't changed, however, so applications aren't supposed to be
doing these things anyway, and as long as application developers care about
testing their applications on Windows 7, they won't.

If you eventually do start seeing applications that do things with console
handles that can't be done with console pseudohandles, well, you can hook more
APIs to preserve the console handle illusion. But that time is a long way away,
if it ever arrives at all.

Attachment: signature.asc
Description: OpenPGP digital signature


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]