Re: Version 1.8.1 does not compile on Cygwin 1.7.14

From: Mark Levedahl <mlevedahl@gmail.com>
Date: Jan 6, 2013, 22:16 UTC
Message-ID: <50E9F7C2.1000603@gmail.com>
In-Reply-To: <7vfw2enl2l.fsf@alter.siamese.dyndns.org>

On 01/06/2013 02:54 PM, Junio C Hamano wrote:

Show 29 quoted lines

> Jonathan Nieder <jrnieder@gmail.com> writes:
>
>> Mark Levedahl wrote:
>>
>>>                                                           However, the newer
>>> win32api is provided only for the current cygwin release series, which can
>>> be reliably identified by having dll version 1.7.x, while the older frozen
>>> releases (dll versions 1.6.x from redhat, 1.5.x open source) still have the
>>> older api as no updates are being made for the legacy version(s).
>> Ah.  That makes sense, thanks.
>>
>> (For the future, if we wanted to diagnose an out-of-date win32api and
>> print a helpful message, I guess cygcheck would be the command to use.)
> Hmph, so we might see somebody who cares about Cygwin to come up
> with a solution based on cygcheck (not on uname) to update this
> part, perhaps on top of Peff's "split default settings based on
> uname into separate file" patch?
>
> If I understood what Mark and Torsten wrote correctly, you will have
> the new win32api if you install 1.7.17 (or newer) from scratch, but
> if you are on older 1.7.x then you can update the win32api part as a
> package update (as opposed to the whole-system upgrade).  A test
> based on "uname -r" cannot notice that an older 1.7.x (say 1.7.14)
> installation has a newer win32api because the user updated it from
> the package (hence the user should not define CYGWIN_V15_WIN32API).
>
> Am I on the same page as you guys, or am I still behind?
>
> In the meantime, perhaps we would need something like this?

It's perhaps worth noting how we got into this mess. The problems have their root in

     adbc0b6b6e57c11ca49779d01f549260a920a97d

Cygwin's entire goal is a completely POSIX compliant environment running under Windows. The above commit circumvents some of Cygwin's API regarding stat/fstat to make things perhaps a bit faster, and definitely not POSIX compliant (The commit message is wrong, the commit definitely breaks POSIX compliance). That code is also what will not compile on different w32api versions. It is curious: the Cygwin mailing list has been absolutely silent since the w32api change was introduced last summer, this is the only piece of code I am aware of that was broken by the new headers, and of course the purpose of this code is to circumvent the Cygwin API (and by extension, Cygwin project goals).

So, perhaps a better path forward is to disable / remove the above code by default. (Those wanting a native Win32 git should just use the native Win32 git).

Mark