Add FFI::LastError.win_error

[graywolf]

Under cygwin both errno and GetLastError are needed. This
adds second getter named #win_error:

Linux
    FFI::LastError.error     == errno
    FFI::LastError.win_error == NoMethodError
Windows
    FFI::LastError.error     == GetLastError
    FFI::LastError.win_error == GetLastError
Cygwin
    FFI::LastError.error     == errno
    FFI::LastError.win_error == GetLastError

Fixed #632

[larskanis]

Makes sense and LGTM. However I would call the new accessor winapi_error to be more precise.

[larskanis]

Why do you declare these functions? This shouldn't be necessary and breaks the MINGW build: https://ci.appveyor.com/project/larskanis/ffi-aofqa/build/1.0.30

[graywolf]

I'm stupid, this should obviously be only on cygwin, fixed

[graywolf]

I don't have strong opinion on the name either way, so I've renamed it.

[larskanis]

LGTM. @tduehr OK for you as well? Although JRuby doesn't work on Cygwin, winapi_error should possibly be added to JRuby later on for API compatibility on Windows.

[tduehr]

Looks pretty good. I'm not sure about the Linux win_error raising but Windows error not... Not sure what the Windows expectation is here as I work in *nix almost exclusively.

Between api compat and the Linux subsystem in win10, this will need to make it into JRuby as well.

[larskanis]

@tduehr It would have been better to not unify errno and GetLastError to error. It seemed logical to me as well - until I though about Cygwin. But now it's there and to keep backward compatibility, I think it's best solved like @graywolf proposed.

[tduehr]

Would it be better to leave error alone, and add last_error and errno to split the two interfaces?

[larskanis]

@tduehr Makes sense. But last_error is too imprecise. I would call it winapi_error or maybe GetLastError like the C function.

[tduehr]

Yeah... I think there may be no proper (Ruby-like) solution here... We'll also need to make similar changes to FFI::Errno. Let's go with GetLastError as it'd be more immediately recognized by windows devs.

[graywolf]

So... should I change to pull request to add LastError#GetLastError? GetLastError is not very ruby-like name for method, so get_last_error?

[damian-m-g]

Hello guys, any news on this? I've checked that under Windows (MinGW), FFI.errno (nor FFI::LastError.error) calls aren't accurate.

UPDATE:

FYI, I've made a workaround from my C code to bypass this problem:

// use SET_ERRNO(x) instead of errno=(x) to set errno
#ifdef __MINGW32__
  #include <afxres.h>
  #define SET_ERRNO(x) SetLastError(x)
#else
  #define SET_ERRNO(x) errno=(x)
#endif

Using that code, from Ruby FFI, I successfuly retrieve the correct errno.

Still, would be nice to have FFI handle this problem (especially for cases where you're using C third hand libs), and I'm not sure how possible this is, so far this thread looked promising.

[larskanis]

To finally solve this issue I merged it as proposed by @graywolf . It is now released as ffi-1.10.0.

I would greatly appreciate if someone can take the time to write some simple tests for the new call FFI::LastError.win_error and FFI::LastError.win_error=.

[graywolf]

@IgorJorobus could you try to take a look and write some tests as @larskanis asks? Sadly I no longer have a system with cygwin readily available.

[damian-m-g]

Hi guys, I hope that this specs cut the cake for you. Honestly, I couldn't run the them, just coded them. The rake tasks ask for several things that are missing in my system, isn't enough to just clone the repo, I've tried to fulfill rake needs but couldn't during the time I set for this task. Let me know if I can do something else for you.