Issues with building on Windows

[phryneas]

Hi there,
I'm trying to use this library on Windows using the msys2 mingw64.

First of all: if you want to try this yourself, you have to install the mingw gpgme library using pacman -S mingw-w64-x86_64-gpgme and you have to do so in the mingw64 environment.

Now the problems:
For one, it complains when building the unsetenvGPGAgentInfo function:

% go run main.go                                                                                                                                                             # github.com/proglottis/gpgme
could not determine kind of name for C.unsetenv
[1]    13260 exit 2     go run src/main/hostApp.go

when I just comment this line out (as I understand the comment, it is just used for testing?), I get this error:

# github.com/proglottis/gpgme
src\github.com\proglottis\gpgme\data.go:184: cannot use C.off_t(offset) (type C.off_t) as type C.gpgme_off_t in argument to func literal

changing the offending line to
n, err := C.gpgme_data_seek(d.dh, C.gpgme_off_t(offset), C.int(whence))
let's me compile everything.

I still haven't come around to really using the library yet, that will be my next step. I just thought I'd report what I've encountered so far.

Thanks for the wrapper, it's really appreciated.

[proglottis]

Thanks for the report. I've never tried compiling this on windows. So I suspect you are paving the way.

Couple of things, unsetenv is the implementation for unsetenvGPGAgentInfo and is only actually used for tests to override GPG_AGENT_INFO. Hopefully there's an alternate way of doing this.

Second, off_t is part of the posix C std. So it's probably missing in windows? I suspect gpgme defines its own in windows for this reason. Happy with that patch, but there're a couple of other usages of off_t that should probably get converted too. It might require a little C wrapper if CGO complains too much. I'm happy to test any patches on my end

[phryneas]

Seems like that is the only instance of off_t preventing me from compiling it, which is quite weird.

But for now, I also was compiling against the gpgme v1.6.1 which is included with msys2, not against the version shipped with the GPG 2.1 installer (which I guess might be around 1.9).
So right now, it's not working with my agent and I'll have to try compiling against the other one. I guess everything will be different there, so we'll see ;)

[proglottis]

I don't think CGO knows about typedefs in C. I pushed a quick little patch so it can compile in both windows and linux/osx. https://github.com/proglottis/gpgme/tree/windows_data_seek

[phryneas]

I'll try that once my libraries finish compiling :)

On the unsetenv issue: that seems to be a mingw issue.
It is documented on https://www.gnu.org/software/autoconf/manual/autoconf-2.64/html_node/Function-Portability.html

unsetenv
On MinGW, unsetenv is not available, but a variable ‘FOO’ can be removed with a call putenv ("FOO="), as described under putenv above.

[phryneas]

just as a quick feedback: your patch seems to compile fine for me

[phryneas]

Okay, just got it to compile against the official downloadable GnuPG binaries (>2.1).

I am using vanilla MinGW now, that requires an additional change in go_gpgme.h:

--- a/go_gpgme.h
+++ b/go_gpgme.h
@@ -8,7 +8,7 @@

 extern ssize_t gogpgme_readfunc(void *handle, void *buffer, size_t size);
 extern ssize_t gogpgme_writefunc(void *handle, void *buffer, size_t size);
-extern off_t gogpgme_seekfunc(void *handle, off_t offset, int whence);
+extern gpgme_off_t gogpgme_seekfunc(void *handle, gpgme_off_t offset, int whence);
 extern gpgme_error_t gogpgme_passfunc(void *hook, char *uid_hint, char *passphrase_info, int prev_was_bad, int fd);
 extern gpgme_error_t gogpgme_data_new_from_cbs(gpgme_data_t *dh, gpgme_data_cbs_t cbs, uintptr_t handle);
 extern void gogpgme_set_passphrase_cb(gpgme_ctx_t ctx, gpgme_passphrase_cb_t cb, uintptr_t handle);

Compiling can be done with this batch file - documenting this here in case anyone else wants to do this:

REM ./include should contain header files from C:\Program Files (x86)\GnuPG\include
REM ./lib should contain files from C:\Program Files (x86)\GnuPG\lib, with extension .imp changed to .a (libgpg-err.a, libassuan.a, libgpgme.a)
REM ./ should contain dll files from C:\Program Files (x86)\GnuPG\bin (libassuan-0.dll  libgpg-error-0.dll  libgpgme-11.dll)

set PATH=%PATH%;C:\Go\bin
set PATH=%PATH%;C:\mingw\mingw32\bin
set GOPATH=%~dp0

set CGO_CFLAGS=-I%~dp0\include
set CGO_LDFLAGS=-L%~dp0\lib

go build yourfile.go

[phryneas]

in case my change does not make real sense for you, here is the original error message:

# github.com/proglottis/gpgme
In file included from $WORK\github.com\proglottis\gpgme\_obj\_cgo_export.c:3:0:
cgo-gcc-export-header-prolog:46:20: error: conflicting types for 'gogpgme_seekfunc'
In file included from src\github.com\proglottis\gpgme\data.go:6:0,
                 from $WORK\github.com\proglottis\gpgme\_obj\_cgo_export.c:3:
src\github.com\proglottis\gpgme/go_gpgme.h:11:14: note: previous declaration of 'gogpgme_seekfunc' was here
 extern off_t gogpgme_seekfunc(void *handle, off_t offset, int whence);
              ^~~~~~~~~~~~~~~~
C:\Users\WEBER~1.VOS\AppData\Local\Temp\go-build897854538\github.com\proglottis\gpgme\_obj\_cgo_export.c:59:13: error: conflicting types for 'gogpgme_seekfunc'
 gpgme_off_t gogpgme_seekfunc(void* p0, gpgme_off_t p1, int p2)
             ^~~~~~~~~~~~~~~~
In file included from src\github.com\proglottis\gpgme\data.go:6:0,
                 from $WORK\github.com\proglottis\gpgme\_obj\_cgo_export.c:3:
src\github.com\proglottis\gpgme/go_gpgme.h:11:14: note: previous declaration of 'gogpgme_seekfunc' was here
 extern off_t gogpgme_seekfunc(void *handle, off_t offset, int whence);
              ^~~~~~~~~~~~~~~~

[phryneas]

Sorry, that must have been some old version I had checked out in-between. Did a clean checkout on another PC and everything works.. :)

[phryneas]

Okay,
to make everything reproducible, I've set up a CI that automatically builds proglottis/gpgme from your repo and runs the test.
https://github.com/phryneas/go_gpgme_build_windows
you can see the latest build & test results here:
https://ci.appveyor.com/project/phryneas/go-gpgme-build-windows

There are still some tests that are failing and I'm not really sure what's happening there. It would be great if you could have a look.
Most things I've tried to use so far worked for me though.

[proglottis]

Hey that's really cool! I didn't think any of those CI places would let me have gpgme.

Could be coincidence but both those tests are using temp dir/file. I wonder if CI is a bit unhappy about that. The only other test to use tmp is AssuanSend but it doesn't make GPG write to it

[phryneas]

I guess you'll be getting a pull request that runs the tests on windows with appveyor and on linux with travis from me ;)

As for the failing tests: that reflects what I'm getting locally, so it doesn't seem to be a CI thing. Even worse, the TestContext_Import sometimes just hangs until I manually quit it.

[wagnerma]

Hi, I also try to use this library under Windows. I am a newbie in programing with go, so perhaps my question is silly ;-)
I installed go for windows and also gnupg-w32. All is usable from the command line. I copied gpgme.h and gpg-error.h from the gnupg installation directory to the build directory.
Then I try to prepare the library with:
go get -u github.com/proglottis/gpgme
I get the same error mentioned above:

could not determine kind of name for C.unsetenv

Should be the library useable under Windows? Is this still a problem? I am missing something?
Thanks for your help and also for the work!
Cheers,
Marcel

[phryneas]

just replace the C.unsetenv(v) in gpgme.go with C.putenv(v) - on MinGW, unsetenv is not available.

PS: oh, and v has to be GPG_AGENT_INFO= instead of GPG_AGENT_INFO

see the part about unsetenv in https://www.gnu.org/software/autoconf/manual/autoconf-2.64/html_node/Function-Portability.html