[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Building for windows. Again.



The last build I know worked for lmdbjni on windows used LMDB_0.9.14.

I was going to try a patch release on LMDB_0.9.16 but the fix doesn't
merge cleanly.

On Fri, Nov 20, 2015 at 10:26 AM, Howard Chu <hyc@symas.com> wrote:
> Kristoffer Sjögren wrote:
>>
>> Yep, the commit fixes the problem. Thanks Howard!
>
>
> The only question now - what is "modified" in the Java UTF-8 encoding?
>
>>
>> On Fri, Nov 20, 2015 at 2:14 AM, Howard Chu <hyc@symas.com> wrote:
>>>
>>> Kristoffer Sjögren wrote:
>>>>
>>>>
>>>> Changing CharacterSet in the build file doesn't affect the garbled path.
>>>>
>>>> I tried to use explicit GetStringChars from the JNI code but with same
>>>> result.
>>>>
>>>> Any pointers? I'm running short of ideas..
>>>
>>>
>>>
>>> Looks like we need to use UTF-8, as in ITS#7992. I've just merged that
>>> patch
>>> into mdb.master, see if that works better. You should use
>>> GetStringUTFChars.
>>>
>>>>
>>>> On Thu, Nov 19, 2015 at 11:55 PM, Kristoffer Sjögren <stoffe@gmail.com>
>>>> wrote:
>>>>>
>>>>>
>>>>> Oh, but the build file [2] have a CharacterSet element set to Unicode.
>>>>> Let me see if I can change this.
>>>>>
>>>>> [2]
>>>>>
>>>>> https://github.com/deephacks/lmdbjni/blob/master/lmdbjni/src/main/native-package/vs2010.vcxproj
>>>>>
>>>>> On Thu, Nov 19, 2015 at 11:45 PM, Kristoffer Sjögren <stoffe@gmail.com>
>>>>> wrote:
>>>>>>
>>>>>>
>>>>>> I can see the problem now from a local test.
>>>>>>
>>>>>> Hmm i'm a bit confused. The JNI code uses GetStringUTFChars calls [1]
>>>>>> for all char * arguments going from Java to C, which is an array of
>>>>>> bytes representing the string in modified UTF-8 encoding. Is this OK?
>>>>>> The other option is to use GetStringChars which is unicode...
>>>>>>
>>>>>> [1]
>>>>>>
>>>>>> http://docs.oracle.com/javase/7/docs/technotes/guides/jni/spec/functions.html
>>>>>>
>>>>>> On Thu, Nov 19, 2015 at 10:32 PM, Howard Chu <hyc@symas.com> wrote:
>>>>>>>
>>>>>>>
>>>>>>> Kristoffer Sjögren wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> The py-lmdb project have a comment [1] about reusing Python.h
>>>>>>>> headers
>>>>>>>> which seems to work for them.
>>>>>>>>
>>>>>>>> # Microsoft Visual Studio 9 ships with neither inttypes.h, stdint.h,
>>>>>>>> or a
>>>>>>>> sane
>>>>>>>> # definition for ssize_t, so here we add lib/win32 to the search
>>>>>>>> path,
>>>>>>>> which
>>>>>>>> # contains emulation header files provided by a third party. We
>>>>>>>> force-include
>>>>>>>> # Python.h everywhere since it has a portable definition of ssize_t,
>>>>>>>> which
>>>>>>>> # inttypes.h and stdint.h lack, and to avoid having to modify the
>>>>>>>> LMDB
>>>>>>>> source
>>>>>>>> # code. Advapi32 is needed for LMDB's use of Windows security APIs.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Note in ITS#8067 we added an ssize_t typedef for MSVC. Also in #8069
>>>>>>> we
>>>>>>> changed to explicitly use the ANSI APIs, so if you're using Unicode
>>>>>>> file
>>>>>>> paths that will certainly be garbled now.
>>>>>>>
>>>>>>>>
>>>>>>>> [1] https://github.com/dw/py-lmdb/blob/master/setup.py
>>>>>>>>
>>>>>>>> On Thu, Nov 19, 2015 at 10:13 PM, Kristoffer Sjögren
>>>>>>>> <stoffe@gmail.com>
>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> The actual build command:
>>>>>>>>>
>>>>>>>>> $ cmd.exe /X /C "vcbuild /platform:x64 vs2008.vcproj release"
>>>>>>>>>
>>>>>>>>> On Thu, Nov 19, 2015 at 10:11 PM, Kristoffer Sjögren
>>>>>>>>> <stoffe@gmail.com>
>>>>>>>>> wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> That's the thing, the build doesn't complain about anything
>>>>>>>>>> missing.
>>>>>>>>>> But the binary seems broken because tests fail randomly.
>>>>>>>>>>
>>>>>>>>>> A user tried the generated binary on his machine and it worked but
>>>>>>>>>> also
>>>>>>>>>> said...
>>>>>>>>>>
>>>>>>>>>> "Hm, almost, the database file is no longer put to the requested
>>>>>>>>>> directory, and its name is totally garbled. It finds it
>>>>>>>>>> consistently
>>>>>>>>>> though, and the contents are correct."
>>>>>>>>>>
>>>>>>>>>> https://github.com/deephacks/lmdbjni/issues/48
>>>>>>>>>>
>>>>>>>>>> On Thu, Nov 19, 2015 at 9:57 PM, Howard Chu <hyc@symas.com> wrote:
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Kristoffer Sjögren wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Hi
>>>>>>>>>>>>
>>>>>>>>>>>> I'm trying to build LMDB with Java/JNI bindings with Visual C++
>>>>>>>>>>>> Project Builder 9.00.30729 (vcbuild).
>>>>>>>>>>>>
>>>>>>>>>>>> Unfortunately, vcbuild don't ship with inttypes.h, stdint.h, or
>>>>>>>>>>>> a
>>>>>>>>>>>> sane
>>>>>>>>>>>> ssize_t. So I searched around and found a few candidates [1] of
>>>>>>>>>>>> inttypes.h and stdint.h that seems to be working for py-lmdb.
>>>>>>>>>>>>
>>>>>>>>>>>> However, unistd.h seems broken for windows - and the python guys
>>>>>>>>>>>> use
>>>>>>>>>>>> python.h instead.
>>>>>>>>>>>>
>>>>>>>>>>>> Any ideas where I might find a good unistd.h?
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> I see you have one already - what is wrong with it, what
>>>>>>>>>>> definitions
>>>>>>>>>>> are
>>>>>>>>>>> missing?
>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Cheers,
>>>>>>>>>>>> -Kristoffer
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> [1]
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> https://github.com/deephacks/lmdbjni/tree/master/lmdbjni-win64/headers
>>>>>>>>>>>>
>
>
>
> --
>   -- Howard Chu
>   CTO, Symas Corp.           http://www.symas.com
>   Director, Highland Sun     http://highlandsun.com/hyc/
>   Chief Architect, OpenLDAP  http://www.openldap.org/project/