RFR 9/8: 8066504: GetVersionEx in java.base/windows/native/libjava/java_props_md.c might not get correct Windows version

[Ivan Gerasimov]

Hi Roger!

On 17.06.2015 17:27, Roger Riggs wrote:
> I updated the webrev to remove the Windows 3.1 case and
> will wait 24hrs to see if there are more comments.
>

I have a little concern.
You're using char-agnostic versions of functions GetSystemDirectory(), 
GetFileVersionInfoSize(), GetFileVersionInfo() with a plain char buffer.
I think it would be better to either declare kernel32_path TCHAR[] or 
explicitly use ANSI versions of the functions.

Personally, I'd prefer if kernel32_path were WCHAR and Unicode versions 
of the functions were used.
I don't know if it's allowed to have non-ASCII chars in the SysDirectory 
path, but it seems reasonable to assume it is.

It would be interesting to setup Windows with non-English name of the 
system directory and see what's starting to fail :)
I'm going to do that as my spare time.

Sincerely yours,
Ivan


> Thanks for the review, Roger
>
>
> On 6/17/2015 9:58 AM, Alan Bateman wrote:
>> On 17/06/2015 14:49, Roger Riggs wrote:
>>> Hi Alan,
>>>
>>> Would there be any concerns about backporting to JDK 8?
>> I don't think so but it will require the jdk8u maintainers to approve.
>>>
>>> The new helper APIs 
>>> <https://msdn.microsoft.com/en-us/library/windows/desktop/ms724451%28v=vs.85%29.aspx> 
>>> only verify that the OS is at least the version requested.
>>> It does not reveal what version the OS is actually running on.
>>> Adding a manifest attribute just declares the minimum version required
>>> and is reported by GetVersionEx.
>>> In the current case, the code uses GetVersionEx as a fallback in
>>> case the version on kernel32.dll is not available.
>>> That fallback could be removed since kernel32.dll must exist on all 
>>> supported versions.
>> Okay, I thought the helper APIs had better support than this. Maybe 
>> the Microsoft folks that are contributing to OpenJDK might have 
>> suggestions. In general I don't have any objections to the approach, 
>> just thought it was a big odd that the OS forces us to look at the 
>> version of kernel32.dll.
>>
>>
>>> :
>>>
>>>>
>>>> Also, is there any update needed in 
>>>> src/windows/resource/java.manifest?
>>> If using the version of kernel32.dll then the manifest version does 
>>> not affect the value of os.name.
>>> But yes, I added the Windows 10 supportedOS uid.
>> I don't think this was in the first webrev that I saw but I see it is 
>> there now, looks fine.
>>
>>
>>>>
>>>> In passing, GetNativeSystemInfo has been in Windows since Windows 
>>>> XP so I don't think we need to use GetModuleHandle to get the 
>>>> address now.
>>> ok.
>>>
>>> Similarly, can the case for Windows 3.1 be removed?   (its not in 
>>> theJDK 8 supported matrix 
>>> <http://www.oracle.com/technetwork/java/javase/certconfig-2095354.html>)
>>> caseVER_PLATFORM_WIN32s:
>> I think that would be a fine cleanup.
>>
>> -Alan.
>
>
>