RFR JDK-8005120

[Chris Hegarty]

On 20/12/2012 13:49, John Zavgren wrote:
> Greetings:
>
> I agree that the "correct" way to fix this problem is to use POSIX data types, e.g., socklen_t. However, when I switch to the doctrinaire data type, the build fails on windows machines:
> ------------- build monologue -----
> c:\jprt\t\p1\032220.jzavgren\s\jdk\src\share\transport\socket\sysSocket.h(39) : error C2146: syntax error : missing ')' before identifier 'len'
> c:\jprt\t\p1\032220.jzavgren\s\jdk\src\share\transport\socket\sysSocket.h(39) : error C2081: 'socklen_t' : name in formal parameter list illegal
> c:\jprt\t\p1\032220.jzavgren\s\jdk\src\share\transport\socket\sysSocket.h(39) : error C2061: syntax error : identifier 'len'
> c:\jprt\t\p1\032220.jzavgren\s\jdk\src\share\transport\socket\sysSocket.h(39) : error C2059: syntax error : ';'
> c:\jprt\t\p1\032220.jzavgren\s\jdk\src\share\transport\socket\sysSocket.h(39) : error C2059: syntax error : ')'
> ....
> ------------- build monologue -----
>
> I used alternative types, e.g., uint32_t, etc. as a way to avoid the limitations of windows.
> What is the recommended way to accommodate this windows limitation? Shall I use a typedef statement to define "socklen_t"?

We don't suffer from this issue in the networking native code. The unix 
and windows implementations are distinct.

I see the vm defines socklen_t in a windows specific header, 
hotspot/src/os/windows/vm/jvm_windows.h, as
    typedef int socklen_t;

  ...and it is then used in shared code, like jvm.cpp, and the hpi, by 
optionally including

    #ifdef TARGET_OS_FAMILY_windows
    # include "jvm_windows.h"
    #endif

We could use a similar, but more simplistic, approach here.

-Chris.

>
> Thanks!
>
>
> ----- Original Message -----
> From: chris.hegarty at oracle.com
> To: david.holmes at oracle.com
> Cc: Alan.Bateman at oracle.com, serviceability-dev at openjdk.java.net, john.zavgren at oracle.com, net-dev at openjdk.java.net
> Sent: Thursday, December 20, 2012 7:41:07 AM GMT -05:00 US/Canada Eastern
> Subject: Re: RFR JDK-8005120
>
> On 19/12/2012 20:52, David Holmes wrote:
>> Real sense of deja-vu here. Didn't we go through this same thing with
>> the HPI socket routines?
>
> Yes, and the networking native code too.
>
> I think it is best to use socklen_t for the unix code. From what I can
> see making these changes, to use socklen_t, should be relatively localized.
>
> -Chris.
>
>>
>> Depending on the OS (and version?) we should be using socklen_t not int
>> and not uint32_t.
>>
>> David
>>
>> On 20/12/2012 2:35 AM, Chris Hegarty wrote:
>>> John,
>>>
>>> I grabbed your patch, and with it I now see different warnings.
>>>
>>> ../../../../src/share/transport/socket/socketTransport.c: In function
>>> 'socketTransport_startListening':
>>> ../../../../src/share/transport/socket/socketTransport.c:310:40:
>>> warning: pointer targets in passing argument 3 of 'dbgsysGetSocketName'
>>> differ in signedness [-Wpointer-sign]
>>> ../../../../src/share/transport/socket/sysSocket.h:58:5: note: expected
>>> 'uint32_t *' but argument is of type 'int *'
>>> ../../../../src/share/transport/socket/socketTransport.c: In function
>>> 'socketTransport_accept':
>>> ../../../../src/share/transport/socket/socketTransport.c:371:33:
>>> warning: pointer targets in passing argument 3 of 'dbgsysAccept' differ
>>> in signedness [-Wpointer-sign]
>>> ../../../../src/share/transport/socket/sysSocket.h:41:5: note: expected
>>> 'uint32_t *' but argument is of type 'int *'
>>>
>>> Do you see these in your build?
>>>
>>> -Chris.
>>>
>>> On 12/19/2012 03:42 PM, Alan Bateman wrote:
>>>>
>>>> John - this is the debugger socket transport so cc'ing the
>>>> serviceability-dev list as that is where this code is maintained.
>>>>
>>>> On 19/12/2012 15:36, John Zavgren wrote:
>>>>> Greetings:
>>>>> Please consider the following change to the two files:
>>>>> src/share/transport/socket/sysSocket.h
>>>>> src/solaris/transport/socket/socket_md.c
>>>>> that eliminate compiler warnings that stem from the fact that the
>>>>> variables that the native code passes to various system calls were not
>>>>> declared correctly. They were declared as integers, but they must be
>>>>> "unsigned" integers because they are used to define buffer lengths.
>>>>> Were one to supply a negative value as an argument, it would be cast
>>>>> into an unsigned "Martian" value and there'd be (hopefully) a system
>>>>> call error.
>>>>>
>>>>> Thanks!
>>>>> John Zavgren
>>>>>
>>>>> http://cr.openjdk.java.net/~mullan/webrevs/jzavgren/8005120/
>>>>