OpenJDK fails to build with GCC when the #include<time.h> inside zip.cpp comes from a non-sysroot path

[Patrick Zhang]

Thanks Florian, the "-isystem /usr/include" is helpful to my case, I see gcc.gnu.org says that "it gets the same special treatment that is applied to the standard system directories". As such the issue gets hidden (error suppressed).

Hi David,
Thanks for your suggestion. My intention was to limit the influence range as far as I could since I don't have other systems except CentOS/Fedora to verify (even just smoke test) all paths.
In addition, if we make below update, does it mean the macro " _REENTRANT " can be removed too? This is probably the only place where _REENTRANT gets used AFAIK.
#ifdef _MSC_VER // Windows
#define gmtime_r(t, s) gmtime(t)
#endif

Regards
Patrick

-----Original Message-----
From: Florian Weimer <fweimer at redhat.com> 
Sent: Thursday, November 29, 2018 8:02 PM
To: David Holmes <david.holmes at oracle.com>
Cc: Patrick Zhang <patrick at os.amperecomputing.com>; jdk-dev at openjdk.java.net; core-libs-dev at openjdk.java.net
Subject: Re: OpenJDK fails to build with GCC when the #include<time.h> inside zip.cpp comes from a non-sysroot path

* David Holmes:

> This should really be being discussed on core-libs-dev.

Okay, moving the conversation.

>> diff -r 70a423caee44 src/share/native/com/sun/java/util/jar/pack/zip.cpp
>> --- a/src/share/native/com/sun/java/util/jar/pack/zip.cpp	Tue Oct 09 08:33:33 2018 +0100
>> +++ b/src/share/native/com/sun/java/util/jar/pack/zip.cpp	Wed Nov 28 22:13:12 2018 -0500
>> @@ -415,9 +415,7 @@
>>       ((uLong)h << 11) | ((uLong)m << 5) | ((uLong)s >> 1);
>>   }
>>   -#ifdef _REENTRANT // solaris
>> -extern "C" struct tm *gmtime_r(const time_t *, struct tm *); -#else
>> +#if !defined(_REENTRANT) // linux
>>   #define gmtime_r(t, s) gmtime(t)
>>   #endif
>>   /*
>
> Under the theme "two wrongs don't make a right" the use of _REENTRANT 
> here is totally inappropriate AFAICS. It seems to be a misguided 
> attempt at determining whether we need the thread-safe gmtime_r or not
> - and the answer to that should always be yes IMHO.
>
> We define _REENTRANT for:
> - linux
> - gcc
> - xlc
>
> So the original code will define:
>
> extern "C" struct tm *gmtime_r(const time_t *, struct tm *);
>
> for linux (and AIX with xlc?) but not Solaris, OS X or Windows.
>
> But Solaris has gmtime_r anyway. So the existing code seems a really 
> convoluted hack. AFAICS we have gmtime_r everywhere but Windows (where 
> gmtime is already thread-safe). So it seems to me that all we should 
> need here is:
>
> -#ifdef _REENTRANT // solaris
> -extern "C" struct tm *gmtime_r(const time_t *, struct tm *); -#else
> +#ifdef _MSC_VER // Windows
>  #define gmtime_r(t, s) gmtime(t)
>  #endif

That looks much cleaner.

Thanks,
Florian