LIBARCHNAME in CompileLauncher.gmk

[Jeremy Manson]

Not really.  We have to keep the patch to the launcher going anyway, so
having this be part of it isn't a big deal.  But it should definitely
either be corrected or removed, because if someone did depend on its
behavior being correct, then they would break 32-bit x86 builds.

(For the curious: this is a deeply uninteresting patch that would take a
couple of paragraphs to explain, so I'm not going to bother.)

Jeremy


On Mon, Nov 18, 2013 at 2:24 AM, Magnus Ihse Bursie <
magnus.ihse.bursie at oracle.com> wrote:

> On 2013-11-15 20:46, Jeremy Manson wrote:
>
>> LIBARCHNAME as defined in CompileLauncher.gmk seems to be i586 in 32-bit
>> x86 instead of being i386.  It's using the wrong make variable.
>>
>> LIBARCHNAME doesn't happen to be used in any of the launchers, so it
>> hasn't
>> caused a problem, but it should probably be either fixed or removed.
>>   Here's a patch to fix it (I'm biased towards this, because our launcher
>> changes use it, but I could certainly understand going the other way).
>>
>
> I'd really prefer to go the other way -- a define that is not used in the
> source code should not be sent in to the compiler. The list of defines tend
> to only grow and never be pruned. :-(
>
> If anything, it should probably have been $(OPENJDK_TARGET_CPU_JLI), since
> that is what LIBARCHNAME is set to when building the jli library.
>
> Would it cause a big problem for you if it was removed instead of
> redefined?
>
> /Magnus
>