Building with MinGW?
raffaello.giulietti at gmail.com
Fri Apr 23 08:42:20 PDT 2010
Let me put things in perspective.
I'm not interested in building OpenJDK7 per se. I would use the binary
snapshots, were it not for the fact that, for my purposes, I need the
latest extensions provided by the MLVM project. Unfortunately, there is
no binary snapshot for that, so I need to download the Mercurial
repository, apply the MLVM specific patches and build it.
Now, I invested two frustrating days in trying to build the "pure"
OpenJDK7, i.e., without the MLVM extensions. I did it according to the
details described in the quite complete "OpenJDK Build README" page. So
I used the expected licensed VisualStudio compiler. The problems I
encountered can be generally grouped in the "path not found" category,
be it because of spaces in the path, because of \ versus /, etc. As a
consequence, I didn't even try a build with the MLVM extensions.
To be clear, I'm not complaining about the README or the like. I'm only
reporting my experience with such a complex system and its build.
So, the real reason behind my request for a MinGW based build is that it
would be a second chance to try a build of the MLVM. But since nobody
seems to have first-hand experience with OpenJDK7/MinGW, I'll gather my
energies and my patience and retry with VisualStudio.
On 2010-04-21 18:40, Kelly O'Hair wrote:
> On Apr 21, 2010, at 5:58 AM, Raffaello Giulietti wrote:
>> I'm wondering if anybody has already tried to build OpenJDK7 on Windows
>> using the MinGW suite.
> If they have, I never heard from them.
>> * Is there anything known to be a hard to circumvent show stopper?
> To me the basic problem is that with "Windows" it is hard to separate
> the code
> dependencies on the OS, some Windows SDK, something specific to Visual
> etc. I'm not saying it would be impossible, but it is not a simple
> change and
> parts of the jdk might be very difficult to disconnect from Visual Studio
> dependencies. The code has assumed Visual Studio for a long long time.
> If someone did it, and we were able to build either way, and the changes
> too outrageous, I'm sure we consider accepting that contribution.
> But I just don't think it will be that simple.
>> * Is it known why Visual C++ is still the reference build system on
> It was probably chosen as the defacto standard on Windows a long time
> ago and
> there was never any value in changing that.
> The performance was probably a key issue, and whether or not you could
> to a different compiler set, before the official builds would ever
> change you
> would need some very detailed performance measurements to verify no loss of
> performance. That's not an easy job, or simple either.
> Any change to the compilers used to create the binary JDKs we distribute
> is always
> a change made very carefully. It might provide significant benefits, but
> hidden dangers are often difficult to find and diagnose.
> I know this binary distribution model is of less interest to some who
> just want
> to build the openjdk source for a particular platform, but it certainly
> is a
> critical issue for us. Compiler changes are carefully tracked.
More information about the build-dev