Hotspot makefile conversion now in build-infra

[Volker Simonis]

Hi Magnus,

while I think this is a good step, I still have some questions:

- is this planned for jdk9 only or will it be backported to jdk8?
- will you replace the Windows nmake-files by GNU makefiles for the
Windows build (I would appreciate if yes and I can say that we' (SAP)
have already done this for the current build)?
- against which HotSpot repository will you synchronize? Is it
hsx/hotspot-main and will it stay the same after jdk9 was branched or
do you synchronize against jdk8 now (and jdk9 later) and just use the
hotspot from there? From my point of view it would be desirable to be
as close as possible to the HotSpot development repository if you want
to change the HotSpot build system which would probably mean that you
should synchronize with  hsx/hotspot-main.
 - what do you think would be the right point in time for me to jump
in with our platform zoo (AIX, HPUX and especially the windows IA64
cross build) ?

Thank you and best regards,
Volker


On Thu, Dec 5, 2013 at 11:49 AM, Magnus Ihse Bursie
<magnus.ihse.bursie at oracle.com> wrote:
> We have just "re-opened" the build-infra project (repos and mailing list)
> for the continuation of the conversion of the old build system.
>
> "Hey, wasn't that finished already?" you might be thinking. Unfortunately,
> no. Due to constraints, we left out a major component in the conversion --
> hotspot.
>
> I and Erik have been working on a prototype for building Hotspot using the
> new build system. It has been developed in a private repository, but is now
> ready to be moved to build-infra.
>
> You are of course welcome to try it out, but unless you're eager, it is
> probably a bit too early. :-) The current prototype has the following
> limitations/properties:
> * It has only been tested (and probably only work) on linux-x86_64.
> * It creates roughly the correct collection of files in the
> build/hotspot/dist directory.
> * It compiles roughly the correct set of native files, with roughly the
> correct set of compiler flags.
> * It creates jar files with roughly the correct contents.
>
> However, the compare script clearly shows that "roughly" does still not mean
> "identical". :-)
>
> Going forward, we need to add support for all platforms, and make sure that
> the built bits indeed end up identical to the old build system. There are
> also several pieces of functionality in the current build system that needs
> to be addressed.
>
> On the bright side, we can already (with no optimizations being done
> whatsoever) see a slight speed increase (2 min 11 sec vs 2 min 35 sec on my
> machine). But most of all, we see a clear reduction of complexity of the
> build system. Compare:
>
> magnusi at ../hotspot/make$ wc `find . -type f` | grep total
>  22527  94124 803058 total
>
> with
>
> magnusi at .../hotspot/makefiles$ wc `find . -type f` | grep total
>   753  3686 32369 total
>
> While this will surely double in size before we're done, it's still an order
> of magnitude simpler.
>
> /Magnus