Re: SableVM SDK - General Info and Gentoo Update

On Tue, May 31, 2005 at 09:12:42AM -0400, Grzegorz B. Prokopski wrote:
> On Sun, 2005-29-05 at 16:02 -0400, David Bélanger wrote:
> > Hi,
> > 
> > I completed a Gentoo ebuild (package build script) that builds
> > everything for SDK 1.12_pre2 and everything seems to work though
> > not tested a lot.
> 
> Great, thanks!
> 
> > Things of general interests to non-Gentoo people:
> > 
> > assert patch: removes assert statements allowing builds on pre-SDK-1.5.
> 
> Weird.  Is your version of jikes recent enough?  Did you try
> --with-jikes=internal ?  I've not noticed any problems.  And assert IIRC
> was introduced somewere in 1.3.
> 

I'm using Gentoo jikes (not the internal one) version 1.22.
I will try with the internal one.

> > bootstrap patch: ensure that SableVM SDK is used to build cp-tools,
> > gjdoc, etc. by setting JAVAC, JAR, JAVA env variables.  Ideally, we
> > might want to have configure options to control this.  On Gentoo
> > systems, JAVAC (and a few others) variable are defined to the current VM
> > SDK selected.  Therefore, cp-tools/gjdoc etc. are build with currently
> > selected system VM instead of SableVM SDK.
> 
> Maybe we should just unset JAVAC/JAR/JAVA in Makefiles, like it's done
> with MAKEFLAGS?
> 
> > I think it should default to SableVM SDK unless otherwise specified with
> > configure flags i.e. the SableVM SDK should always bootstrap itself.
> > I don't think it should rely on pre-set environment variables as it
> > is currently the case.
> 
> The idea was that if there's no installation of jikes or fastjar found
> (NOT: javac/jar!) then they will be built and installed.  But if they
> are found then these already installed ones should be used.
> 

I will think more about what to do about this one before committing
anything.

The thing is that on a Gentoo system, if IBM SDK (should be
same for Sun or other) is installed and selected, the gjdoc/cp-tools will use
the IBM SDK and not jikes since JAVAC will be set by Gentoo to the IBM
SDK javac.  If JAVAC unset should work as it should autodetect and find
jikes first (need to test it).  I am setting it to the wrapper script
which, at least in the case of jikes, works even if the system installed
jikes is used.

> > chmod patch: fixes earlier bug reported.  When autoconf tools rerun and
> > recreate files - permissions were not always set for BUILDIR/bin.
> 
> Hope it's commited now.
> 

Will do it shortly.


> > dir patch: One way to fix DESTDIR bugs (undo Greg's Makefile.am changes
> > and uses config options).  An alternative way is to keep Greg's changes
> > and remove DESTDIR from all bindir,libdir,...
> 
> My mistake IIRC.  I thought the DESTDIR was missing, so I added it, but
> apparently it was all right.  So undoing my change seems like the right
> thing to do.
> 

I committed a fix yesterday - not the same patch.  I basically fixed the
DESTDIR in the Makefile.am and left the bindir,libdir there.

> > Note: fastjar will not rebuild automatically Makefile.in if Makefile.am
> > is built unlike all other code.  By adding the config switch:
> > --enable-maintainer-mode
> > to fastjar, it will make the build dep behaviour ok like the others.
> > This should not affect end users but is a little bit confusing to dev
> > when they patch Makefile.am and expect Makefile.in to be regenerated
> > like the others. :)
> 
> IIRC we could either remove the MAINTAINER_MODE macro from configure or
> add --enable-maintainer-mode to the flags passed to fastjar's configure
> (you can do that in top-level configure).  I am unsure what's the
> benefit from using MAINTAINER_MODE anyway, but it might be just my lack
> of knowledge on that topic.
> 

I committed yesterday by passing the config option.  No benefits for
end-users really as Makefile.in will be provided in releases.  For devs,
it makes the behaviour more consistent.


David

---

David Bélanger
Web page:   http://www.cs.mcgill.ca/~dbelan2/
Public key: http://www.cs.mcgill.ca/~dbelan2/public_key.txt