No subject

[Julian Waters]

Hi Eric,

I see, sorry for the misunderstanding. I'm not sure if there's any
practical implementation for this either, but I'll look into it more anyway

best regards,
Julian

On Sat, Mar 5, 2022 at 12:02 AM <erik.joelsson at oracle.com> wrote:

> Hello Julian,
>
> You are misunderstanding Magnus here. What he means is that we are
> compiling the version numbers into Java classes and native libraries
> (through generating Java source files with the version information in
> them and by setting -D flags when compiling C/C++). Because of this, if
> we let the version string dynamically update at make time, we would
> always have to recompile those files on incremental builds, which
> triggers quite a bit of downstream rebuilding (relinking jibjvm,
> recreating jmods and images etc). This makes the incremental build slow
> and unwieldy and would be impractical for most developers.
>
> What Magnus is suggesting is that in the final JDK image, we would have
> a text file with the version information in it, and every native lib or
> Java class in the JDK that currently has this compiled into it, would
> instead know how to reference it from this text file.
>
> It's not clear to me if this is a feasible proposal.
>
> /Erik
>
> On 2022-03-04 07:46, Julian Waters wrote:
> > Hi Magnus,
> >
> > While looking through the now resolved JDK-8274980
> > <https://bugs.openjdk.java.net/browse/JDK-8274980> one discussed
> > possibility in particular caught my eye:
> > "The one thing remaining now to make me fully happy wrt version numbers,
> is
> > if we could stop hard-coding them into files during build, and instead
> have
> > interested parties read it from a text file."
> >
> > By hardcoded files, are you referring to the files generated when
> configure
> > is run? In my opinion a separate text file wouldn't be needed if a flag
> is
> > set somewhere, and we could instead modify the generated spec.gmk to run
> > other support scripts to determine VERSION_BUILD if it hasn't been set
> yet
> > (ie shifting automated version detection to when make is called by
> changing
> > spec.gmk.in to an absolute value if --with-version-opt was passed, and
> > compile time check otherwise). perhaps by escaping bash variables into
> make
> > while it's running. What do you think?
> >
> > Thanks for your time and I'm really sorry if I'm troubling you on this,
> > have a great day! :)
> >
> > best regards,
> > Julian
>