RFR: 8352689: Allow for hash sum overrides when linking from the run-time image [v2]

[Severin Gehwolf]

On Wed, 26 Mar 2025 14:50:11 GMT, Alan Bateman <alanb at openjdk.org> wrote:

>>> > The cacerts issue mentioned in the JBS issue relates to an RPM installation of the JDK where the cacerts file is a symlink to the distro provided one. So I think that's "use system" issue.
>>> > TZ updates would potentially break this too. If an external tool updates `tzdb.dat` then the hash sum computed at JDK build-time will no longer match. I believe this could also be solved with a sha-override (e.g. by coming from a file `@${java.home}/sha-override.txt`) which would get the update along with `tzdb.dat`.
>>> 
>>> I think one direction to explore is configuring which files or directory are "upgradable". Upgradable modules aren't excluded from the hash computation to allow upgrade via the upgrade module path. Something similar here would allow jlink to report that tzdb.dat has been upgraded. Maybe cacerts is the same but need to look closer as the hash computation shouldn't be following sym links outside of the runtime image.
>> 
>> I'll keep looking into this specific case. However, it sounds a bit orthogonal to the patch at hand which I do believe we still need for the original reasons mentioned (RPM changing binaries after the JDK build is long done and the windows issue of the JDK build itself placing different *.pdb files into the image than was present at jlink time). So perhaps we should explore this in parallel?
>
>> I'll keep looking into this specific case. However, it sounds a bit orthogonal to the patch at hand which I do believe we still need for the original reasons mentioned (RPM changing binaries after the JDK build is long done and the windows issue of the JDK build itself placing different *.pdb files into the image than was present at jlink time). So perhaps we should explore this in parallel?
> 
> I think upgradable files is something we can deal with. I'm not sure yet on the PDB issue, need to think more about about the scenarios to see what might make sense.

@AlanBateman Any more thoughts on this? We'd need to include a patch like this one for getting the Fedora JDK 24+ builds to work with JEP 493 enabled. Thanks!

-------------

PR Comment: https://git.openjdk.org/jdk/pull/24190#issuecomment-2768744746