Here's what I/we found when running "jmod hash".

As background:

    jdk.jdeps requires jdk.compiler
    jdk.compiler requires java.compiler

Running this:

% cd build/windows*/images/jmods
% jmod.exe hash --mp . --hash-modules \
    "jdk.jdeps|jdk.compiler|java.compiler"

updates two files:

jdk.compiler.jmod
    adds "hash jdk.jdeps"
    updates timestamp (zip file offset 10-14)

java.compiler.jmod
    adds "hash jdk.[compiler,jdeps]"
    updates timestamp (zip file offset 10-14)

But note that when the hash for jdk.compiler.jmod was calculated to add to java.compiler.jmod, it used the original file/internal timestamp, not the updated/soon-to-be-written one. So if you now try to run jlink, the stored java.compiler signature for jdk.compiler.jmod won't match.

So my question is:

Mandy wrote:
> But I would expect only java.compiler will record the hashes.

Was jdk.compiler.jmod also supposed to be updated in the "jmod hash" above?

If no, you have a bug because it is being updated.
If yes, you have a bug in that you're calculating jdk.compiler pre-update.

I'm just not sure which one is the expected behavior.