Commits on Jun 14, 2022

1. Have image.FromUnparsedImage return a SourcedImage instead of types.I… …

   …mage
   
   This will allow us to add more methods to SourcedImage without breaking
   external implementations of types.Image, if any.
   
   Should not change behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   0cf416b View commit details

   Browse the repository at this point in the history

2. Make SourcedImage.{ManifestBlob,ManifestMIMEType} public …

   We know these values are available without having to worry about
   errors, which allows simplifying users in c/image/copy.  We don't really
   need accessor functions, all users are internal (and a public field
   is actually better protected than a public method, because a public
   method can be used by third parties through a caller-defined interface).
   
   It's also usefo to make them available as individual values; various
   users only need one or the other.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   2874e99 View commit details

   Browse the repository at this point in the history

3. Introduce manifestConversionPlan …

   To make determineManifestConversion easier to test,
   we'll want to separate it from imageCopier and make it a pure function.
   
   To do that, start with introducing a struct for its return value;
   that way we won't need three unnamed return values, and three
   variables with long names at the call site.
   
   Should not change behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   cba0127 View commit details

   Browse the repository at this point in the history

4. Move the manifest conversion setup out of determineManifestConversion …

   Have determineManifestConversion just set a flag in the returned plan;
   then copier.copyOneImage will actually act on that flag.
   
   Should not change behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   225269f View commit details

   Browse the repository at this point in the history

5. Use manifestConversionPlan for expected results in TestDetermineManif… …

   …estConversion
   
   This is not smaller, but a bit more readable and consistent
   with the real call site.
   
   Also, on test failures we can see the full returned plan as a single unit,
   making it possibly easier to see what went wrong.
   
   Should not change (test) behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   e117048 View commit details

   Browse the repository at this point in the history

6. Introduce determineManifestConversionInputs …

   ... and split the actual logic (determineManifestConversion)
   from imageCopier (imageCopier.determineManifestConversion).
   
   determineManifestConversion is now a pure function (well,
   apart from logging).
   
   Should not change behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   f395eeb View commit details

   Browse the repository at this point in the history

7. Mostly test the pure function in TestDetermineManifestConversion …

   Keep the "reading manifest failed" case, for now.
   
   Should not change (test) behavior, except that the test coverage
   is now a tiny bit smaller - but then the test not depending
   on imageCopier is the primary reason for this refactoring.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   7602cbd View commit details

   Browse the repository at this point in the history

8. Record an *image.SourcedImage in imageCopier.src …

   This is surprisingly somewhat convoluted.
   
   We've had to do a bunch of work to make determineManifestConversion
   not depend on a complete (and mostly valid!) image, to be able
   to make this change.
   
   Then, so that we don't lose test coverage (and to actually benefit
   from the new private type), use SourcedImage.ManifestMIMEType
   in imageCopier.determineManifestConversion.  That allows us to get
   rid of the last imageCopier.determineManifestConversion call
   in tests.
   
   Should not change behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   cce475b View commit details

   Browse the repository at this point in the history

9. Use StoredImage.StoredManifest() where appropriate …

   This eliminates some unreachable error paths.
   
   Should not change behavior.
   
   Signed-off-by: Miloslav Trmač <mitr@redhat.com>

   

   mtrmac committed Jun 14, 2022
   Copy the full SHA

   f074600 View commit details

   Browse the repository at this point in the history

10. Inline imageCopier.determineManifestConversion into the caller …

    It's now basically a trivial call forwarder we don't need for
    anything, so just have the caller set determineManifestConversionInputs
    directly (and let it benefit from labeling the inputs).
    
    Should not change behavior.
    
    Signed-off-by: Miloslav Trmač <mitr@redhat.com>

    

    mtrmac committed Jun 14, 2022
    Copy the full SHA

    acbde71 View commit details

    Browse the repository at this point in the history