Add methods for setting file timestamps

[TadCordle]

Adds overloads for LayerConfiguration.Builder#addEntryRecursive that take providers for file permissions/file modification timestamps to apply permissions and timestamps on a per-file basis.

Fixes #1607.

[chanseokoh]

This looks good to me.

One thing for a thought: previously the file timestamp was part of LayerConfiguration, which makes sense as it is a property of a file like permissions which is still in LayerConfiguration. Now the timestamp provider is set in BuildConfiguration. I think this is fine, but I wonder if others have different opinions.

[briandealwis]

Thanks for prototyping this. It seems to a bit more complicated than I imagined. Here are my thoughts:

• I think we should probably make the file-permissions configurable too (so the below should extend to FilePermissions)
• I think we want to use a provider rather than single value, so Function<Path,Instant>; we should probably have this on the Jib class so that it's easy to reference (class Jib { static Function<Path,Long> defaultFileModificationTimeProvider = path -> 1000; })
  • if someone wants to use real timestamps set to current then they can just use p -> Files.getLastModifiedTime(p)
• LayerConfiguration.Builder would require a file modification time provider to be used for subsequent calls to addEntry without an explicit file modification time
• LayerConfiguration.builder() would use the our default epoch+1s provider
• JibContainerBuilder and JavaContainerBuilder would take and cascade a provider down — so LayerConfiguration.builder() would just be a convenience for Jib Core users, whereas our plugins should always explicitly pass down the provider

[chanseokoh]

I think we want to use a provider rather than single value, so Function<Path,Instant>

The PR uses a provider (effectively Function<AbsoluteUnixPath, Instant), if you haven't noticed.

[briandealwis]

Weird — I must have somehow looking from an earlier commit. This current state looks way better. But I think we should use this provider to provide defaults which can still be explicitly over-ridden in addEntry(...).

[TadCordle]

If we do allow setting a timestamp provider on a per-LayerEntry basis, I wonder if we even need the AbsoluteUnixPath argument in the provider anymore?

Edit: Actually, source path might be more useful for something like git commit history / getting the actual mod time of the file.

[TadCordle]

Sorry to keep changing this so much. Trying this now:

• addEntry may take a single FilePermissions and Instant
• addEntryRecursive may take a Function BiFunction that determines the permissions/timestamp on a per-file basis
• Also moved the default FilePermissions stuff out of LayerEntry, which meant changing the constructor, which meant changing a bunch of tests.

[chanseokoh]

Generally looks good to me.