do not strip "classpath:" when normalizing the path

[andrus]

Impact

This may or may not be a breaking change.

Description

This PR addresses the issue #5878 . It ensures that path "normalization" never strips the classpath: prefix when creating DatabaseChangeLog for child resources. The solution is submitted per @tati-qalified advise.

Things to be aware of

The altered method is static and is widely used across Liquibase. This may or may not be a breaking change.

Things to worry about

I don't fully understand the consequences of NOT stripping classpath: prefix when resolving relative paths for other implementations of ResourceAccessor. Quite possibly some of them will need to be updated as well. But it seems logical: in a hierarchy of resources, child resource resolution should follow the same algorithm as the root resource.

Additional Context

See #5878

[MalloD12]

Hi @andrus,

How are you doing? We very much appreciate you trying to fix #5878 and create this PR for it, but we can't merge it as it is. This PR is breaking some tests, but besides that, I have been discussing it with the team and we think this is a breaking change and we don't want to change the current behavior.

I think what we can do is define a new global configuration (you can have a look at src/main/java/liquibase/GlobalConfiguration.java to see some existent examples) and depending on the value set for this configuration then we can decide to strip/don't strip classpath: from the path. I think this way we would prevent breaking somebody else code.

If you have any questions/concerns about it please let us know and me or someone else from the team would be keen to help with you. Last thing, if you are keen to do address these modifications could you please add some unit tests to test the code changes made.

Thanks,
Daniel.

[andrus]

Hi @MalloD12 , I suspected we can't just change the policy globally. The suggested idea of doing it conditionally via GlobalConfiguration property should work. Not sure how hard it will be to write a test on the LB site, but I will definitely be happy to test it on Bootique side (and I already have unit tests there).

[MalloD12]

You can try adding the tests also in Liquibase repo and I can help with it, or you can share the scenarios you would like to add and I can assist you on adding them.

Thanks,
Daniel.

[andrus]

Let me poke around

[andrus]

@MalloD12, I updated the PR. Now it has two commits:

• c0d203b - a unit test that shows how relative paths do not work for a certain type of dynamic ResourceAccessors
• 94950ca - my original "fix" that will be replaced with GlobalConfiguration property as you suggested.

Hope this helps with that property implementation.

[MalloD12]

I think we should leave this code as it is for now.

[MalloD12]

Can you please move this test class to a separate file?

[MalloD12]

Hi @andrus,

I've been looking at this PR and it's still in a not mergeable state. By removing:

if (filePath.startsWith("classpath:")) {
    filePath = filePath.substring("classpath:".length());
}

will change and break the current behavior which is what we are trying to avoid at the moment. I think it is a reasonable approach to create a new resource accessor class as you are doing and handling there the decision to according the global configuration we were discussing previously decide whether we want to remove or not the classpath: prefix from the path when normalizing.

The global configuration I was talking about would be something like:

INCLUDE_CLASSPATH_PREFIX = builder.define("includeClasspathPrefix", Boolean.class)
                .setDescription("If false Liquibase will remove 'classpath:' prefix when normalizing a resource path. By default, path will keep the classpath prefix")
                .setDefaultValue(true)
                .build();

Thanks,
Daniel.

[andrus]

Yes, this is what I said in my comment. The second commit in the PR is for information purposes so to speak. Anyways, I will remove it shortly to avoid any confusion.

[andrus]

Just updated the PR:

• Removed the changes from DatabaseChangeLog
• Moved the test ResourceAccessor into a top-level class as requested in the previous review.