Core: Fix relativize() to handle path equal to prefix (#15172)

[Baunsgaard]

What

Fix RewriteTablePathUtil.relativize() to handle the edge case where path equals prefix exactly.

Why

Currently, relativize() fails when a path equals the prefix (e.g., write.data.path set to the table root). This breaks rewrite_table_path for tables with properties pointing to the table location itself.

Example failure:
// Throws IllegalArgumentException instead of returning ""
RewriteTablePathUtil.relativize("/path/to/table", "/path/to/table");

Use case:
Storage migration or replication where write.data.path = table location.

How

Added a check for exact match after normalizing trailing separators:

// Handle exact match where path equals prefix (without trailing separator)
if (maybeAppendFileSeparator(path).equals(toRemove)) {
  return "";
}

Changes

• RewriteTablePathUtil.java: Fix relativize() + updated Javadoc for relativize() and newPath()
• TestRewriteTablePathUtil.java: Added 10 test methods covering:
  • Normal relativization
  • Path equals prefix (the fix)
  • Trailing separator variations
  • Invalid path rejection
  • Subdirectory scenarios (backup/restore)
  • Overlapping name rejection (/table vs /table-old)

Testing

./gradlew :iceberg-core:test --tests "org.apache.iceberg.TestRewriteTablePathUtil"

[szehon-ho]

thanks, discussed offline with @Baunsgaard , I believe this was an initial oversight of the code

[dramaticlly]

should this pass? I guess /a cannot be relative to the given prefix of /a/, right? Unless we want to normalize file separator on both side

I think those 2 are ok as of now

    assertThat(RewriteTablePathUtil.relativize("/a/", "/a")).isEqualTo("");
    assertThat(RewriteTablePathUtil.relativize("/a/", "/a/")).isEqualTo("");

[Baunsgaard]

I would have expected that case to work. I added explicit handling for it, by modifying your suggested code to do the maybe append file separator :
if (!path.startsWith(toRemove) && !maybeAppendFileSeparator(path).equals(toRemove)) {

[Baunsgaard]

Thanks for the review feedback @szehon-ho and @dramaticlly!

I have updated the implementation, though with a slight variation from the suggested approach.

What was suggested

if (!path.startsWith(toRemove) && !path.equals(prefix)) {
  throw ...
}

What I implemented

if (!path.startsWith(toRemove) && !maybeAppendFileSeparator(path).equals(toRemove)) {
  throw ...
}

Why the difference

The suggested approach using !path.equals(prefix) was stricter - it would reject relativize("/a", "/a/") because "/a" != "/a/".

I chose to normalize both sides because:

• /a and /a/ represent the same directory
• Table properties may come from different sources with inconsistent trailing separators
• Being strict about this mismatch seemed overly pedantic for a utility method

This means all trailing separator combinations now work: