[JainAyush-01]

This Pull request:

Fixes silent data loss when merging ROOT files containing nested subdirectories (depth >= 2) with chunking / batching enabled (TFileMerger::SetMaxOpenedFiles or hadd -n).

Changes or fixes:

Root Cause:

When TFileMerger merges files incrementally in batches (type & kIncremental), TFileMerger::MergeOne cleans up in-memory directory objects at lines 830–834:

if (ownobj && (!(type & kIncremental) || dirobj->GetFile() != target)) {
   dirobj->ResetBit(kMustCleanup);
   delete dirobj;
}

For top-level subdirectories (depth 1), target is the TFile* itself, so dirobj->GetFile() == target evaluates to true and dirobj is preserved across batches until the final fOutputFile->Write() flushes all accumulated objects.

However, for nested subdirectories at depth $\ge 2$ (e.g., nested/dir):

• target is the parent TDirectoryFile* (nested)
• dirobj->GetFile() returns the root TFile* (fOutputFile)

Because a TDirectoryFile* does not equal a TFile*, dirobj->GetFile() != target always evaluated to true. This prematurely deleted the in-memory subdirectory dirobj before subsequent batches could write and flush into it, causing silent loss of objects merged in later batches.

Fix:

Change dirobj->GetFile() != target to dirobj->GetFile() != fOutputFile. This correctly checks whether dirobj belongs to the output file being accumulated across incremental batches, ensuring nested directories and their contents remain intact in memory until written.

Testing:

• Added GoogleTest regression test TEST(TFileMerger, NestedDirMaxOpenedFiles) in io/io/test/TFileMergerTests.cxx.
• Validated with multi-level directory depths (depth 2 and depth 3) and batch configurations (-n 2, -n 3).
• Verified 100% pass on full gtest-io-io-TFileMerger test suite via ctest.

Checklist:

• tested changes locally
• updated the docs (if necessary)

This PR fixes #23666

[github-actions]

Test Results

    23 files      23 suites   3d 18h 51m 33s ⏱️
 3 899 tests  3 895 ✅ 0 💤 4 ❌
79 664 runs  79 660 ✅ 0 💤 4 ❌

For more details on these failures, see this check.

Results for commit 65ae899.

[JainAyush-01]

Thanks @vepadulano and @enirolf for improving the test and adding the comprehensive roottest coverage!

I noticed in the CI run for commit 65ae899 that the Alma10 build failed in gtest-io-io-TFileMerger due to histogram auto-registration being off in that configuration. Please let me know if you would like me to adjust the test or if you prefer to handle it directly. Happy to help!