How are "adjacent files" identified

[zwhitten]

I'm curious how files are identified as being "adjacent" in a scenario where no partitioning is present on a table.
I have a test application that performs a number of small inserts every X minutes. Initially I had a few hundred smaller files (expected) so I ran the merge_adjacent_files() function and combined them. This created a larger file ~82KB in size.
After performing additional inserts into the same table and again running the merge_adjacent_files() function, I then had the original 82KB file and a new 10KB file.

-rw-r--r--  1 zwhitten 82126 Jul 15 14:46 ducklake-01980f69-1692-753a-928c-94099833d188.parquet
-rw-r--r--  1 zwhitten 10209 Jul 15 15:37 ducklake-01980f97-d63c-7310-9051-718ba645b455.parquet

In this case I was expecting the new records get combined into the larger file (since both are much less than the configured max file size)
Is this the expected behavior? I'm assuming that DuckLake determined these files were not adjacent but I'm curious why? There's no partitioning configured on the table

Update:
After further testing this seems to be related to expiring snapshots. It seems that once the snapshot that created a file(s) has been expired the files can no longer be merged with adjacent files? I'm not sure if this is expected behavior or not.

Steps:

1. A series of many inserts into a table creating a number of small files
2. merge_adjacent_files() successfully combining all the small files
3. expire snapshots before "now"
4. perform another series of small inserts creating small files
5. merge_adjacent_files() at this point only the files created in step 4 get combined and they do not combine with the larger file from step 2