Arvados

Customer wants to search for a file by name, and probably also by path.

Currently, they can use the trigram index on the full Collection record.

This turned up an unexpected behavior.

Searching for "dmel_r6.16/Bowtie2Index/genome.fa" doesn't work

Searching for "dmel_r6.16/Bowtie2Index" works

This is because the file_names column is populated with each file name and each stream name ("directory") from the manifest. The file name and stream name are not joined to get a full path. So the file_names table looks like:

genome.fa
./dmel_r6.16/Bowtie2Index

It would better fit user expectations if it was formatted like this:

./dmel_r6.16/Bowtie2Index/genome.fa

Propose migrating the column contents to use full paths.