Skip to content

Browsable folders

The report and the dashboard's file index both link to folders as well as to files, and S3 serves objects rather than directories, so those links land on nothing once the results are published. index_directories.sh fills that gap in before the upload: it walks the results folder and renders templates/listing.html into every directory of it, the results folder included — subfolders first, then files with their sizes, every entry linked, and a link back up.

Names are HTML-escaped for the page and percent-encoded for the href beside it, a byte at a time, since a filename is bytes and a # in one would otherwise cut its own link short.

The listings are never called index.html

They are directory_listing.html, and that is the whole convention:
index.html a page something wrote to be read — the landing page at the top of a run, a Krona chart under a shotgun run's krona/
directory_listing.html the listing of the folder it sits in, written here

Because the two never collide, the script has no special cases: it writes a listing into every directory it finds, whether or not the pipeline published a page of its own there, and a second pass over the same results folder just rewrites them. A folder holding a tool's own index.html ends up with both — that page, and a listing of the per-level CSVs beside it.

Every link is written by name. A folder row points at subfolder/directory_listing.html, not at subfolder/, and "up one folder" points at ../directory_listing.html. Nothing depends on the server rewriting a folder URL, which is what lets an unpacked copy of the run's zip browse exactly the way the published run does.

Folder URLs are reachable too, because CloudFront maps one onto the listing beneath it — see CloudFront. That is what catches the folder links inside nf-core's own report, which this system does not write, and a reader editing the address bar.

Where "up one folder" ends

At the results folder, whose listing is the one page that leaves the frame: its link out is ↑ Results dashboard, carrying target="_top". The listings below it link up normally and stay in the frame.

That matters because the landing page reads these listings inside itself, and the top of a run is the dashboard — so a chain that walked up into it without breaking out would open a second copy of the dashboard in the dashboard's own frame. The root listing also leaves index.html out of its rows for the same reason.

The one link that would otherwise reach the same dead end is in ampliseq's report: its Final notes section says the read count report "can be found in the base results folder" and writes that as href="../", which CloudFront resolves to the dashboard. ampliseq_upload.sh repoints it at ../directory_listing.html before the upload.

Whether it opens or downloads is not decided here. It follows from the content type each object was uploaded with — see What a link does when you click it. A .tsv or .log opens in the browser because it was uploaded as text/plain, not because the listing says anything about it.

Like the progress page, none of this is worth failing a run over: a listing that cannot be built is warned about and whatever rows did come out are kept, so the link that led there still resolves. Results nobody can browse are still results.