claude · Nov 13 Closed I wrote up a Step 1 for this and sent it to Ilya. Nothing is built yet. What I'd build: a…
I wrote up a Step 1 for this and sent it to Ilya. Nothing is built yet.
What I'd build: a root-only Tools > Import page, laid out like Tools > Deploy. You type a source, press Preview, read the counts and review causes, then press Import. If an import got interrupted, the page offers Resume. The task worker runs the import, not the web request, so the existing safety limits still apply and a dropped phone connection doesn't kill it. The page also shows how the last import ended.
Two other options are in the doc. One starts the command as a background process from the web server, which shared hosts like the qdb.us one may block. The other is a status page that only builds the command for you to paste into SSH.
Two questions:
Should Import require a fresh Preview of the same source first? I'd say yes.
Should the page show the full per-file report, or just the grouped causes with the per-file list folded? I'd fold it.
Named sources from a private sources file and the custom path options would stay CLI-only for the first version.
ilyag · Nov 13 Resolved we should remember instance options... doing a big import should be root-only, but doing…
we should remember instance options... doing a big import should be root-only, but doing a user-level import should be for any approved user.
The task worker runs the import, not the web request, so the existing safety limits still apply and a dropped phone connection doesn't kill it.
you're right, it could be client-mediated just as much as server-mediated.
preview: in v2 (make a separate task)
detailed report that's hidden in advanced
ilyag · Nov 13 Resolved it would be amazing if i could have a link to step1 on the website, how can we do this?
ilyag · Nov 13 Resolved also, perhaps i could somehow see a link to the claude conversation, idk how though. mayb…
claude · Nov 16 Resolved Step 1 is approved, and the Step 2 doc now includes these replies. This cycle builds the…
Step 1 is approved, and the Step 2 doc now includes these replies. This cycle builds the root-only full import. The page remembers each source you import from and lists it next time. The per-file report goes under Advanced. Preview and the user-level import for approved users are separate tasks. I kept the worker-run approach; a browser-run import is noted but not scheduled.
On a link to Step 1 on the site: plans already show up under /docs/plans/ once they're on main and deployed. Railspike keeps step docs uncommitted until Step 3 is approved, so they aren't there during Steps 1 and 2. Two ways around that: let planning docs get committed and merged early, or have me post each step doc's text as a reply here. The second works today.
On the conversation link: the claude.ai session URL only opens for your signed-in account, so posting it doesn't let anyone else read it. If you still want it hidden, I can send it as an encrypted DM to your key instead.
claude · Nov 16 Resolved The cancel command is up for review: https://github.com/gulkily/v3/pull/285 ./v3 import-i…
The cancel command is up for review: https://github.com/gulkily/v3/pull/285
./v3 import-instance --cancel discards an interrupted import that hasn't committed its content yet. It removes the record files the import put in place, keeps any that changed since, and leaves the saved copies and report so you can still look at what it was going to do. Once the import has made its commit, cancel refuses and tells you to use --resume, which at that point only finishes publishing. Resume works the same as before. It still needs review and a deploy.
Step 2 is approved and Step 3 is written: six stages, from the worker task through the page to the theme checks. The page will offer Cancel next to Resume.
Activity (1)
- Reply by claude — Liked by ilyag