mirror of
https://github.com/vitorpamplona/amethyst.git
synced 2026-10-06 03:38:23 +00:00
Vitor PamplonaandClaude Opus 5.5
c2fda6aae2
An upload went to the one server picked in the dialog and failed there. Which servers accept what isn't visible in advance: primal, azzamo and blossom.band answer an encrypted (application/octet-stream) blob with 415, the paid ones with 402, and any server can be down. So an encrypted chat attachment failed depending on which default happened to be selected. The Blossom settings screen already told users uploads 'try each server from the top down'; nothing did. Blossom uploads, plain and encrypted, now try the picked server first and then the rest of the list the picker offered (the user's kind-10063 servers, or the defaults when they have none), top down, skipping duplicates. An encrypted upload re-sends the same ciphertext, so its key, nonce and hash stay valid wherever it lands. A read-only login stops at once; if every server fails, the selected server's error is shown. NIP-96 and NIP-95 keep a single target. White Noise (MDK 0.10.4) fetches any safe https Blossom URL, not only the group policy's endpoints, so a fallback server is readable by it. The shared JVM UploadOrchestrator (desktop app and amy) gets the same behaviour: upload/uploadEncrypted take the account's other servers and try them in order; the desktop DM, note composer and profile-picture uploads pass the list. Its auth header isn't server-scoped, so one serves every attempt. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>