VersionControl:PRs/magenta/441: Difference between revisions
From PickiPedia: A knowledge base of bluegrass, old time psychedelic jams, and other public domain music
Jump to navigationJump to search
cryptograss/magenta#441, as GitHub has it |
cryptograss/magenta#441, as GitHub has it |
||
| Line 16: | Line 16: | ||
|deploy= | |deploy= | ||
}} | }} | ||
<pre class="pr-body">From Justin's voice memo, 8 Nov in #interfaces, after catching up in voice mode on his phone. | |||
**The top line flickered.** `renderVoiceMode` runs every second, and anything that changes (an agent's clock) replaced the whole thing. The top bar ("heard", or "recording voice memo") played its `voiceSlideIn` animation again every time. Now it slides in only when it's new: an unchanged top bar gets `.still`, which has no animation. | |||
**"Magent, with tools, says" and then nothing, again and again.** A message with nothing to say aloud still got its narrator line, made and played. That covers a picture alone (`` reads as empty) and words only inside [brackets], which the voice performs as a tag rather than says. Then its words came back `400 nothing in that message to read aloud`, so the reader retried it, narrator and all, before giving up. | |||
- `voice.has_words`: whether any piece has text outside [tags]. `speak_piece` uses the same test. | |||
- `?check=1` on the narrator's ask answers `{"nothing": true}` (400) before any audio is made. A failed ask for the words carries `nothing: true` too. | |||
- The page passes over such a message quietly (`passOver`: reported `skipped`, its place saved, on to the next) instead of retrying it. | |||
- Test: a picture alone and a bracket-only message each answer `nothing` for the narrator and for the words, with no text-to-speech request made. A message with words still gets its narrator. It fails without the server change. | |||
**The six buttons overlapped the list.** On a phone, the buttons now fill the bottom fifth (`20dvh`, two rows of three, tight) and the list scrolls above them, never under. Checked at 402×874 with the CSS injected on live magenta: the buttons span 693–868 and the list ends at 685. | |||
**"leave voice mode" moves to ☰** as its first item, big, while in voice mode. ☰ already opens the Moods over voice mode. On a wide screen, which has no ☰, the line under the buttons stays. | |||
Not in this PR: the cross-device "catch up from where I left off" and notifications that only show what's new since the last read on any device. Both build on `ReadMark`, which already exists, and get their own PR. Also not here: the later clips playing slower; I haven't found the cause yet. | |||
The whole suite: 1,786 pass. | |||
Redeploy maybelle. | |||
Worked on by: magent and justin, in #interfaces</pre> | |||
Revision as of 16:06, 9 November 2026
| State | 🟢 open |
|---|---|
| Worked on by | magent, justin |
| Moods | #interfaces |
| Deploy | not said |
| Branch | magent/voice-mode-phone → main, by magent-cryptograss
|
| Opened · last pushed | 2026-11-08 19:36:58 · 2026-11-08 19:35:14 |
From Justin's voice memo, 8 Nov in #interfaces, after catching up in voice mode on his phone.
**The top line flickered.** `renderVoiceMode` runs every second, and anything that changes (an agent's clock) replaced the whole thing. The top bar ("heard", or "recording voice memo") played its `voiceSlideIn` animation again every time. Now it slides in only when it's new: an unchanged top bar gets `.still`, which has no animation.
**"Magent, with tools, says" and then nothing, again and again.** A message with nothing to say aloud still got its narrator line, made and played. That covers a picture alone (`` reads as empty) and words only inside [brackets], which the voice performs as a tag rather than says. Then its words came back `400 nothing in that message to read aloud`, so the reader retried it, narrator and all, before giving up.
- `voice.has_words`: whether any piece has text outside [tags]. `speak_piece` uses the same test.
- `?check=1` on the narrator's ask answers `{"nothing": true}` (400) before any audio is made. A failed ask for the words carries `nothing: true` too.
- The page passes over such a message quietly (`passOver`: reported `skipped`, its place saved, on to the next) instead of retrying it.
- Test: a picture alone and a bracket-only message each answer `nothing` for the narrator and for the words, with no text-to-speech request made. A message with words still gets its narrator. It fails without the server change.
**The six buttons overlapped the list.** On a phone, the buttons now fill the bottom fifth (`20dvh`, two rows of three, tight) and the list scrolls above them, never under. Checked at 402×874 with the CSS injected on live magenta: the buttons span 693–868 and the list ends at 685.
**"leave voice mode" moves to ☰** as its first item, big, while in voice mode. ☰ already opens the Moods over voice mode. On a wide screen, which has no ☰, the line under the buttons stays.
Not in this PR: the cross-device "catch up from where I left off" and notifications that only show what's new since the last read on any device. Both build on `ReadMark`, which already exists, and get their own PR. Also not here: the later clips playing slower; I haven't found the cause yet.
The whole suite: 1,786 pass.
Redeploy maybelle.
Worked on by: magent and justin, in #interfaces