My Emacs configuration https://git.cjennings.net/dotemacs/atom?h=main 2026-08-06T16:50:47+00:00 2026-08-06T16:50:47+00:00 Craig Jennings c@cjennings.net 2026-08-06T16:50:47+00:00 urn:sha1:36c737f45efc4f9459a7923d62ccf2cf5bf2686a - Keeps the stream-copy loop technique and its measurements. - Notes the amix normalize=0 trap that silently discards the mix ratio. - Records the loop-period arithmetic behind the crossfade. - Explains why the openly-licensed set sounds better but its rain is wrong. The audio itself stays untracked. Claude-Session: https://claude.ai/code/session_01WpQhUr9mNrXss3HiaN6G2s 2026-08-03T20:27:17+00:00 Craig Jennings c@cjennings.net 2026-08-03T20:27:17+00:00 urn:sha1:6f90ec5b660c6a2552ac7284a3d9d963771fbbc2 cj/music--append-track-to-m3u-file decided whether to prepend a newline by seeking to a byte offset from file-attributes, which doesn't follow symlinks. On a stow-deployed playlist that measures the link string instead of the file. Two failure modes, and I only went looking for the second. All 100 symlinked playlists read the wrong byte, so a file already ending in a newline looked unterminated and gained a blank line on every append. On the 31 whose link string is longer than their content the range fell outside the file entirely. Nothing was inserted, and char-after handed nil to a numeric comparison, so the append died with a wrong-type error. The probe now reads the file and checks its last character. That takes file-attributes out of the path, so this class can't come back. The largest real playlist is 11 KB and the read costs 0.11 ms, so being clever about offsets bought nothing. A test per mode, because one fixture can't show both. Seed the blank-line case with content shorter than the link string and it lands on the error path instead, going red for the wrong reason and proving nothing. The cheap patch was guarding char-after against nil. That would have silenced the crash on 31 playlists and left the blank line live on all 100. 2026-08-03T19:38:20+00:00 Craig Jennings c@cjennings.net 2026-08-03T19:38:20+00:00 urn:sha1:a457ade9207be48f6bfaa519046d7acea390ff1b A track under the playlist's own directory still writes relative. Anything outside it now writes the absolute path instead of a ../ chain. Playlists in the mpd directory point into ~/music, so the relative form there was four levels of .. and broke the moment anything moved. I took the trade knowingly. ambience.m3u is the only playlist anywhere using ../ lines, and it reaches its audio through a sibling directory. So a newly appended track there writes absolute while its hand-written lines stay relative, and its header comment needs a matching edit. The alternative was an exception for tracks inside the deployed tree, which would be a rule serving exactly one file. The comment I wrote for the earlier half of this argued one direction only. Absolute survives the playlist moving, which I said. Relative survives the library moving, which I didn't, and for a track in the same tree the ../ form survives strictly more moves rather than fewer. Rewritten to name the trade instead of implying there isn't one. The boundary case is a directory whose name begins with two dots. The check looks for a leading "../", so a real subdirectory called ..hidden sits under the playlist and must stay relative. It has a test now, because loosening that check to ".." would break that case and nothing else. 2026-08-03T17:57:36+00:00 Craig Jennings c@cjennings.net 2026-08-03T17:57:36+00:00 urn:sha1:90aa02c5cd7dc78729d4362d4962005627069e60 cj/music--append-track-to-m3u-file wrote every relative path against cj/music-root wherever the playlist actually sat. Both readers resolve against the playlist's own directory: cj/music--m3u-file-tracks, and EMMS's emms-source-playlist-parse-m3u. So the writer and the readers disagreed whenever those two directories were not the same one. That is not only the playlist-outside-the-music-root case I went looking for. 39 of the 66 m3u files under ~/music are album playlists sitting beside their tracks, storing bare filenames. Appending to any of them wrote a line like Artist/Album/track.flac that resolved to a path which never existed. The fix is the base: (file-name-directory m3u-file). Worth recording how the tests went, because the first three I wrote all passed against the broken writer. I had put cj/music-root and the playlist directory at the same depth, so file-relative-name returned the same string for both bases. The fixture reached the bug's code path and still could not see it. The discriminating cases put the two bases at different depths, and reverting the one-line fix now fails them. Eight existing tests moved with it. They built track paths from the real cj/music-root while the playlist sat in a temp directory, which pinned the old base and also read my actual music root. Rebinding the root to the fixture directory restores every assertion verbatim and makes them hermetic. One thing this does not change: MPD resolves relative playlist entries against music_directory and rejects entries that escape it, so a ../ line is for EMMS rather than MPD. That costs nothing today, since EMMS drives mpv here and MPD is not running. 2026-08-03T14:59:58+00:00 Craig Jennings c@cjennings.net 2026-08-03T14:59:58+00:00 urn:sha1:fa1e63e4b3e4dbe3841247c8efd2096d2d75390c The two toggles were only on C-; r v and C-; r a. F9 has been free since ai-term moved its family to M-SPC, so it goes to what I reach for mid-work. Both keep their prefix argument, so C-u F9 still prompts for a location. Reaching them from inside a terminal buffer turns on which key categories each EAT input mode claims. My first read of that was backwards. Semi-char mode is the default and where agent buffers sit. It's built from :ascii, :arrow and :navigation, so it never claims function keys and F9 already reached Emacs there. That entry is redundant. Char mode is the one that mattered. It adds :function, binding f1 through f63 to eat-self-input. It's also a minor mode, so its map outranks eat-mode-map. Worse, :function claims only the unmodified keys. Leaving char mode alone would have split the pair rather than broken it: Shift-F9 toggling audio while F9 went to the program under the cursor. That is a recording I believe I started and didn't. Both chords are claimed in eat-char-mode-map and eat-eshell-char-mode-map, which costs a char-mode program the use of F9. I'd rather pay that than ship a toggle that half works. The tests resolve the char-mode cases through key-binding in a fixture that reproduces minor-mode precedence, not by reading the binding back out of the map the module just wrote. A positive control asserts F8 still reaches eat-self-input, since F8 sits in the same category and nothing here touches it. Without that control the fixture could be inert and every assertion would still pass, because F9 resolves through the global map either way. 2026-08-03T14:39:08+00:00 Craig Jennings c@cjennings.net 2026-08-03T14:39:08+00:00 urn:sha1:eb0c2191b77e795cca7884e0690ea51c515527fd The hourly timer armed itself from org-agenda-mode-hook, so a session where I never opened the agenda never synced at all. After a reboot that is every session until the first agenda call. I found it with all three calendar files still frozen at the pre-reboot write, five hours stale. The deferral had a real reason. Feed URLs behind :secret-host live in authinfo.gpg, and starting at load would prompt for a passphrase on a cold gpg-agent. The cost was worse than what it bought. A user timer owns the schedule now, every ten minutes. calendar-sync-auto-start defaults to nil so the editor no longer arms its own. scripts/calendar-sync-run is the batch entry point, and it blocks until every calendar leaves the syncing state. The pipeline is asynchronous end to end, so a batch Emacs that returns early exits zero having written nothing. Installing it on a machine that resolves its feeds through :secret-host surfaced two bugs the inline-url machine could never show: - cj/auth-source-secret-value called auth-source-search behind nothing but a declare-function, which quiets the compiler and loads nothing. An interactive Emacs always has auth-source in by the time anyone calls, so the omission stayed invisible until a batch -Q sync died on a void function. The require is guarded on fboundp. An unconditional one reloads auth-source over whatever is already there, which replaced a caller's stubbed search mid-call and sent an existing test out to the real authinfo. - A synchronous failure in one calendar aborted the whole loop. The async callbacks record their own failures, but they never run when the error lands before a process starts. Resolving a :secret-host feed signals outright on a cold agent. Failures are contained per calendar now, so one bad feed no longer costs the other two. A failed row prints the reason it recorded. Batch Emacs discards the *Messages* buffer the interactive path logs to, so without it the journal shows only "error", with no way to tell a cold agent from a revoked token. 2026-08-01T20:17:41+00:00 Craig Jennings c@cjennings.net 2026-08-01T20:17:41+00:00 urn:sha1:0eb65b6dcc2db54663b39dc7201c3dcf151add74 wp adds a slash to pix-dir, which already ends in one, so it resolves to pictures//wallpaper/. I built this entry off videos-dir without the extra slash and left wp alone. 2026-07-31T17:39:27+00:00 Craig Jennings c@cjennings.net 2026-07-31T17:39:27+00:00 urn:sha1:0b93863689038a385fd320440ac0a413ce81c4e1 I had to kill an Emacs session today over a prompt I could not answer. A second agent session held the selected window while "Overwrite hiroshi.m3u? (yes or no)" waited in another frame, so my keystrokes went to a terminal and the minibuffer stayed empty. Losing unsaved buffers is a far worse outcome than a mis-keyed confirm. So cj/confirm-strong becomes cj/confirm-destructive, and instead of binding use-short-answers to nil for a typed "yes" it reads a single y or n. Two things I kept: - No default. Only y and n answer, so a stray RET or space re-prompts rather than confirming a shutdown. - Pending input is discarded before the read. Without that the change would have been a bad trade: read-char-choice reads the input queue, so a key typed before the prompt painted would confirm instantly, where the old typed-"yes" absorbed it harmlessly. Worth being honest about what this does not fix. One keystroke does not make a prompt reachable when focus is elsewhere, and C-g is still the escape. It lowers the cost of the situation rather than preventing it. Six call sites: shutdown and reboot, shred, playlist overwrite and delete, and the two file overwrites. 2026-07-31T16:04:42+00:00 Craig Jennings c@cjennings.net 2026-07-31T16:04:42+00:00 urn:sha1:445a0be315f3d6fff585f1558262fb5dacd10d12 The surface reading this has to stay correct while Emacs is down, and emacsclient is exactly the thing that cannot. So the writer is a batch Emacs and a user timer runs it every five minutes. - scripts/agenda-render-cache writes the cache from a batch Emacs. It resolves the agenda file list through the config's own resolver rather than restating it, so adding a calendar source stays a one-place change. - The timer uses OnCalendar so Persistent actually applies. On a monotonic schedule it is silently ignored, and a machine that slept would come back to a stale file with nothing to trigger a catch-up. - The window is now three days, yesterday through tomorrow. A consumer drawing a rolling window centred on now has nothing to draw for the part of its span outside today, which late in the evening is half the surface. The keyword list moved to user-constants, where a batch Emacs can reach it without this config's package dependencies. Without it org did not recognise DOING or VERIFY, so it stopped parsing those headlines as tasks at all: the keyword and priority cookie stayed glued to the front of every title and each row reported no keyword and not-done. The file parsed cleanly and was wrong, which is the failure worth guarding. The bats tests run against a fixture and this checkout, not the machine's real agenda and installed config. Asserting over live data let the same suite pass on a day that happened to have no keyworded entries. 2026-07-31T14:59:29+00:00 Craig Jennings c@cjennings.net 2026-07-31T14:59:29+00:00 urn:sha1:37f21839ca62617fef07684b2f98268de47a71c2 The surface that reads this went live, and it reads a different shape than the canonical profile emits. - cj/agenda-render-json adds s and e in epoch milliseconds, plus t for the title. The canonical fields ride along, so one file serves both readers. - Every row gets an end it can be drawn with. An all-day entry spans its day. A point event has zero width. The canonical profile still reports null there, which is faithful to the source but not a width. - cj/agenda-render-cache-update writes today to the XDG cache path. Input stays epoch seconds on both profiles. Only the output converts, and the guards that catch a milliseconds-for-seconds caller still apply. Nothing schedules the cache write yet, so the file is only as fresh as its last call. Picking a refresh mechanism is its own decision.