The truly free, open-source, local-first, cross-platform CapCut replacement. Scriptable through a JSON-RPC API & MCP.
Rust ffmpegfree-video-editormodel-context-protocolnon-linear-editoroffline-firstopen-source-video-editorself-hostedvideo-editingvideo-editing-automationvideo-editing-softwarevideo-editing-toolsvideo-editorvoice-cloningyoutube-automationyoutube-shorts
build: `cargo dev` runs the window in twenty seconds, not three minutes
Jareer6495f1f69967
diff --git a/src/.cargo/config.toml b/src/.cargo/config.tomlindex 0fae32c9..ba66476b 100644--- a/src/.cargo/config.toml+++ b/src/.cargo/config.toml@@ -4,10 +4,15 @@# at an unpacked shared build. Set those in your own environment, never# here - a machine path committed to this file breaks every other checkout.-# No incremental caches. They are a speed-up for the second build of one-# crate and a tax on the disk for everything else: a session of edits across-# the workspace leaves tens of gigabytes of them behind. The shipping builds-# run on GitHub Actions; a machine here needs `cargo check` and the tests of-# the crates it is changing, and those are quick without.-[build]-incremental = false+# No incremental caches for the dev and test profiles: they are a speed-up+# for the second build of one crate and a tax on the disk for everything+# else, and a session of edits across the workspace leaves tens of+# gigabytes of them behind. That is set per profile in Cargo.toml rather+# than as `build.incremental` here, because the config key overrides every+# profile - it was silently switching off the `quick` profile's+# incremental, which is the one build that wants it.++# `cargo dev`: the window, built to try a change in. See `[profile.quick]`+# in Cargo.toml for what it is and why it is quick.+[alias]+dev = "run --bin concat --profile quick"diff --git a/src/Cargo.toml b/src/Cargo.tomlindex fbe5587d..b833704b 100644--- a/src/Cargo.toml+++ b/src/Cargo.toml@@ -52,7 +52,8 @@ opt-level = 3lto = "thin"codegen-units = 1-# For trying a change in the window: `cargo run --profile quick -p concat`.+# For trying a change in the window: `cargo dev` (an alias for+# `cargo run --bin concat --profile quick`, see .cargo/config.toml).# Optimised enough that video plays as it would in release, without the# thin LTO and single codegen unit that make a release rebuild a# minutes-long relink. Incremental, so a change in one crate is seconds.@@ -62,6 +63,23 @@ lto = falsecodegen-units = 16incremental = true+# The window's own crate unoptimised. It is where every edit lands, and+# most of it is the Slint compiler's output: one enormous module that the+# front end takes fifty-odd seconds to check and that LLVM then took two+# minutes over at level 3 - and, measured, the same two minutes at level+# 1, since the time is a few huge generated functions and not the level.+# At 0 that is ten seconds. With the profile's incremental honoured (see+# .cargo/config.toml) a Rust edit rebuilds in twenty seconds and a .slint+# edit in twenty-five, from three minutes twenty. The engine crates keep+# the profile's level 3: decoding and compositing are theirs, and they+# only rebuild when they change. What this crate does itself - publish the+# lanes, run the panes, format a waveform - is light enough that nothing+# on screen tells the difference; if something ever does, the shipping+# build is `--profile app`.+[profile.quick.package.concat]+opt-level = 0+codegen-units = 256+# The shipping build of the editor window: `cargo build --profile app -p concat`.## Every knob here is set for the fastest possible binary, and each one costs@@ -101,3 +119,11 @@ strip = "symbols"# the rest of the engine is being debugged.[profile.dev.package."*"]opt-level = 2++# No incremental caches for the everyday check-and-test builds; see+# .cargo/config.toml for why it is said here and not there.+[profile.dev]+incremental = false++[profile.test]+incremental = false