Nested Hyprland worked but had real limits: the outer compositor decided
the nested window's pixel size (needing an outer-session float+resize
dispatch per capture), occluded surfaces got no frame callbacks (so grim
hung unless the nested window was also focused/raised), and there was no
way to fully suppress a brief real, visible flash of that window on the
operator's desktop.
wlroots' WLR_BACKENDS=headless (Sway, not Hyprland, is built on wlroots
directly) has a genuine headless backend: no seat/DRM-master claim, no
window anywhere, ever. Confirmed empirically: zero visible footprint,
both zwlr_layer_shell_v1 and zwlr_screencopy_manager_v1 present, grim
completes instantly with no focus dance needed.
This drops the Hyprland-specific plumbing that no longer applies:
- bread-screenshots now exposes one compositor-agnostic capture_region
primitive instead of capture_layer/capture_output, since the isolated
canvas size is always known up front rather than queried via hyprctl.
- bread-utils::hypr loses the Monitor scale/transform/logical_size and
Layer/find_layer additions that only existed to support that querying.
- bread-capture's isolation module spawns headless Sway instead of a
nested Hyprland instance, and passes --width/--height through to the
target app so it knows the canvas size without asking anyone.
Also fixes a socket leak in isolation teardown: killing the compositor
(Hyprland or Sway) doesn't unlink the wayland-N/.lock files it created,
so every capture run was orphaning a socket pair in the runtime dir.
Drop now removes them explicitly.
New bread-screenshots crate captures a layer-shell surface (by namespace+pid,
to disambiguate from an already-running instance) or the whole focused
output via grim, using bread-utils::hypr/proc. bread-utils::Monitor gains a
scale field and logical_size() so output geometry accounts for HiDPI/
transform, matching breadshot's proven math. bread-utils::hypr gains
find_layer() over hyprctl layers -j.
bread-capture is a small orchestrator that drives an app's --screenshot mode
and collects the resulting PNGs; hardcoded to breadbar's two views for now.
home_dir() was already computed internally (home_or_root) but not exposed
— breadarr-shared's own expand_home() helper needs it to fix the same bug
class documented in this module: its own fallback (when HOME itself isn't
set) returned the literal unexpanded "~/..." input string instead of a
real path. Also documented breadmon/src/profile.rs's profiles_dir(),
fixed in that repo's own commit.
Neither original implementation this replaces (breadbox's
get_active_workspace, breadclip's hyprctl_json) set a read/write timeout
on the Hyprland IPC socket — a wedged or mid-reload Hyprland instance
could hang the call indefinitely, and every current caller runs it on the
GTK main thread, so a hang here freezes the whole UI. Found while
reviewing the crate for this pass's "any existing bugs" sweep; same class
of bug as everything else bread_utils::proc/hypr exists to fix, just one
I'd introduced myself by porting the original code faithfully without
also porting the missing safety net.
singleton.rs's tests used to call the real kill() on the current test
process itself (the recorded "other instance" PID was our own, since
tests run single-process) — split try_acquire (side-effect-free) out from
toggle_or_kill (sends the signal) so the concurrency assertions no longer
risk SIGTERM-ing the test binary.
Separately, hypr.rs's env-var test mutates HYPRLAND_INSTANCE_SIGNATURE/
XDG_RUNTIME_DIR process-globally; cargo runs tests in parallel threads by
default, so it could race a concurrently-running singleton or xdg test
expecting the real XDG_RUNTIME_DIR, intermittently failing them with
ENOENT. Added a shared env_test_lock() all env-var-touching tests now
acquire for their duration.
Verified via 5 repeated full test runs with zero flakes (28/28 passing
each time).
bread-utils extracts genuinely duplicated logic found across breadbox,
breadclip, breadmon, breadcrumbs, bos-settings, and breadhelp:
- hypr: Hyprland socket1 request/response client (breadbox's
get_active_workspace + breadclip's position.rs hyprctl_json were
near-identical), socket2 path resolution (breadmon), and a
version-tolerant `fullscreen` field parser (Hyprland has shipped both
bool and int representations across versions).
- singleton: correct flock-based single-instance toggle, replacing the
TOCTOU-prone read-pid/check-proc/kill/write-pid pattern duplicated
verbatim between breadbox and breadclip (breadclip's own comment says
"matches breadbox pattern").
- proc: breadcrumbs' timeout-guarded subprocess runner, promoted verbatim
as the one implementation in the ecosystem that already got this right.
- atomic + xdg: atomic (temp-then-rename) file writes with an optional
.bak-before-overwrite variant, and XDG path helpers that never fall back
to a literal "~/..." string (the exact breadclip-core and
breadpad-shared bug: PathBuf never expands `~`).
- tomlcfg (feature "toml"): the load_doc/save_doc TOML-editing discipline
bos-settings and breadhelp both implemented byte-for-byte identically in
the same fix pass that introduced it.
- gtk_popup (feature "gtk"): layer-shell overlay window setup, visible-row
navigation, and click-outside-close, deduplicated from breadbox and
breadclip (~150 duplicated lines, per both apps' own "same as breadbox"
comments).
bread-onnx extracts the embedding pipeline (tokenize -> tensor build ->
mean-pool -> L2-normalize) duplicated near-verbatim between breadarr and
breadsearch, a shared execution-provider session builder with loud EP-
registration logging, and a model download+integrity helper. Defaults AMD
iGPU acceleration to ort::ep::MIGraphX (not ROCm) per this machine's own
breadsearch-gpu-backends lesson: ROCMExecutionProvider silently no-ops to
CPU on distro ROCm onnxruntime builds compiled with --use_migraphx.
Both crates build and pass their own test suites standalone. Consumer
migrations follow in subsequent commits.
