dtourolle/DarkRoom
1
0
Fork 0
Code Issues 39 Pull Requests Actions Packages Projects Releases 20 Wiki Activity

Export a whole selection, on a thread that is not the interface's

Browse Source 
The export button rendered, resampled and encoded a 24 MP frame on the UI
thread and the window was dead for all of it. That was written down as a known
compromise, on the grounds that a batch is what makes the wait intolerable
rather than merely noticeable. This is the batch, so the compromise comes due.

The grid's selection now exports (FR-EXP-7). A worker thread takes a clone of
the `GpuContext` — an `Arc` pair over a device and a queue — and opens each
photograph for itself: fetch, sidecar, decode, demosaic, render at full size,
resample, sharpen, encode, write. Nothing of that touches the interface, which
keeps drawing throughout, and the progress goes where every other background
job's does: one row in the activity register, with a count and a bar.

Why the worker does not borrow the session it could have had. A
`DevelopSession` owns the `AdjustPass` the canvas renders from, so handing it
to a worker would stop the develop view drawing for the length of the batch —
the same freeze, moved. Opening a session per image instead costs a
`Demosaicer` and an `AdjustPass` each time round, and the pipeline cache is
per-pass so the composed shader is recompiled per image rather than once for
the run. Against a full-resolution decode, render and encode that is a few
percent, and it keeps this file out of the pipeline `develop` owns. A reusable
export pass is the obvious next economy if a profile ever says so.

The open image is the exception, and it is why the develop button is not simply
a one-image batch. Its edit lives in the interface's session and may not have
reached a sidecar yet, so a worker that re-opened the file would export the
saved version rather than the one on screen. That frame is therefore rendered
by the caller and handed over as `Source::Rendered`; everything after the
render — the Lanczos reduction, the encode, the write, which is the larger half
of the wait and all of its variance — still leaves the UI thread. So the
develop export is no longer synchronous, but it is not fully off-thread either,
and the doc comment says so rather than claiming otherwise.

Cancellation (NFR-ARCH-3) is an `AtomicBool` read between stages, and the
export button becomes the cancel button while a run is live — a batch that
could only be stopped by not touching the selection would be a trap. Waits on
another worker use `recv_timeout` rather than `recv`, so a cancelled batch
sitting on a forty-megabyte download gives up within 100 ms instead of when the
transfer finishes. The honest bound is worse than that: a frame already in
render has no interior stopping point, so the worst case is one image. Closing
that needs the render itself to become interruptible, which is NFR-ARCH-2's
scheduler and not a finer poll here.

Failures are per image and typed (NFR-ARCH-4). One unreadable body, one folder
that cannot be written, one server that went away — each is a message on the
channel, a line in the log, and a count in the summary, and the batch carries
on. A run with any failure keeps its row until it is cleared, because that is
the row somebody came to the list to find; a cancelled run does not, because
they asked for it.

Two collisions that look alike and are not. `CollisionPolicy` is the user's
answer to "a file of this name was already there", and Overwrite is a fine
answer to that. It is not an answer to "the frame I exported four seconds ago
was also called this" — two folders in a library each holding an IMG_0001 is
ordinary — so a name the run has already issued is always stepped past whatever
the policy says about the folder. Both halves are held by tests.

Supporting changes, each smaller than it sounds. `open_session` comes out of
`load_bytes` so the worker shares the JPEG-versus-RAW routing rather than
carrying a copy that would drift; the half that builds a `slint::Image` stays
behind, where it belongs. `LibraryController::selected_image_paths` answers
from the catalog rather than from the loaded window, because selection is by id
and survives a scrub — a selection made before scrolling routinely names
photographs no row holds. `cache_context_for` takes an id for the same reason,
so a batch reads the originals cache instead of re-downloading three hundred
files. `format_date` is shared so `{date}` and the timeline agree about what
day a photograph was taken.

Left undone, deliberately: the batch is sequential, where FR-EXP-7 asks for all
available cores. Four full-resolution frames in flight is tens of megabytes
each and a straightforward way to exhaust a tablet, and the GPU is shared with
the interface in any case. Also undone: exporting with a chosen preset rather
than the current export settings — that is FR-EXP-5's machinery, which does not
exist yet.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
dtourolle
2026-08-17 09:59:57 +02:00
co-authored by Claude Opus 5
parent 2330ed25e9
commit 7d3c8c521f
6 changed files with 1410 additions and 141 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ui/dr-ui/src/export.rs
+949 -2
View File
@@ -27,13 +27,45 @@
//! promise and never do. An export waiting to upload is a promise — the user
//! was told the export succeeded — and sweeping it away to reclaim disk would
//! destroy work that no longer exists anywhere else.
//!
//! # The batch (FR-EXP-7)
//!
//! The second half of this file is a worker that exports a whole selection.
//! It runs on a thread of its own and reaches the interface through the same
//! two mechanisms everything else here does: an `mpsc` channel drained by a
//! Slint timer, and a row in [`crate::activity`].
//!
//! **Why the worker opens its own sessions.** A [`crate::DevelopSession`] owns a GPU
//! `AdjustPass`, and the one the interface is holding is the one the canvas
//! renders from — handing it to a worker would mean the develop view could not
//! draw while a batch ran, which is the freeze this exists to remove. A
//! `GpuContext` is an `Arc` pair over a device and queue and is cheap to
//! clone, so the worker takes a clone and opens each photograph for itself.
//! The cost is a `Demosaicer` and an `AdjustPass` per image rather than one
//! for the run; against a full-resolution decode, render and encode it is
//! small, and it keeps this file out of the pipeline that `develop` owns.
//!
//! **The open image is the exception.** Its edit lives in the interface's
//! session and may not have reached a sidecar yet, so a worker that re-opened
//! the file for itself would export the *saved* version rather than the one on
//! screen. That one frame is therefore rendered by the caller and handed over
//! as [`Source::Rendered`]; everything after the render still moves off the UI
//! thread.
use std::collections::HashSet;
use std::path::{Path, PathBuf};
use std::rc::Rc;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::mpsc::{Receiver, RecvTimeoutError, Sender};
use std::sync::Arc;
use std::time::Duration;
use dr_export::Encoded;
use dr_export::{Encoded, NameContext};
use dr_sync::{RemoteBackend, RemotePath};
use dr_sync_nextcloud::{AppCredentials, NextcloudBackend};
use dr_types::ExportTarget;
use dr_types::{ExportSettings, ExportTarget};
use crate::AppWindow;
/// Where exports wait for a server that is not there yet.
///
@@ -371,6 +403,619 @@ pub fn spawn_upload(
rx
}
// ---------------------------------------------------------------------------
// Batch export (FR-EXP-7)
// ---------------------------------------------------------------------------
/// TRACES: NFR-ARCH-3
/// The stop flag a running batch reads.
///
/// One bit rather than a channel. Cancellation is read thousands of times more
/// often than it is written, and a message would only be seen if the worker
/// happened to be listening at the moment the button was pressed rather than at
/// the next point it is safe to stop.
///
/// `Relaxed` throughout: the flag guards no other data, so there is nothing for
/// an acquire/release pair to publish. The only ordering that matters is that
/// the write eventually becomes visible, which every ordering guarantees.
#[derive(Clone, Default)]
pub struct Cancel(Arc<AtomicBool>);
impl Cancel {
pub fn cancel(&self) {
self.0.store(true, Ordering::Relaxed);
}
pub fn is_cancelled(&self) -> bool {
self.0.load(Ordering::Relaxed)
}
}
/// How long the worker blocks on another worker before rereading the flag.
///
/// TRACES: NFR-ARCH-3
/// This is the cancellation bound for everything the batch spends its time
/// *waiting* on — a download of tens of megabytes would otherwise hold a
/// cancelled batch open until the transfer finished. It is not the bound for
/// the frame being rendered and encoded when the button is pressed: that has no
/// interior stopping point, so the true worst case is one image, and on a 24 MP
/// frame that is well past NFR-ARCH-3's 100 ms target. Closing that gap needs
/// the render itself to become interruptible (NFR-ARCH-2's scheduler), not a
/// finer poll here.
const CANCEL_POLL: Duration = Duration::from_millis(100);
/// How often the drain looks at what the worker has said.
const DRAIN_INTERVAL: Duration = Duration::from_millis(120);
/// Where one image's pixels come from.
pub enum Source {
/// A photograph in the library, fetched and rendered by the worker.
///
/// Carries its own cache context because that is assembled from the catalog
/// and the session, and a worker thread can reach neither.
Library {
path: String,
cache: Option<crate::library::CacheContext>,
},
/// A frame the caller has already rendered — the image open in develop.
///
/// See the module docs for why this one cannot be left to the worker.
Rendered {
stem: String,
frame: dr_export::Frame,
},
}
impl Source {
/// What to call this image in a progress row.
fn describe(&self) -> String {
match self {
Source::Library { path, .. } => path.rsplit('/').next().unwrap_or(path).to_string(),
Source::Rendered { stem, .. } => stem.clone(),
}
}
}
/// Everything a batch needs, assembled on the UI thread.
///
/// Assembled there and not in the worker because most of it is only reachable
/// from a `RefCell` the interface owns: the session, the settings record, the
/// catalog. Passing the finished answers means the worker borrows nothing.
pub struct BatchRequest {
pub sources: Vec<Source>,
/// Credentials for the account the library is open on. `None` where no
/// library is open, which is fine for a [`Source::Rendered`] and fatal for
/// anything that has to be fetched.
pub creds: Option<(AppCredentials, String)>,
pub settings: ExportSettings,
pub outbox: PathBuf,
pub sidecar_cache: PathBuf,
pub offline: bool,
/// `None` on a build with no adapter, where a library image cannot be
/// rendered at all — reported per image rather than refused up front, so
/// the reason lands in the same place every other failure does.
pub gpu: Option<dr_gpu::GpuContext>,
}
/// TRACES: NFR-ARCH-4
/// Why one image of a batch produced no file.
///
/// One variant per stage. "Export failed" repeated forty times is not a report
/// anybody can act on: a destination folder that cannot be written is one fix,
/// a body rawler cannot decode is another, and a server that went away is a
/// third — and only the first is worth stopping the batch to correct.
#[derive(Debug, thiserror::Error)]
pub enum ItemError {
#[error("could not be fetched: {0}")]
Fetch(String),
#[error("could not be opened for editing: {0}")]
Open(String),
#[error("could not be rendered: {0}")]
Render(String),
/// The name was taken and the collision policy is Skip. Not really a
/// failure — the user asked for exactly this — but it is still an image
/// that produced no file, and a batch reporting "40 exported" when it wrote
/// 31 would be lying.
#[error("a file of that name is already there, and the collision setting is Skip")]
NameTaken,
#[error(transparent)]
Encode(#[from] dr_export::ExportError),
#[error("could not be written: {0}")]
Place(String),
}
/// What the worker says as it goes.
#[derive(Debug)]
pub enum BatchMessage {
/// Work has begun on one image; `done` counts the ones behind it.
Started {
done: usize,
total: usize,
name: String,
},
/// One image is finished, for better or worse.
Item {
name: String,
outcome: Result<Placed, ItemError>,
},
Finished {
exported: usize,
failed: usize,
cancelled: bool,
},
}
/// Export a selection on a thread of its own.
pub fn spawn_batch(request: BatchRequest, cancel: Cancel) -> Receiver<BatchMessage> {
let (tx, rx) = std::sync::mpsc::channel();
std::thread::spawn(move || run(request, &cancel, &tx));
rx
}
/// The batch itself, one image at a time.
///
/// Sequential rather than parallel, which FR-EXP-7's "uses all available cores"
/// does not yet get. One reason and one excuse: the reason is that a
/// full-resolution frame is tens of megabytes and four in flight is a
/// straightforward way to exhaust a tablet; the excuse is that the GPU is
/// shared with the interface, and the render is where the time goes.
fn run(mut request: BatchRequest, cancel: &Cancel, tx: &Sender<BatchMessage>) {
let sources = std::mem::take(&mut request.sources);
let total = sources.len();
// Names this run has already written. See [`resolve_batch_name`] for why
// the destination alone is not enough to keep two exports apart.
let mut issued: HashSet<String> = HashSet::new();
let (mut exported, mut failed) = (0usize, 0usize);
for (i, source) in sources.into_iter().enumerate() {
if cancel.is_cancelled() {
break;
}
let name = source.describe();
let _ = tx.send(BatchMessage::Started {
done: i,
total,
name: name.clone(),
});
// `None` is cancellation mid-image, which is not an outcome for this
// photograph: nothing failed, the user simply stopped asking.
let Some(outcome) = export_one(&request, source, i as u32 + 1, &mut issued, cancel) else {
break;
};
// Counted, reported, and then on to the next one. A failure here must
// not end the run — the whole point of exporting three hundred frames
// unattended is that the one unreadable file is a line in a report
// rather than an evening lost (FR-EXP-7).
match &outcome {
Ok(_) => exported += 1,
Err(_) => failed += 1,
}
let _ = tx.send(BatchMessage::Item { name, outcome });
}
let _ = tx.send(BatchMessage::Finished {
exported,
failed,
cancelled: cancel.is_cancelled(),
});
}
/// One image, start to finish. `None` where the run was cancelled part way.
fn export_one(
request: &BatchRequest,
source: Source,
sequence: u32,
issued: &mut HashSet<String>,
cancel: &Cancel,
) -> Option<Result<Placed, ItemError>> {
let (stem, date, frame) = match source {
Source::Rendered { stem, frame } => (stem, String::new(), frame),
Source::Library { path, cache } => {
match render_from_library(request, &path, cache, cancel)? {
Ok(rendered) => rendered,
Err(e) => return Some(Err(e)),
}
}
};
Some(place_frame(request, &stem, &date, sequence, &frame, issued))
}
/// Fetch a photograph, apply its stored edit, and render it at full size.
fn render_from_library(
request: &BatchRequest,
path: &str,
cache: Option<crate::library::CacheContext>,
cancel: &Cancel,
) -> Option<Result<(String, String, dr_export::Frame), ItemError>> {
let Some((creds, user_id)) = request.creds.clone() else {
return Some(Err(ItemError::Fetch("no library is open".into())));
};
let Some(gpu) = request.gpu.as_ref() else {
return Some(Err(ItemError::Open(
"this build found no GPU adapter".into(),
)));
};
// Both started before either is waited on, exactly as opening an image in
// develop does: the sidecar is three orders of magnitude smaller than the
// RAW, so it costs nothing to have in hand by the time there is a session
// to apply it to.
let sidecar_rx = crate::library::spawn_sidecar_fetch(
creds.clone(),
user_id.clone(),
path.to_string(),
request.sidecar_cache.clone(),
request.offline,
);
let bytes_rx = crate::library::spawn_full_fetch(creds, user_id, path.to_string(), cache);
let bytes = match wait_for(&bytes_rx, cancel) {
Waited::Got(Ok(bytes)) => bytes,
Waited::Got(Err(e)) => return Some(Err(ItemError::Fetch(e.message))),
Waited::Cancelled => return None,
Waited::Silent => {
return Some(Err(ItemError::Fetch(
"the download ended without answering".into(),
)))
}
};
let sidecar = match wait_for(&sidecar_rx, cancel) {
Waited::Got(sidecar) => sidecar,
Waited::Cancelled => return None,
// An unedited photograph has no sidecar and a fetch that died looks
// identical from here. Exporting at defaults is what opening it would
// do, and refusing the image over a missing edit it may never have had
// would fail the common case.
Waited::Silent => None,
};
let meta = dr_decode::metadata(&bytes).unwrap_or_default();
let orientation = meta.orientation.unwrap_or_default();
// `{date}` is the capture date, not today's: a template naming exports by
// when the shutter fired is the reason the token exists.
let date = meta
.captured_at
.map(crate::library_ui::format_date)
.unwrap_or_default();
let mut session = match crate::open_session(gpu, &bytes, orientation) {
Ok(session) => session,
Err(e) => return Some(Err(ItemError::Open(e))),
};
// TRACES: FR-CAT-8
// The stored edit. FR-EXP-9 is about resolution; this is the other half of
// exporting what the user actually has, and a batch that skipped it would
// write out three hundred unedited frames without saying so.
if let Some(version) = sidecar
.as_ref()
.and_then(dr_pipeline::Sidecar::default_version)
{
session.apply_version(version);
}
// The last cheap place to stop. Everything past here is a full-resolution
// render and an encode with no interior stopping point — see [`CANCEL_POLL`].
if cancel.is_cancelled() {
return None;
}
let frame = match session.render_for_export() {
Ok(frame) => frame,
Err(e) => return Some(Err(ItemError::Render(e))),
};
let stem = Path::new(path)
.file_stem()
.map(|s| s.to_string_lossy().into_owned())
.unwrap_or_else(|| "export".into());
Some(Ok((stem, date, frame)))
}
/// Name, encode and write one rendered frame.
///
/// The tail every source shares, whoever rendered it.
fn place_frame(
request: &BatchRequest,
stem: &str,
date: &str,
sequence: u32,
frame: &dr_export::Frame,
issued: &mut HashSet<String>,
) -> Result<Placed, ItemError> {
// The size is resolved before the name because `{dimensions}` is one of the
// tokens a template can carry.
let (width, height) = dr_export::target_size(
frame.width,
frame.height,
request.settings.sizing,
request.settings.allow_upscaling,
);
let ctx = NameContext {
source_stem: stem,
sequence,
date,
width,
height,
preset: "",
};
let name = resolve_batch_name(&request.settings, &ctx, issued).ok_or(ItemError::NameTaken)?;
let encoded = dr_export::export(frame, &request.settings, name)?;
place(
&encoded,
request.settings.target,
&request.settings.destination,
&request.outbox,
)
.map_err(ItemError::Place)
}
/// The name this export takes, avoiding both what was in the destination and
/// what this run has already written.
///
/// Those are two different collisions and they deserve two different answers.
/// [`dr_types::CollisionPolicy`] is the user's answer to "a file of this name
/// was already there", and Overwrite is a perfectly reasonable one. It is not
/// an answer to "the frame I exported four seconds ago was also called this":
/// two photographs of the same stem in different folders are ordinary in a
/// library, and a batch that quietly handed back fewer files than images —
/// having destroyed its own output — is not something anybody asked for. So a
/// name this run issued is always stepped past, whatever the policy says about
/// the folder.
fn resolve_batch_name(
settings: &ExportSettings,
ctx: &NameContext<'_>,
issued: &mut HashSet<String>,
) -> Option<String> {
let dir = PathBuf::from(&settings.destination);
let taken = |name: &str| -> bool {
match settings.target {
dr_types::ExportTarget::Device => dir.join(name).exists(),
// A queued export cannot see the server, and may never be able to.
// Names are kept apart in the outbox instead — see [`stage`].
dr_types::ExportTarget::Remote => false,
}
};
let name = dr_export::resolve_name(
&settings.filename_template,
ctx,
settings.format,
settings.collision,
&|name| taken(name) || issued.contains(name),
)?;
// Only reachable under Overwrite, which hands back the taken name by
// design. Skip and Increment have already been through the closure above.
let name = if issued.contains(&name) {
step_past(&name, &|candidate| {
taken(candidate) || issued.contains(candidate)
})?
} else {
name
};
issued.insert(name.clone());
Some(name)
}
/// `photo.jpg` → `photo-1.jpg`, and on until the name is free.
fn step_past(name: &str, taken: &dyn Fn(&str) -> bool) -> Option<String> {
let path = Path::new(name);
let stem = path
.file_stem()
.map(|s| s.to_string_lossy().into_owned())
.unwrap_or_else(|| "export".into());
let ext = path
.extension()
.map(|s| s.to_string_lossy().into_owned())
.unwrap_or_default();
// Bounded for the same reason `resolve_name`'s own search is: a destination
// that reports every name as taken has to fail rather than spin.
(1..10_000)
.map(|n| {
if ext.is_empty() {
format!("{stem}-{n}")
} else {
format!("{stem}-{n}.{ext}")
}
})
.find(|candidate| !taken(candidate))
}
/// What waiting on another worker produced.
///
/// Three answers rather than an `Option`, because "the user cancelled" and "the
/// worker died without a word" have the same shape and want opposite treatment:
/// one ends the batch quietly, the other is a failure belonging to one image.
enum Waited<T> {
Got(T),
Cancelled,
Silent,
}
/// Block on another worker's answer without going deaf to cancellation.
fn wait_for<T>(rx: &Receiver<T>, cancel: &Cancel) -> Waited<T> {
loop {
match rx.recv_timeout(CANCEL_POLL) {
Ok(value) => return Waited::Got(value),
Err(RecvTimeoutError::Timeout) => {
if cancel.is_cancelled() {
return Waited::Cancelled;
}
}
Err(RecvTimeoutError::Disconnected) => return Waited::Silent,
}
}
}
/// Which status line a batch writes its commentary to.
///
/// One drain serves both entry points rather than two near-identical timers, so
/// it has to be told where the words go: develop has its own line beside the
/// export button, and the grid has the header's.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Reporting {
Develop,
Library,
}
/// TRACES: FR-EXP-7 | NFR-P9
/// Drain a batch's progress on the UI thread.
///
/// `slot` holds the timer, and dropping what was there before is what stops a
/// superseded batch's drain — and, through [`crate::activity::Activity`]'s
/// `Drop`, takes its row with it.
pub fn drain_batch(
weak: slint::Weak<AppWindow>,
activity: &Rc<crate::activity::ActivityLog>,
slot: &Rc<std::cell::RefCell<Option<slint::Timer>>>,
rx: Receiver<BatchMessage>,
total: usize,
reporting: Reporting,
) {
let job = activity.begin(
crate::activity::Kind::Export,
if total == 1 {
"Exporting".to_string()
} else {
format!("Exporting {total} images")
},
);
job.total(total);
// The first failure, kept for the summary. Only the first: a run where
// every image failed for the same reason should say that reason once, and
// the rest are in the log.
let mut first_failure: Option<String> = None;
let timer = slint::Timer::default();
let held = slot.clone();
timer.start(slint::TimerMode::Repeated, DRAIN_INTERVAL, move || {
let Some(w) = weak.upgrade() else { return };
loop {
let message = match rx.try_recv() {
Ok(m) => m,
Err(std::sync::mpsc::TryRecvError::Empty) => return,
Err(std::sync::mpsc::TryRecvError::Disconnected) => {
// A worker that died without a word must not leave the
// button saying "Cancel export" for the rest of the session.
job.fail("ended unexpectedly");
settle(&w, reporting, "Export ended unexpectedly");
stop_timer(&held);
return;
}
};
match message {
BatchMessage::Started { done, total, name } => {
job.progress(done, total);
job.detail(name.clone());
report(
&w,
reporting,
&format!("Exporting {name} ({}/{total})", done + 1),
);
}
BatchMessage::Item { name, outcome } => match outcome {
Ok(placed) => log::info!("{name}: {}", placed.describe()),
Err(e) => {
// Every failure is logged, not only the first: the
// summary is a sentence and this is the record of which
// photographs it is about (NFR-ARCH-4).
log::warn!("exporting {name}: {e}");
first_failure.get_or_insert_with(|| format!("{name}: {e}"));
}
},
BatchMessage::Finished {
exported,
failed,
cancelled,
} => {
let (text, is_failure) =
summarise(exported, failed, cancelled, first_failure.as_deref());
if is_failure {
job.fail(text.clone());
} else {
job.finish(text.clone());
}
settle(&w, reporting, &text);
stop_timer(&held);
return;
}
}
}
});
*slot.borrow_mut() = Some(timer);
}
/// How a finished batch reads, and whether it counts as a failure.
///
/// A cancelled run is never a failure however little it exported: the user
/// stopped it, and a red row telling them so is the application arguing. A run
/// with even one failure is, and stays in the list until it is cleared — that
/// is the row somebody came looking for.
fn summarise(
exported: usize,
failed: usize,
cancelled: bool,
first_failure: Option<&str>,
) -> (String, bool) {
if cancelled {
return (format!("Cancelled after {exported}"), false);
}
if failed == 0 {
return (format!("Exported {exported}"), false);
}
let detail = first_failure.unwrap_or("see the log");
(
format!("Exported {exported}, {failed} failed — {detail}"),
true,
)
}
/// Say what the batch is doing on the line the caller came from.
fn report(window: &AppWindow, reporting: Reporting, text: &str) {
match reporting {
Reporting::Develop => window.set_export_status(text.into()),
Reporting::Library => window.set_library_status(text.into()),
}
}
/// Report, and put the buttons back.
fn settle(window: &AppWindow, reporting: Reporting, text: &str) {
report(window, reporting, text);
match reporting {
Reporting::Develop => window.set_export_busy(false),
Reporting::Library => window.set_library_exporting(false),
}
}
fn stop_timer(slot: &Rc<std::cell::RefCell<Option<slint::Timer>>>) {
if let Some(timer) = slot.borrow().as_ref() {
timer.stop();
}
}
#[cfg(test)]
mod tests {
use super::*;
@@ -596,4 +1241,306 @@ mod tests {
.describe()
.contains("Exported"));
}
// -----------------------------------------------------------------------
// The batch (FR-EXP-7)
// -----------------------------------------------------------------------
/// Settings that write PNGs into `dir`, so a test can look at what landed.
fn to_folder(dir: &Path) -> ExportSettings {
ExportSettings {
format: dr_types::ExportFormat::Png,
target: ExportTarget::Device,
destination: dir.to_string_lossy().into_owned(),
..Default::default()
}
}
fn request(settings: ExportSettings, sources: Vec<Source>) -> BatchRequest {
BatchRequest {
sources,
creds: None,
settings,
outbox: std::env::temp_dir().join("dr-batch-test-outbox"),
sidecar_cache: std::env::temp_dir().join("dr-batch-test-sidecars"),
offline: true,
gpu: None,
}
}
/// A frame of flat pixels — enough for the encoder, cheap for a test.
fn frame(width: u32, height: u32) -> dr_export::Frame {
dr_export::Frame::new(width, height, vec![128; (width * height * 4) as usize])
.expect("well-formed")
}
fn drive(request: BatchRequest, cancel: Cancel) -> Vec<BatchMessage> {
let (tx, rx) = std::sync::mpsc::channel();
run(request, &cancel, &tx);
drop(tx);
rx.into_iter().collect()
}
fn finished(messages: &[BatchMessage]) -> (usize, usize, bool) {
messages
.iter()
.find_map(|m| match m {
BatchMessage::Finished {
exported,
failed,
cancelled,
} => Some((*exported, *failed, *cancelled)),
_ => None,
})
.expect("a batch always says it has finished")
}
#[test]
fn a_rendered_frame_reaches_the_destination_folder() {
// The whole tail of the batch — name, size, sharpen, encode, write —
// with no GPU and no network, which is what makes it testable at all.
let dir = tmp();
let out = dir.join("exports");
let messages = drive(
request(
to_folder(&out),
vec![Source::Rendered {
stem: "IMG_0001".into(),
frame: frame(16, 12),
}],
),
Cancel::default(),
);
assert_eq!(finished(&messages), (1, 0, false));
assert!(out.join("IMG_0001.png").exists());
}
#[test]
fn two_images_of_one_name_both_survive_the_batch() {
// Two folders in a library holding an IMG_0001 each is ordinary, and a
// batch that wrote one file for two photographs would destroy work
// without saying anything. Overwrite is set deliberately: it is the
// user's answer about the *folder*, not about this run's own output.
let dir = tmp();
let out = dir.join("exports");
let mut settings = to_folder(&out);
settings.collision = dr_types::CollisionPolicy::Overwrite;
let messages = drive(
request(
settings,
vec![
Source::Rendered {
stem: "IMG_0001".into(),
frame: frame(16, 12),
},
Source::Rendered {
stem: "IMG_0001".into(),
frame: frame(16, 12),
},
],
),
Cancel::default(),
);
assert_eq!(finished(&messages), (2, 0, false));
assert!(out.join("IMG_0001.png").exists());
assert!(out.join("IMG_0001-1.png").exists());
}
#[test]
fn a_name_that_was_already_there_still_obeys_the_collision_setting() {
// The other half of the rule above: a file that existed *before* the
// batch is exactly what the policy is about, and Overwrite must still
// mean overwrite or the setting would do nothing.
let dir = tmp();
let out = dir.join("exports");
std::fs::create_dir_all(&out).expect("temp dir");
std::fs::write(out.join("IMG_0001.png"), b"older").expect("seed");
let mut settings = to_folder(&out);
settings.collision = dr_types::CollisionPolicy::Overwrite;
let mut issued = HashSet::new();
let name = resolve_batch_name(
&settings,
&NameContext {
source_stem: "IMG_0001",
sequence: 1,
..Default::default()
},
&mut issued,
);
assert_eq!(name.as_deref(), Some("IMG_0001.png"));
}
#[test]
fn a_skipped_name_is_reported_rather_than_silently_dropped() {
// Skip is a legitimate answer, but the image still produced no file —
// and a batch claiming to have exported it would be lying.
let dir = tmp();
let out = dir.join("exports");
std::fs::create_dir_all(&out).expect("temp dir");
std::fs::write(out.join("IMG_0001.png"), b"older").expect("seed");
let mut settings = to_folder(&out);
settings.collision = dr_types::CollisionPolicy::Skip;
let messages = drive(
request(
settings,
vec![Source::Rendered {
stem: "IMG_0001".into(),
frame: frame(8, 8),
}],
),
Cancel::default(),
);
assert_eq!(finished(&messages), (0, 1, false));
assert!(matches!(
messages.iter().find_map(|m| match m {
BatchMessage::Item { outcome, .. } => Some(outcome),
_ => None,
}),
Some(Err(ItemError::NameTaken))
));
assert_eq!(std::fs::read(out.join("IMG_0001.png")).unwrap(), b"older");
}
#[test]
fn one_failure_does_not_abandon_the_rest_of_the_batch() {
// FR-EXP-7's central promise. The first source cannot be fetched — no
// library is open — and the two after it must still be attempted and
// still be counted.
let dir = tmp();
let out = dir.join("exports");
let messages = drive(
request(
to_folder(&out),
vec![
Source::Library {
path: "Photos/broken.CR2".into(),
cache: None,
},
Source::Rendered {
stem: "good-a".into(),
frame: frame(8, 8),
},
Source::Rendered {
stem: "good-b".into(),
frame: frame(8, 8),
},
],
),
Cancel::default(),
);
assert_eq!(finished(&messages), (2, 1, false));
assert!(out.join("good-a.png").exists());
assert!(out.join("good-b.png").exists());
}
#[test]
fn every_image_gets_its_own_outcome() {
// The report is per image, not one verdict for the run: two failures
// for two different reasons have to arrive as two messages, or a
// three-hundred-frame batch is unreportable (NFR-ARCH-4).
let dir = tmp();
let messages = drive(
request(
to_folder(&dir.join("exports")),
vec![
Source::Library {
path: "Photos/a.CR2".into(),
cache: None,
},
Source::Rendered {
stem: "b".into(),
frame: frame(8, 8),
},
],
),
Cancel::default(),
);
let outcomes: Vec<&Result<Placed, ItemError>> = messages
.iter()
.filter_map(|m| match m {
BatchMessage::Item { outcome, .. } => Some(outcome),
_ => None,
})
.collect();
assert_eq!(outcomes.len(), 2);
assert!(outcomes[0].is_err());
assert!(outcomes[1].is_ok());
}
#[test]
fn a_cancelled_batch_stops_and_writes_nothing_more() {
// TRACES: NFR-ARCH-3
// Cancelled before it began, which is the strongest form of the
// property: not one file, and the run still reports itself finished
// rather than leaving the interface waiting for a message.
let dir = tmp();
let out = dir.join("exports");
let cancel = Cancel::default();
cancel.cancel();
let messages = drive(
request(
to_folder(&out),
vec![Source::Rendered {
stem: "IMG_0001".into(),
frame: frame(8, 8),
}],
),
cancel,
);
assert_eq!(finished(&messages), (0, 0, true));
assert!(!out.join("IMG_0001.png").exists());
}
#[test]
fn a_cancelled_wait_gives_up_instead_of_blocking_for_ever() {
// TRACES: NFR-ARCH-3
// The bound on cancelling a batch that is waiting on a download. With
// a plain `recv` this test would hang, which is precisely the bug.
let (tx, rx) = std::sync::mpsc::channel::<u8>();
let cancel = Cancel::default();
cancel.cancel();
assert!(matches!(wait_for(&rx, &cancel), Waited::Cancelled));
drop(tx);
}
#[test]
fn a_worker_that_dies_is_not_mistaken_for_a_cancellation() {
// The two look identical from a channel and mean opposite things: one
// ends the batch, the other fails one image and moves on.
let (tx, rx) = std::sync::mpsc::channel::<u8>();
drop(tx);
assert!(matches!(wait_for(&rx, &Cancel::default()), Waited::Silent));
}
#[test]
fn a_cancelled_run_is_not_reported_as_a_failure() {
// The user stopped it. A red row saying so is the application arguing
// with something it was told to do.
let (text, failed) = summarise(5, 0, true, None);
assert!(!failed, "{text}");
assert!(text.contains('5'), "{text}");
}
#[test]
fn a_run_with_a_failure_keeps_its_row_and_names_one() {
// This is the row somebody comes to the activity list to find, so it
// has to survive being trimmed — and it has to say which photograph.
let (text, failed) = summarise(38, 2, false, Some("IMG_0007.CR2: could not be rendered"));
assert!(failed);
assert!(text.contains("IMG_0007.CR2"), "{text}");
assert!(text.contains("38"), "{text}");
}
}