Skip to main content

file_engine/
progress.rs

1use std::path::PathBuf;
2
3use tokio::sync::mpsc;
4
5use crate::profiler::Entry;
6
7/// Discrete per-entry events rather than a cumulative snapshot;
8/// `EntryFailed` carries only the `Entry`, not the `Error` — `Error` isn't
9/// `Clone` (it wraps `std::io::Error`), and the failure detail is already
10/// available from the operation's final `OperationOutcome.failed` once the
11/// handle resolves.
12/// `#[non_exhaustive]`: adding a variant here is otherwise a breaking
13/// change for any downstream exhaustive `match`, which is exactly what
14/// adding `Planned` was. Marked now so the next addition isn't.
15#[derive(Debug, Clone)]
16#[non_exhaustive]
17pub enum Progress {
18    /// The shape of the work about to be performed, emitted once per
19    /// phase *before* `DirectoriesStarted` and `Started` — i.e. before
20    /// the directory pre-pass that `Started` doesn't cover.
21    ///
22    /// Exists for cost estimation (`EtaEstimator`): the small/large
23    /// split is the difference between work whose cost is per-file
24    /// (syscall-bound) and work whose cost is per-byte (bandwidth-bound),
25    /// and `Started`'s single `bytes_total` can't distinguish them. A
26    /// consumer that only wants a progress bar can ignore this variant
27    /// entirely.
28    ///
29    /// Not emitted by the delete sweeps (`sync`'s orphan sweep,
30    /// `move_path`'s source cleanup) — those are metadata-only phases
31    /// with no byte-sized work to model, so they emit a bare `Started`.
32    Planned {
33        /// Directories needing an explicit `create_dir_all` — already
34        /// filtered to those with no file beneath them (see
35        /// `pipeline.rs`'s `directories_covered_by_files`), so this
36        /// matches the `DirectoriesStarted { total }` that follows
37        /// rather than the total directory count in the source tree.
38        directories: usize,
39        small_files: usize,
40        small_bytes: u64,
41        large_files: usize,
42        large_bytes: u64,
43        /// The threshold that produced the split above. Carried so a
44        /// consumer can classify each subsequent `Entry` the same way
45        /// the planner did, without having to know what the builder was
46        /// configured with.
47        small_file_threshold: u64,
48    },
49    /// Emitted once per phase, before any entries in that phase start.
50    /// `bytes_total` is `None` for phases with nothing byte-sized to
51    /// report (the delete sweeps). Can be emitted more than once per
52    /// operation — `sync` emits it once per phase (copy, then delete).
53    Started {
54        bytes_total: Option<u64>,
55        entries_total: usize,
56    },
57    EntryStarted {
58        entry: Entry,
59    },
60    /// Bytes written so far for an entry still in flight, sampled by
61    /// watching the destination file grow. Emitted only for large
62    /// (streamed) entries, and only while they take long enough to be
63    /// sampled at all — a copy that the filesystem satisfies by
64    /// copy-on-write finishes before the first sample and emits none.
65    ///
66    /// `bytes_copied` is cumulative, not a delta, and is clamped to the
67    /// entry's size. It is monotonically non-decreasing per entry.
68    ///
69    /// Exists because `tokio::fs::copy` is opaque while it runs: without
70    /// this, a single large file emits `EntryStarted` and then nothing
71    /// until it finishes, so its transfer rate is unmeasurable for exactly
72    /// as long as it takes to copy.
73    EntryProgress {
74        entry: Entry,
75        bytes_copied: u64,
76    },
77    EntryCompleted {
78        entry: Entry,
79    },
80    /// Emitted instead of `EntryCompleted` when the destination already
81    /// held byte-identical content, so the entry was left untouched
82    /// rather than copied (`.skip_if_identical()`, `checksum` feature).
83    EntrySkipped {
84        entry: Entry,
85    },
86    EntryFailed {
87        entry: Entry,
88    },
89    /// The directory-creation pre-pass (`operations/pipeline.rs`'s
90    /// `ensure_directories_exist`), separate from `Started`/`Entry*`
91    /// since it operates on `DirEntry`, not `Entry` — carrying only the
92    /// destination path (not the full `DirEntry`) is enough for a
93    /// caller to show "N/M directories created" without extending
94    /// every `Entry`-typed variant to also accept `DirEntry`. Added
95    /// after a real run against a USB-connected exFAT drive spent about
96    /// a minute silently creating ~7,700 directories one at a time
97    /// before `Started` (for the file-copy phase) was ever emitted.
98    DirectoriesStarted {
99        total: usize,
100    },
101    DirectoryCompleted {
102        path: PathBuf,
103    },
104    DirectoryFailed {
105        path: PathBuf,
106    },
107}
108
109/// `Send + Sync + Clone` sender wrapper threaded into every execution
110/// path that processes entries. Backed by an unbounded channel
111/// specifically so `.send()` is synchronous, not `async` — needed
112/// because `compress.rs`'s writer runs inside `spawn_blocking`, a
113/// non-async context that can't await a bounded channel's backpressure.
114#[derive(Clone)]
115pub(crate) struct ProgressReporter {
116    tx: mpsc::UnboundedSender<Progress>,
117}
118
119impl ProgressReporter {
120    pub(crate) fn new(tx: mpsc::UnboundedSender<Progress>) -> Self {
121        Self { tx }
122    }
123
124    /// A reporter whose receiver is immediately dropped — for tests that
125    /// don't care about progress and don't want to plumb a receiver
126    /// through just to ignore it.
127    #[cfg(test)]
128    pub(crate) fn noop() -> Self {
129        let (tx, _rx) = mpsc::unbounded_channel();
130        Self { tx }
131    }
132
133    /// A closed receiver means nobody's listening (the caller never
134    /// called `.progress()`, or dropped the stream) — not a failure,
135    /// just nothing to report to.
136    pub(crate) fn send(&self, progress: Progress) {
137        let _ = self.tx.send(progress);
138    }
139}
140
141#[cfg(test)]
142mod tests {
143    use std::path::PathBuf;
144
145    use super::*;
146
147    fn entry(name: &str) -> Entry {
148        Entry {
149            path: PathBuf::from(name),
150            relative_path: PathBuf::from(name),
151            size: 1,
152            modified: None,
153        }
154    }
155
156    #[test]
157    fn send_after_receiver_dropped_does_not_panic() {
158        let (tx, rx) = mpsc::unbounded_channel();
159        let reporter = ProgressReporter::new(tx);
160        drop(rx);
161
162        reporter.send(Progress::EntryStarted { entry: entry("a") });
163    }
164
165    #[tokio::test]
166    async fn sends_are_received_in_order() {
167        let (tx, mut rx) = mpsc::unbounded_channel();
168        let reporter = ProgressReporter::new(tx);
169
170        reporter.send(Progress::Started {
171            bytes_total: Some(10),
172            entries_total: 2,
173        });
174        reporter.send(Progress::EntryStarted { entry: entry("a") });
175        reporter.send(Progress::EntryCompleted { entry: entry("a") });
176
177        drop(reporter); // otherwise the last recv() below blocks forever
178
179        assert!(matches!(rx.recv().await, Some(Progress::Started { .. })));
180        assert!(matches!(
181            rx.recv().await,
182            Some(Progress::EntryStarted { .. })
183        ));
184        assert!(matches!(
185            rx.recv().await,
186            Some(Progress::EntryCompleted { .. })
187        ));
188        assert!(rx.recv().await.is_none());
189    }
190}