Treat a placeholder as the photograph, not as a one-byte file

The folder connector was pointed at a Nextcloud VFS tree and got three
things wrong, the first of which loses work.

**A dehydrated sidecar read as absent.** `a.drsc` does not exist when the
client has dehydrated it — only `a.drsc.nextcloud` does — so `get` missed,
`.ok()` swallowed the `NotFound`, and the sidecar writer took that for
"there is no sidecar yet" and wrote a fresh document over the existing
one. Every edit another device had put there went with it. That function's
own doc comment calls this the exact loss the format's unknown-key
preservation exists to prevent.

**A stub was catalogued as a 1-byte image**, and ARCH §9.0 measured this
machine at 121,785 placeholders against 10,267 real files — so a folder
library on a synced tree was ~92% broken rows.

**Identity changed on hydration**, so downloading a photograph looked like
a delete and an add, orphaning its thumbnail and its face rows.

Entries now carry the photograph's own name and a `materialised` flag;
`get` on a stub returns the new `RemoteError::NotMaterialised`, which is
distinct from `NotFound` precisely because the sidecar writer must treat
them differently — it fetches the sidecar and merges, or leaves the entry
queued.

Hydration is a **borrow**. `BorrowPool` records what was on disk before it
asked, so `release_all` dehydrates only what a pass brought and leaves
what the user already had. Reference counted: the thumbnail pass and the
face pass meet on the same RAW, and without counting the first to finish
dehydrates the file the second is reading. A borrow against a plain folder
or a server does nothing, so a pass written for VFS runs everywhere.

Releasing means asking the client to dehydrate and never deleting: a
deletion inside a synced tree propagates to the server and removes the
photograph from every device.

Not a second backend — the capability is per *connection*, not per type,
since the same folder hydrates only while the client runs. The convention
arrives through a detector the registry supplies, so `dr-sync-folder`
still knows nothing about any client's protocol.

ARCH §9.0a records this as an amendment: finding 3 rejected hydration
because it costs 100× a range read, and that comparison assumed a
connector was available. A folder library has none.
This commit is contained in:
2026-08-29 09:57:52 +02:00
parent 6c363cee97
commit c102ba9df2
22 changed files with 1555 additions and 53 deletions
+49
View File
@@ -38,6 +38,50 @@ pub struct ChunkConstraints {
pub max_chunks: u32,
}
/// TRACES: FR-NC-6c
/// Whether every listed object's content is actually reachable.
///
/// Every backend but a virtual-filesystem folder answers [`Always`](Self::Always).
/// A VFS folder is the case this exists for: the sync client leaves a
/// placeholder where a file is catalogued but not downloaded, so the name is
/// listable and the bytes are not (ARCH §9.0).
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Materialisation {
/// Listing an object means its content can be read. Every server backend,
/// and a plain directory.
Always,
/// Some objects are placeholders, and nothing this process can do will
/// change that — the sync client is not running, or the platform offers no
/// way to ask. Such an object reads as
/// [`RemoteError::NotMaterialised`](crate::RemoteError::NotMaterialised)
/// and is shown as offline rather than broken.
Placeholders,
/// Some objects are placeholders, and this backend can ask for their
/// content — and give it back.
///
/// **Whole-file, and that is the whole difficulty.** Hydration has two
/// states, one byte or all bytes, so using it to fill a grid transfers the
/// entire library to produce thumbnails (ARCH §9.0). It belongs to the
/// originals tier — an image opened in develop, exported, or deliberately
/// pinned — and to passes the user has asked for and been quoted a price
/// on. Never to browsing.
OnDemand,
}
impl Materialisation {
/// Whether content can be fetched on request.
pub fn can_materialise(self) -> bool {
matches!(self, Materialisation::OnDemand)
}
/// Whether some objects may have no content locally.
pub fn has_placeholders(self) -> bool {
!matches!(self, Materialisation::Always)
}
}
/// TRACES: FR-NC-3
/// Whether the server can render thumbnails, and for what.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -69,6 +113,8 @@ pub struct Capabilities {
/// Conditional write (If-Match), for conflict-safe sidecar updates.
pub conditional_write: bool,
pub server_previews: ServerPreviews,
/// Whether a listed object's content is necessarily present.
pub materialisation: Materialisation,
}
impl Capabilities {
@@ -83,6 +129,9 @@ impl Capabilities {
bulk_upload: false,
conditional_write: false,
server_previews: ServerPreviews::None,
// The weakest backend still answers for everything it lists;
// placeholders are a property a backend opts into.
materialisation: Materialisation::Always,
}
}
+16
View File
@@ -36,6 +36,22 @@ pub enum RemoteError {
#[error("not found: {0}")]
NotFound(String),
/// TRACES: FR-NC-6c
/// The object exists, but its content is not on this device.
///
/// A virtual-filesystem placeholder: the sync client holds the name and a
/// stub, and the bytes are still on the server (ARCH §9.0).
///
/// **Emphatically not [`NotFound`](Self::NotFound), and the distinction is
/// what stops a silent data loss.** The sidecar writer reads before it
/// writes, and treats a miss as "there is no sidecar yet, create one" — so
/// a dehydrated sidecar reported as absent makes it write a fresh document
/// over an existing one, discarding every edit another device had put
/// there. It is also the difference between an error a user can act on
/// (fetch it) and one they cannot (it is gone).
#[error("not on this device: {0}")]
NotMaterialised(String),
/// The backend does not support this operation. Expected, not a bug —
/// callers check capabilities and adapt.
#[error("operation unsupported by this backend: {0}")]
+46 -1
View File
@@ -35,7 +35,9 @@ pub mod types;
pub mod upload;
pub use account::{Account, AccountError, AccountStore, Connection, Secret, LEGACY_BACKEND};
pub use capability::{Capabilities, ChangeDetection, ChunkConstraints, ServerPreviews};
pub use capability::{
Capabilities, ChangeDetection, ChunkConstraints, Materialisation, ServerPreviews,
};
pub use error::RemoteError;
pub use provider::{BackendProvider, BackendRegistry, SignIn};
pub use reachability::{Connectivity, Reachability};
@@ -151,6 +153,49 @@ pub trait RemoteBackend: Send + Sync {
/// destination, not to claim they created it.
async fn create_dir(&self, path: &RemotePath) -> Result<(), RemoteError>;
// ---- materialisation --------------------------------------------------
/// TRACES: FR-NC-6c
/// Ask for a placeholder's content to be brought to this device.
///
/// Only meaningful where [`Capabilities::materialisation`] is
/// [`Materialisation::OnDemand`]; others return
/// [`RemoteError::Unsupported`].
///
/// **Whole-file, and slow.** There is no partial hydration: a placeholder
/// becomes one byte or all of them, so this transfers a 27 MB RAW to
/// answer a question a 256 KB range read would have answered (ARCH §9.0
/// finding 3). It is for the originals tier — develop, export, a pin the
/// user asked for — and for passes the user has been quoted a price on and
/// agreed to. **Never for filling a grid**: doing so downloads the entire
/// library to produce thumbnails.
///
/// Returns once the content is readable. Callers that borrowed it should
/// hand it back with [`dematerialise`](Self::dematerialise).
async fn materialise(&self, _id: &RemoteId) -> Result<(), RemoteError> {
Err(RemoteError::Unsupported(
"this backend has no placeholders to materialise",
))
}
/// Give a placeholder's content back, freeing the disk it held.
///
/// The counterpart that makes hydration a *borrow* rather than an
/// acquisition: a pass that hydrates a library to index it can return each
/// file as it finishes, so peak disk is the working set rather than the
/// library.
///
/// **Never destructive.** On a synced folder this asks the client to
/// dehydrate; it must not delete, because a deletion in a synced tree
/// propagates to the server and removes the photograph everywhere. An
/// implementation that cannot dehydrate must return
/// [`RemoteError::Unsupported`] rather than approximating it.
async fn dematerialise(&self, _id: &RemoteId) -> Result<(), RemoteError> {
Err(RemoteError::Unsupported(
"this backend has no placeholders to release",
))
}
// ---- optional ---------------------------------------------------------
/// Server-rendered thumbnail, where available.
+3
View File
@@ -253,6 +253,7 @@ mod tests {
size: 0,
modified: None,
has_preview: false,
materialised: true,
}
}
@@ -265,6 +266,7 @@ mod tests {
size: 1000,
modified: None,
has_preview: false,
materialised: true,
}
}
@@ -301,6 +303,7 @@ mod tests {
bulk_upload: false,
conditional_write: true,
server_previews: ServerPreviews::None,
materialisation: crate::Materialisation::Always,
},
lists: RefCell::new(0),
probes: RefCell::new(0),
+17
View File
@@ -97,6 +97,23 @@ pub struct RemoteEntry {
/// Whether the server claims a renderable preview exists. Advisory: stock
/// Nextcloud reports none for RAW (ARCH §6.7).
pub has_preview: bool,
/// TRACES: FR-NC-6c
/// Whether [`get`](crate::RemoteBackend::get) can produce this object's
/// content right now.
///
/// True for everything a server backend lists — the bytes are remote, but
/// they are reachable. False only for a virtual-filesystem placeholder,
/// where the name is on this device and the content is not (ARCH §9.0).
///
/// The catalog maps this to [`Availability::Offline`](dr_types::Availability),
/// which is the difference between a photograph shown as *not downloaded*
/// and one shown as broken.
///
/// **`size` is not meaningful when this is false.** A Linux suffix-mode
/// stub is one byte and carries no record of what it stands for, so there
/// is nothing to report but zero.
pub materialised: bool,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+1
View File
@@ -237,6 +237,7 @@ mod tests {
size: *size,
modified: None,
has_preview: false,
materialised: true,
})
.collect())
}