Merge branch 'fix/sim-flakes-and-open-race' into chore/sdk-0.10.1

# Conflicts:
#	docs/BUGS.md

Cargo.lock

@@ -3779,9 +3779,9 @@ dependencies = [
 
 [[package]]
 name = "proton-drive-rs"
-version = "0.10.0"
+version = "0.10.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8f70318394b2815b009c52f804d75747086db93a338ceb6eae62541f2b25d5bf"
+checksum = "a8fa2a88130883010b32035e208aaa36fb275f6d7d78ed44a1069b1d33d24481"
 dependencies = [
  "async-trait",
  "base64 0.23.1",
@@ -3806,9 +3806,9 @@ dependencies = [
 
 [[package]]
 name = "proton-sdk"
-version = "0.10.0"
+version = "0.10.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4196392032f56b3bbaac0bb1f24eea62fa7e2a0af72004de8d08e518d14720fb"
+checksum = "af0bf8d50fa829d1a65b09c6855b606578d02734d4089874c971d1063c23c353"
 dependencies = [
  "aes-gcm 0.11.1",
  "async-trait",

Cargo.toml

@@ -15,8 +15,8 @@ keywords = ["proton", "drive", "fuse", "sync", "linux"]
 categories = ["filesystem", "command-line-utilities"]
 
 [workspace.dependencies]
-proton-sdk = { version = "0.10.0" }
-proton-drive-rs = { version = "0.10.0" }
+proton-sdk = { version = "0.10.1" }
+proton-drive-rs = { version = "0.10.1" }
 
 # Internal crates.
 pdfs-core = { path = "crates/pdfs-core" }

crates/pdfs-fuse/src/filesystem.rs

@@ -1434,6 +1434,14 @@ impl ProtonFs {
                     // (`ContentCache::clone_or_copy`).
                     if let Err(e) = ContentCache::clone_or_copy(&pending.path, &path) {
                         let _ = std::fs::remove_file(&path);
+                        // The drain lets go of the pending entry before it
+                        // discards the blob, so a blob that went away under the
+                        // copy is a revision that landed or was replaced: the
+                        // same race the re-check below catches, only earlier.
+                        if self.core.pending_blob(&uid).as_ref() != Some(&pending.path) {
+                            warn!(%uid, attempt, "queued revision went away while opening for write; retrying");
+                            continue;
+                        }
                         error!(%uid, source = %pending.path.display(), error = %e,
                             "copy queued revision into write scratch failed");
                         break 'install Err(Errno::EIO);

crates/pdfs-fuse/src/sim/daemon.rs

@@ -592,14 +592,22 @@ mod tests {
             daemon.pending(),
             drive.tree()
         );
-        // The bytes kept for the handle went once the trash landed with it closed.
-        let cached: Vec<_> = std::fs::read_dir(dir.join("cache"))
-            .unwrap()
-            .filter_map(|e| e.ok())
-            .filter(|e| e.file_type().is_ok_and(|t| t.is_file()))
-            .map(|e| e.file_name())
-            .collect();
-        assert!(cached.is_empty(), "{cached:?}");
+        // The bytes kept for the handle went once the trash landed with it
+        // closed. The queue empties as the trash's row goes, a moment before
+        // the eviction, so this waits too.
+        let cached = || -> Vec<_> {
+            std::fs::read_dir(dir.join("cache"))
+                .unwrap()
+                .filter_map(|e| e.ok())
+                .filter(|e| e.file_type().is_ok_and(|t| t.is_file()))
+                .map(|e| e.file_name())
+                .collect()
+        };
+        assert!(
+            wait_until(Duration::from_secs(30), || cached().is_empty()),
+            "{:?}",
+            cached()
+        );
 
         assert!(matches!(daemon.stop(), Ok(MountOutcome::Shutdown)));
         let _ = std::fs::remove_dir_all(&dir);
@@ -1998,6 +2006,21 @@ mod tests {
         daemon.write_new(&path, b"first").unwrap();
         assert!(wait_until(Duration::from_secs(30), || held.reached()));
         std::fs::write(&path, b"second").unwrap();
+        // The kernel sends the release after `close` returns. Attaching the
+        // bytes resets the create's attempts, so a release that came after the
+        // lost reply had it retried at once, and the retry landed before
+        // anything here saw it fail. The bytes it replaced go only once they
+        // are attached.
+        let staged = dir.join("cache").join("staging");
+        assert!(wait_until(Duration::from_secs(30), || {
+            let blobs: Vec<Vec<u8>> = std::fs::read_dir(&staged)
+                .into_iter()
+                .flatten()
+                .filter_map(|e| e.ok())
+                .filter_map(|e| std::fs::read(e.path()).ok())
+                .collect();
+            blobs.iter().any(|b| b == b"second") && !blobs.iter().any(|b| b == b"first")
+        }));
         drop(held);
         assert!(wait_until(Duration::from_secs(30), || {
             daemon

docs/BUGS.md

@@ -68,10 +68,10 @@ original wording.
 
 | Entry | Title | What is left |
 |---|---|---|
+| [B275](#b275--on-a-slow-link-a-files-last-write-can-land-only-in-a-conflict-copy) | On a slow link, a file's last write can land only in a conflict copy | Cause not found; seen once in CI |
 | [B273](#b273--a-restore-proposes-another-computers-paths-when-the-two-share-a-hostname) | A restore proposes another computer's paths when the two share a hostname | Not fixed |
 | [B270](#b270--one-file-left-in-a-mirror-folder-clears-the-wipe-guard-for-all-the-others) | One file left in a mirror folder clears the wipe guard for all the others | Not fixed |
 | [B267](#b267--a-key-failure-at-a-start-from-a-warm-cache-shows-as-offline-for-good) | A key failure at a start from a warm cache shows as offline for good | Not fixed |
-| [B265](#b265--address-keys-whose-token-is-for-an-inactive-account-key-never-open) | Address keys whose token is for an inactive account key never open | Needs an SDK change; not reproduced |
 | [B134](#b134--a-file-deleted-online-while-the-drain-checks-its-write-is-kept-as-a-conflict-copy) | A file deleted online while the drain checks its write is kept as a conflict copy | Only with `local_first` off |
 | [B91](#b91--the-daemon-hung-overnight-and-nothing-could-say-what-it-was-waiting-on) | The daemon hung overnight, and nothing could say what it was waiting on | The cause is not diagnosed; the daemon now reports where it is stuck and is restarted |
 | [B67](#b67--fresh-state-restore-can-silently-select-a-new-device-high-22) | Fresh-state restore can silently select a new device (HIGH-22) | An ambiguous device after a loss of local state is not asked about |
@@ -109,6 +109,8 @@ a live account since.
 
 | Entry | Title |
 |---|---|
+| [B265](#b265--address-keys-whose-token-is-for-an-inactive-account-key-never-open) | Address keys whose token is for an inactive account key never open |
+| [B274](#b274--opening-a-file-for-writing-fails-with-eio-while-its-queued-revision-lands) | Opening a file for writing fails with EIO while its queued revision lands |
 | [B271](#b271--one-failed-listing-makes-a-large-first-sync-start-over-from-nothing) | One failed listing makes a large first sync start over from nothing |
 | [B266](#b266--pin-unpin-search-and-node-loads-report-a-database-error-as-offline) | Pin, unpin, search and node loads report a database error as "offline" |
 | [B264](#b264--a-mount-that-fails-after-starting-its-workers-leaves-them-running) | A mount that fails after starting its workers leaves them running |
@@ -375,6 +377,56 @@ a live account since.
 
 Newest first.
 
+### B275 — On a slow link, a file's last write can land only in a conflict copy
+
+**Status:** Open
+**Found:** 2026-10-31, CI on #60: `one_client_on_a_slow_link`, profile `one-client-wifi` seed 2,
+failed with `Drive does not hold what the clients wrote, 120s after the links came up`. The
+same seed passed 4 times locally, and #60 changes no daemon logic, so the failure depends on
+timing.
+**Where:** not located; the drain's conflict check in `crates/pdfs-fuse/src/drain.rs` is the
+first suspect
+
+**Repro:** Only the CI run above. The last steps created `x/z/y/a.txt` (37 bytes), wrote 42
+bytes into it at offset 20, and brought every link up. Drive then held the file at its first
+37 bytes, and the full content was in `x/z/y/a (sync-conflict …).txt`. No other client wrote
+to it.
+
+**Cause.** Not known. Nothing was lost, since the conflict copy holds the bytes, but the file
+conflicted with its own create. B183 and B184 found the same symptom after a lost create
+reply.
+
+**Fix.** Not done. Needs a reproduction: run the seed in a loop under load and keep the state
+directory of a failing run.
+
+---
+
+### B274 — Opening a file for writing fails with EIO while its queued revision lands
+
+**Status:** Fixed, not verified live
+**Found:** 2026-10-31, CI: the simulation
+`a_file_whose_upload_was_not_read_back_takes_two_writes_at_once` failed once with
+`Input/output error` on a write, after `copy queued revision into write scratch failed … No
+such file or directory`. It did not fail in 54 local runs.
+**Where:** the write open in `crates/pdfs-fuse/src/filesystem.rs`
+
+**Repro:** Write a file, and open it for writing again at the moment its queued revision
+finishes uploading. The open fails with EIO, and the program sees its save fail.
+
+**Cause.** The open samples the queued revision, then copies its blob into the new handle's
+scratch file. The drain lets go of the pending entry and then deletes the blob when the upload
+lands. A blob deleted between the sample and the copy failed the open, although the re-check
+just after the copy already retries a revision that changed during it.
+
+**Fix.** A failed copy checks whether the pending entry still names that blob. When it does
+not, the revision landed or was replaced, and the open goes around again with the base the
+file has now, as the re-check does. Since the drain drops the entry before the blob, a blob
+that is gone and still named is a real failure and still returns EIO.
+
+**Test:** none deterministic; the simulation above covers the window.
+
+---
+
 ### B273 — A restore proposes another computer's paths when the two share a hostname
 
 **Status:** Open
@@ -566,7 +618,7 @@ error, reads as offline. That covers every caller at once, the database ones inc
 
 ### B265 — Address keys whose token is for an inactive account key never open
 
-**Status:** Open
+**Status:** Fixed, not verified live
 **Found:** 2026-10-31, code audit after issue #40: a Duo account whose mount failed on every
 start with `fetch My Files root: cryptography error: failed to decrypt: missing key`, after three
 `failed to decrypt address key token` warnings. The web app worked. B210's fix does not apply,
@@ -583,8 +635,17 @@ already reports "token is for inactive account key", so the case is known. Separ
 `addresses()` fails as a whole through `?` when one address has no usable primary key, which
 takes the working addresses down with it.
 
-**Fix.** Not done, and it belongs in the SDK. Use every account key that opens for decrypting
-tokens, and only the active ones for signing. Skip and log an address that cannot be opened.
+**Fix.** In `proton-sdk` 0.10.1 (narrrl/proton-sdk-rs#3). Tokens are tried with every account
+key that opens, active ones first, and signing uses active keys only. An address that cannot be
+opened is left out with a warning, and the others work; `addresses()` fails only when none
+opens. When the left-out address is the default, `default_address` fails with the reason
+rather than signing with another address. `key_report` walks the same chain, so `pdfs diagnose`
+reports such a key as opening. Not verified against an account with an inactive key, since
+none is at hand; issue #40's reporter can confirm.
+
+**Test:** `an_address_key_whose_token_is_for_an_inactive_account_key_opens`,
+`an_address_whose_keys_do_not_open_leaves_the_others_working` and
+`an_account_whose_addresses_all_fail_to_open_is_an_error` in `proton-sdk`.
 
 ---