Skip to content

chore(deps): bump dartssh2 from 2.22.5 to 4.1.0 - #24

Open
dependabot[bot] wants to merge 1 commit into
mainfrom
dependabot/pub/dartssh2-4.1.0
Open

chore(deps): bump dartssh2 from 2.22.5 to 4.1.0#24
dependabot[bot] wants to merge 1 commit into
mainfrom
dependabot/pub/dartssh2-4.1.0

Conversation

@dependabot

@dependabot dependabot Bot commented on behalf of github Sep 7, 2026 •
edited
Loading

Copy link
Copy Markdown

Bumps dartssh2 from 2.22.5 to 4.1.0.

Release notes

Sourced from dartssh2's releases.

4.1.0

What's Changed

Full Changelog: vicajilau/dartssh2@v4.0.1...v4.1.0

4.0.1

What's Changed

New Contributors

Full Changelog: vicajilau/dartssh2@v4.0.0...v4.0.1

4.0.0

What's Changed

New Contributors

Full Changelog: vicajilau/dartssh2@v3.3.1...v4.0.0

3.3.1

What's Changed

Full Changelog: vicajilau/dartssh2@v3.3.0...v3.3.1

3.3.0

What's Changed

... (truncated)

Changelog

Sourced from dartssh2's changelog.

[4.1.0] - 2026-09-04

  • Added SSHClient.pipelineChannelRequests, off by default, which sends all of a session's channel requests before reading any reply instead of waiting for each one in turn. execute and shell send env, agent forwarding, pty-req and x11-req ahead of exec or shell, and each of them cost a round trip: against a server 40 ms away, execute with a pty measured 246 ms before and 206 ms after. RFC 4254 §5.4 permits sending further messages without waiting and §4 requires the peer to answer a channel's requests in the order it received them, which is what ssh(1) relies on when it does the same thing. Setting it also adopts OpenSSH's reporting, because it has to: the command is on the wire before a refusal can come back, so a refused pty-req, env, agent forwarding or x11-req is reported through printDebug and the command runs, the way ssh(1) prints PTY allocation request failed on channel 0 and carries on, rather than throwing an error that means the command has already run. Only a refused exec or shell still throws SSHChannelRequestError, because nothing has run when that one fails. Leaving it unset changes nothing, down to the order the requests go out and the message of every error #243.
  • Added a dartssh2-nopty account to the interop server, which PermitTTY no applies to, so both sides of a refused pty-req are exercised against a real OpenSSH: by default the command does not run, and with pipelining it does #243.
  • Fixed the package reading as web-incompatible on pub.dev. SftpFile.downloadToRandomAccess takes a dart:io RandomAccessFile, and naming that type from the SFTP library was enough for pub.dev to drop platform:web from the whole package, which also keeps it out of any search filtered to web. Everything else already compiled and ran there, and 4.0.0 made the ciphers and SFTP work. The method moves to an SftpFileDownload extension in its own library, exported conditionally the way SSHSocket and dynamic forwarding already are, so nothing changes for a caller on the VM: same import, same call. On the web the extension is simply absent, as is the RandomAccessFile it would need. Confirmed with pana, the tool pub.dev scores with: platform:web is present after and absent before #248.
  • Added browser interop tests, so "web is supported" is something CI checks rather than something the pieces individually suggest. Everything that ran under -p chrome until now proved the AEAD arithmetic, the SFTP encoding and the HTTP parsing compile and behave, and none of it put a packet on a wire, which is the gap that let every AEAD cipher and the whole of SFTP sit broken on the web until 4.0.0. The new tests run the browser against the same OpenSSH server the VM interop tests use, through tool/ws_bridge.dart, and cover a handshake, a command, an aes256-gcm session and an SFTP round trip. The SSHSocket they connect through is the WebSocket one the README tells web users to write, so it doubles as a worked example #248.

[4.0.1] - 2026-09-03

  • Fixed a channel stalling permanently when data arrived before the application subscribed to it. StreamController.isPaused is true until something listens, which suppressed every SSH_MSG_CHANNEL_WINDOW_ADJUST, and Dart delivers the first subscription as onListen rather than onResume — the only hook wired — so a peer that legally filled the advertised window in that gap was left at zero credit with no further data able to arrive and trigger a grant. Reproduced at the client's own sizes with 64 packets of 32 KiB into a 2 MiB window, before and after the first listener and through .map(), in every case zero adjustments. The remote forwarding example in the README reaches it: it awaits Socket.connect() for each connection before subscribing, and the forwarded channel is created with no listener attached #244.
  • Changed the receive window to be granted back once half of it has been consumed rather than after every data packet. Any consumed byte made the outstanding grant positive, so each SSH_MSG_CHANNEL_DATA was answered with a window adjust of its own, and the smaller the packets the closer the uplink packet count got to the downlink one: counted on SSHChannelController, 256 inbound packets of 32 KiB produced 256 adjustments, and 4096 of 64 B produced 4096. They now produce 8 and 0. This is one of the two rules OpenSSH applies in channels.c, which refills at half the window or once local_window_max - local_window > local_maxpacket * 3, whichever comes first. Only the first is implemented here, so at a 2 MiB window with 32 KiB packets this defers further than OpenSSH would, thirty-two packets against its three or four, which is the point of the change. The threshold never defers past one maximum packet of remaining credit, so a window smaller than twice the packet size cannot leave a conforming peer holding a chunk it may neither send nor is obliged to split. Pausing the stream still suppresses the grant, which is the documented backpressure mechanism #244.

[4.0.0] - 2026-08-31

  • Breaking change. Removed the legacy algorithms from the default proposals, leaving them implemented but off unless asked for: the SHA-1 key exchange methods diffie-hellman-group14-sha1 and diffie-hellman-group-exchange-sha1, the ssh-rsa host key signature, and the aes256-cbc and aes128-cbc ciphers. This matches what OpenSSH proposes in myproposal.h, which drops all of them and keeps only hmac-sha1 at the end of the MAC list, as this does. ssh-rsa signs host keys with SHA-1 and is open to chosen-prefix collisions, which is why OpenSSH disabled it by default in 8.8, and CBC in SSH is vulnerable to the plaintext recovery of CVE-2008-5161. A server that offers nothing but these will now fail to negotiate rather than connect weakly, which for older routers, NAS boxes and embedded servers is a real change: pass the algorithm through SSHAlgorithms to keep talking to it #236. Thanks [@​GT-610].
  • Changed SFTP uploads to pipeline a bounded number of outstanding write requests, 64 by default, matching OpenSSH's DEFAULT_NUM_REQUESTS. Writes were issued and awaited one at a time, so every chunk cost a full round trip and a high latency link spent most of its time idle. Acknowledgements are now accepted out of order without resubmitting an offset, and offsets are still assigned in stream order so concurrent writes cannot overlap. SftpFile.write and SftpFileWriter take chunkSize and maxPendingRequests, and reject a negative offset or a non-positive setting rather than misbehaving later #237. Thanks [@​GT-610].
  • Changed a failing SFTP upload to stop scheduling new writes, drain the ones already in flight and report the first error through the returned future, instead of surfacing whichever error happened to arrive last. This supersedes the narrower SftpFileWriter error handling added in #230, whose regression tests all still pass #237.
  • Deprecated chunkSize in favour of defaultChunkSize, and maxBytesOnTheWire, which nothing reads any more now that uploads are bounded by request count rather than by a byte window #237.
  • Changed ChunkBuffer to grow by reallocating with headroom instead of copying the whole buffer on every append, making add() amortised O(1) rather than O(n). Accumulating 64 MB in 8 KiB chunks without draining, which is what a burst of packets arriving faster than they are consumed looks like, drops from about 185 seconds to about 72 milliseconds. The steady drained case costs about 25% more, roughly 162 ms to 202 ms over 20k packets of 32 KiB, because consume() now returns a copy rather than an alias into a buffer a later add() may reallocate; both figures are around 3 GB/s, so neither is visible next to a network #231.
  • Removed the dead MinChunkSize stream transformer and two stale markers in ssh_mac_type.dart, neither of which anything referenced #231.
  • Removed the legacy analyzer plugin entry from analysis_options.yaml. Dart 3.13.2 warns on it and dart analyze exits non-zero on a warning, so every job on every branch started failing with nothing in the code having changed. It was dead configuration in any case: dart_code_metrics_presets ships preset YAML meant for include:, no analyzer plugin, so nothing was ever loaded through it. The now unused dev dependency went with it #232 #233.
  • Scoped the encrypted key tests to the VM. They use dart:io for their ssh-keygen interoperability checks, which the web job cannot load #239.
  • Breaking change. SftpFileAttrs now drops a uidgid or acmodtime pair when only one half of it is set, instead of writing the flag with a single value. The pair is two fields under one flag in the SFTP wire format, so a half-filled one produced a packet the server misparsed, applying the wrong ownership or timestamps. Anyone calling setStat with only modifyTime set will find the value is now ignored rather than sent alongside a garbage access time; set both to change either #230. Thanks [@​klc].
  • Breaking change. A host key that changes during a rekey now terminates the connection with SSHHostkeyError, as OpenSSH does. The signature was already re-checked on every exchange, but that only proved the key presented was self-consistent, not that it was the key onVerifyHostKey had already approved, so a server could hand out one key at connect time and a different one on the first rekey. A connection that used to survive this will now drop. onVerifyHostKey is consulted once per connection, not once per exchange #229. Thanks [@​klc].
  • Fixed the version exchange failing when the banner arrives split across TCP segments or WebSocket frames. It was treated as a framing error rather than a partial read, which made the WebSocket transport the README recommends for web unreliable by construction, and any slow or proxied connection intermittently so. Lines of text sent before the identification line are also skipped now, which RFC 4253 §4.2 says clients MUST be able to process; at most 1024 of them, matching OpenSSH, so a server streaming them forever cannot keep a client busy #229.
  • Fixed keyboard-interactive responses being written to the trace log in plaintext, which put the user's password in any log a caller collected with printTrace set #229.
  • Changed MAC comparison to constant time, and removed both the received and the expected MAC from the failure message #229.
  • Fixed three of the four packet send paths using a fixed padding pattern instead of random bytes, contrary to RFC 4253 §6 #229.
  • Added validation of the peer's public value in every key exchange. Finite field Diffie-Hellman now requires 1 < f < p - 1, the NIST curves reject points that fail to decode, lie off the curve or are the point at infinity, and X25519 rejects small-order points via the RFC 8731 §3 all-zero shared secret check and requires the key to be exactly 32 bytes. Without these a peer could force a shared secret it knew in advance #229.
  • Added the missing lower bound and AEAD path checks to packet length validation, and made a zero-length payload raise SSHPacketError instead of a RangeError that no SSHError handler would catch #229.
  • Changed the non-ETM receive path to verify the MAC before parsing the padding, so a forged packet is rejected before its length fields are trusted #229.
  • Fixed the non-ETM receive path accepting a packet whose encrypted length is not a multiple of the cipher block size. The decrypt loop pulled whole blocks until it had enough, so an unaligned length made it read past the ciphertext and on into the MAC, and the RangeError that followed was not an SSHError any handler would catch. The length is now rejected up front, as OpenSSH does in ssh_packet_read_poll2(), and the remaining ciphertext is decrypted in one pass instead of a block at a time #234. Thanks [@​GT-610].
  • Fixed SSH_MSG_KEX_ECDH_REPLY being encoded with its fields in the wrong order. RFC 5656 §4 specifies K_S, Q_S, signature, which is what this library's own decoder already expected, so only a peer decoding what dartssh2 sent as a server was affected #229.
  • Fixed writeMpint(BigInt.zero) emitting 00 00 00 01 00 where RFC 4251 §5 requires a zero-length string, and readNameList returning [''] for an empty name-list #229.
  • Documented that leaving onVerifyHostKey null accepts any host key, which makes the connection trivially interceptable. The parameter is optional and the behaviour was not stated anywhere #229.
  • Added encrypted OpenSSH private-key writing. OpenSSHKeyPair.toPem() takes an optional passphrase, and a non-empty one encrypts the private section with aes256-ctr keyed by bcrypt_pbkdf, following what sshkey_private_to_blob2() writes: a 16-byte salt, 24 rounds, key and IV derived together in one call, the check integer written twice and the block padded with 1, 2, 3 and so on. A null or empty passphrase still writes the unencrypted form. OpenSSHKeyPairs.encrypted builds the container directly for callers that need it. The SSHKeyPair interface is unchanged, so the new arguments are reachable only through OpenSSHKeyPair #235. Thanks [@​GT-610].
  • Fixed a failing bcrypt KDF being ignored when reading an encrypted OpenSSH key. bcrypt_pbkdf reports invalid parameters through its return value, which was discarded, so a key with a zero round count or an empty salt carried on with an underived key and failed later as a check-integer mismatch. It now raises SSHKeyDecryptError at the point the KDF fails #235. Thanks [@​GT-610].
  • Added SSHClient.rekey(), which starts a new key exchange on an established connection and returns a Future<void> completing once the new keys are in effect, so a caller rekeying between transfers can await it instead of watching done. If an exchange is already running, whether this side or the server started it, no second one is sent and the future tracks the one in flight; if the connection ends first the future carries the error that ended it. SSHTransport.rekey() was already public and returns the same future now, where it used to return void #229.
  • Fixed every AEAD cipher and the whole of SFTP being dead on the web. ByteData.getUint64 and setUint64 throw Unsupported operation: Uint64 accessor not supported by dart2js, and four call sites went through them: the message reader, the int helper, the AES-GCM nonce and the ChaCha20-Poly1305 nonce. Since [email protected] sits first in the default cipher list, a browser connection died at the first encrypted packet even with a correct custom SSHSocket. Each now reads and writes two 32-bit words instead, bit-for-bit identical on the VM; compiled to JS a value needing more than 53 bits raises UnsupportedError rather than silently rounding, since these back SFTP file sizes and offsets #230.
  • Added a test-web CI job running dart test -p chrome, with @TestOn markers on the suites that genuinely need the VM. Nothing ran against dart2js before, which is how the above went unnoticed while the README listed web as supported #230.
  • Fixed virtual files such as those under /proc and /dev being handed back as empty. They report a stat size of 0 while still returning data, and read, downloadTo and downloadToRandomAccess all trusted the reported size and returned before issuing a read. Reads are now EOF-driven when the size cannot be trusted #230.
  • Fixed SftpFileWriter hanging forever when the local stream raised. Errors had no handler and _handleLocalDone completed the done future unguarded, so a failing upload never returned and could also double-complete #230.
  • Fixed an HTTP response body being discarded when the server sends no Content-Length. The body was read only up to a length that stayed 0, so whether it survived depended on TCP segment boundaries #230.
  • Fixed HttpHeaders.host and HttpHeaders.port returning null for a perfectly valid Host header #230.
  • Fixed a response that carries no body being read until the peer closes the connection. A 1xx, 204 or 304 response, and any response to HEAD, has no body whatever its framing headers say, so a server that keeps the connection open after sending one left the read waiting for bytes that were never coming. This was the sharp edge of reading unframed bodies to end of stream, added above #230.
  • Added SSHHttpClient.idleTimeout, which bounds the wait between two pieces of a response. A body delimited by connection close still has to be read to end of stream, and nothing else in the client bounded that, so a peer that stopped sending without closing could hang a request indefinitely. It is an inactivity timeout rather than a deadline for the whole response, so a large body that keeps arriving is never cut short; leaving it null keeps the unbounded behaviour #230.
  • Added overflow protection to the SFTP request id counter #230.

[3.3.1] - 2026-08-19

  • Removed the background isolate offload from X25519 and NIST curve key exchange, which cost more than the work it was hiding. Generating an ephemeral key or computing the shared secret on these curves is one fixed-size scalar multiply, well under a millisecond, while Isolate.run takes several times that to spawn and tear down, and a client pays it twice per handshake. On a memory constrained Android device the spawn delay was long enough for the server to time out the key exchange and close the connection before SSH_MSG_NEWKEYS went out, surfacing as SSHAuthAbortError with a null reason #226. Thanks [@​cesarcamps].
  • Kept the offload for finite field Diffie-Hellman, which is the one exchange whose cost the peer controls: group exchange lets the server name a modulus of up to 8192 bits, and modular exponentiation grows steeply with it.
  • Added debug logging around host key signature verification and the onVerifyHostKey callback. Everything between receiving the key exchange reply and sending SSH_MSG_NEWKEYS used to run without a single printDebug call, so a slow user callback and a slow shared secret were indistinguishable in a trace, and both looked like a hung handshake #226.

... (truncated)

Commits
  • 20b39cc Merge pull request #249 from vicajilau/docs/web-support-readme
  • 3500b8b chore: release 4.1.0
  • fd0078a docs: lead the web section with what works
  • 982a793 Merge pull request #248 from vicajilau/feat/web-platform
  • bdcece8 feat(sftp): list the package as web-compatible, and prove it in CI
  • 55ef90f Merge pull request #246 from michnovka/perf/pipeline-channel-requests
  • 557de3c perf(client): add SSHClient.pipelineChannelRequests, default off
  • 0a5a115 Merge pull request #245 from vicajilau/chore/release-4.0.1
  • cc4dca7 chore: release 4.0.1
  • f34badf Merge pull request #244 from michnovka/perf/window-adjust-threshold
  • Additional commits viewable in compare view

@dependabot @github

dependabot Bot commented on behalf of github Sep 7, 2026

Copy link
Copy Markdown
Author

Labels

The following labels could not be found: dependencies, pub. Please create them before Dependabot can add them to a pull request.

Please fix the above issues or remove invalid values from dependabot.yml.

@dependabot dependabot Bot mentioned this pull request Sep 7, 2026
Closed
@changeset-bot

changeset-bot Bot commented Sep 7, 2026 •
edited
Loading

Copy link
Copy Markdown

⚠️ No Changeset found

Latest commit: bea6dbe

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

@dependabot
chore(deps): bump dartssh2 from 2.22.5 to 4.1.0
bea6dbe 
Bumps [dartssh2](https://github.com/vicajilau/dartssh2) from 2.22.5 to 4.1.0.
- [Release notes](https://github.com/vicajilau/dartssh2/releases)
- [Changelog](https://github.com/vicajilau/dartssh2/blob/main/CHANGELOG.md)
- [Commits](vicajilau/dartssh2@v2.22.5...v4.1.0)

---
updated-dependencies:
- dependency-name: dartssh2
  dependency-version: 4.1.0
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <[email protected]>
@dependabot dependabot Bot changed the title chore(deps): bump dartssh2 from 2.22.2 to 4.1.0 chore(deps): bump dartssh2 from 2.22.5 to 4.1.0 Sep 14, 2026
@dependabot
dependabot Bot force-pushed the dependabot/pub/dartssh2-4.1.0 branch from 060d443 to bea6dbe Compare September 14, 2026 13:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

0 participants