| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Currently prefetch will send all requests for all chunks for the file in
one shot. With the default chunk size of 32k, this can result in many
thousand outstanding requests for large files. Some servers like Serv-U
15.2.3.742 seem to be dropping requests after a certain number, which
results in the file download hanging indefinitely (or until the server
closes the connection).
Fix this issue by letting the user specify a limit for the number of
concurrent requests. This is similar to openssh's sftp, which limits
the number of concurrent requests to 64 by default.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
- New subclass(es) for opt-in use. Most below messages refer to them,
not parent classes.
- In parent classes, make handler tables instance attributes for easier
subclass twiddling.
- Refactor Transport-level session check
- Refactor Transport-level auth handler instantiation (but keep behavior
the same, for now)
- Add service-request handler to Transport subclass, and remove from
AuthHandler subclass
- Remove manual event injection from the handful of Transport auth
methods which supported it. Suspect unused, don't need the extra
complexity, and wasn't consistent anyways - can add back smarter later
if anyone needs it.
- Not bothering with gssapi at all for now as I cannot easily test it
- Primarily tested against the new AuthStrategy architecture
|
|
- Comment was being read-but-not-stored from the agent reply. wat?
- Use newly added PKey constructor to instantiate a key subclass for the
'inner'/proxied key, this way client code can obtain stuff like bit
size, fingerprint, etc.
- Proxy to inner_key with __getattr__ so clients don't have to know
whether they're dealing with an AgentKey or a regular one
- Add `__repr__` to PKey instead of doing it in AgentKey. (wow, how did
we not have this ever?)
|
|
- add .name to eventually replace .get_name
- use that in a bunch of spots to avoid some duplication
- add .identifiers classmethod to extend existing idea from ECDSAKey
- add from_type_string alt constructor which uses .identifiers
- use that in HostKeys (includes hopefully-minor refactoring)
- no longer giving outdated init kwarg to ECDSA host key loading
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
=== Do not change lines below ===
{
"chain": [],
"cmd": "codespell -w",
"exit": 0,
"extra_inputs": [],
"inputs": [],
"outputs": [],
"pwd": "."
}
^^^ Do not change lines above ^^^
|
|
|
|
|
|
|
|
Also make 'em f-strings, which AFAIK should be comparable
Closes #2110
|
|
|
|
Additionally, other bits of code that retry EINTR can similarly just go
away.
|
|
Also apparently the old, old test for this had the wrong issue number in
it :(
|
|
|
|
|
|
Plus related updates to tests, docstrings, changelog
|
|
I feel like we should be able to just nuke byte_chr and friends at this
point, but it's not entirely obvious, so let's rock that boat later.
|
|
and bump top end test matrix cell to 40ish
|
|
- Realized at this date that _moving_ Ed25519 deps behind an entrypoint
is going the wrong direction, that key type should be used _more_ now,
not _less_. I no longer even remember why folks were pissy about
pynacl/bcrypt exactly.
- We can in fact just use ed25519 from cryptography itself now, so no
more pynacl anyways (but still bcrypt)
- Twiddle install language a bit re: invoke used in more places, and
again removing the notes about ed25519 being special
|
|
NOTE: imports not fixed yet so this will break CI!
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
