Motivation:
We didn't have a test that tested that streaming really streams, ie.
that request body part N arrives before N+1 is sent.
Modification:
Test it.
Result:
Better test coverage.
Motivation:
AHC's tests were split in HTTPClientTests (which only use the public
API) and HTTPClientInternalTests (which can use `internal` API by using
`@testable`). At some point, the public API tests had `@testable` added
to the AHC import which breaks this idea.
Modification:
Restore the intent by removing `@testable` and moving the 2 tests that
needed it over.
Result:
Cleaner test suite.
This is the continuation of the good work of @Yasumoto and @weissi in #135
The following code adds support for NIO Transport services. When the ConnectionPool asks for a connection bootstrap it is returned a NIOClientTCPBootstrap which wraps either a NIOTSConnectionBootstrap or a ClientBootstrap depending on whether the EventLoop we are running on is NIOTSEventLoop.
If you initialize an HTTPClient with eventLoopGroupProvider set to .createNew then if you are running on iOS, macOS 10.14 or later it will provide a NIOTSEventLoopGroup instead of a EventLoopGroup.
Currently a number of tests are failing. 4 of these are related to the NIOSSLUncleanShutdown error the others all seem related to various race conditions which are being dealt with on other PRs. I have tested this code with aws-sdk-swift and it is working on both macOS and iOS.
Things look into:
The aws-sdk-swift NIOTS HTTP client had issues with on Mojave. We should check if this is the case for async-http-client as well.
Co-authored-by: Joe Smith <yasumoto7@gmail.com>
Co-authored-by: Johannes Weiss <johannesweiss@apple.com>
* Enable clients to call URLs that include %2F as an escaped backslash
Previously `percentEncodedPath` was using `path.addingPercentEncoding(withAllowedCharacters: .urlPathAllowed)`
which converts %2F to a literal '/'. This prevented users fetching URLs like https://api.travis-ci.org/repo/github/rails%2Frails
which use %2F as part of a path segment.
Migrating to `URLComponents(url: self, resolvingAgainstBaseURL: false)?.percentEncodedPath` has the desired behaviour
for the couple of test cases that exist.
Updated the test server to switch on the `percentEncodedPath` so it's easier
to understand the desired behaviour.
motivation:
TLS handshake promise was leaked in some cases of failure (see #179)
changes:
- Avoid leaking promise
- Clearer completion flow for related futures
- Add testAvoidLeakingTLSHandshakeCompletionPromise test
* Close idle pool connections
Motivation: Pooled connections should close at some point (see #168)
Changes:
- Add new poolingTimeout property to HTTPClient.Configuration, it's
default value is .seconds(60), it can be set to nil if one wishes to
disable this timeout.
- Add relevant unit test
Co-authored-by: Johannes Weiss <johannesweiss@apple.com>
Motivation:
I wasn't sure if we tested that we would successfully recover from
a server that sometimes closes the connection on us.
Modification:
Added a test case where we do three requests: First one succeeds,
second one gets a close from the server, third one succeeds again.
So we're testing that although AsyncHTTPClient doesn't auto-retry, the
user can just retry again.
Result:
More test coverage.
motivation: Better performance thanks to connection reuse
changes:
- Added a connection pool for HTTP/1.1
- All requests automatically use the connection pool
- Up to 8 parallel connections per (scheme, host, port)
- Multiple additional unit tests
Previously, UNIX Domain Sockets would only work if the URL also had a
"base URL". If it didn't have a base URL, it would try to connect to the
empty string which would fail :).
Now, we support both cases:
- URLs with a baseURL (the path to the UDS) and a path (actual path)
- URLs that just have an actual path (path to the UDS) where we'll just
use "/" as the URL's path
* Bugfix HTTPS SNI and IP Address
Motivation:
Solving the SNI Bug
Modifications:
Added an internal extension on String for checking if the hostname is an IP Address -- see the private extension on SNI. Additionally using the IPv4Address and IPv6Address Function from Network above 10.14 as protecting with #availabe.
Adding the test for HTTPS and IP in as hostname
Result:
We get results with an IP as Hostname
Motivation:
We need a test that we can handle servers that repeatedly send
connection: closes.
Modification:
Add a unit test that always makes the server send connection: close.
Result:
More test coverage.
Motivation:
It's important to also have a test case where async-http-client is
actually used to do multiple parallel requests on multiple workers.
Modification:
Add a test where 5 workers are doing 100 requests each.
Result:
Better test coverage.
Motivation:
More unit tests are good and now that SwiftNIO shipd
`NIOHTTP1TestServer`, writing integration tests for async-http-client is also
more straightforward.
Modification:
Demonstrate some tests using NIOHTTP1TestServer.
Result:
More tests.
motivation: the Swift Server Workgroup is not a legal entity and cannot hold copyrights. with this change, code authors continue and retain their copyrights under the apache license and previous copyrights note, but Apple steps up instead of the workgroup which has no legal status
changes:
* update header files to say "Apple Inc. and the AsyncHTTPClient project authors" instead of "Swift Server Workgroup and the AsyncHTTPClient project authors"
* update validation scripts to check for the correct header
* add CONTRIBUTING.md file to explain how to make contributions and include a legal notice about licensing the contribution to Apple and the project
* regenerate CONTRIBUTORS.md to reflect most recent contributions
This commit fixes#95 by always hopping event loop futures received from
the delegate to the right event loop. This could be a source of bugs if
the library users forgot to hop(to:) futures from their delegates
implementations.
Motivation:
Right now, HTTPClient only asserts that it's shut down if it was started with its own EventLoopGroup.
That however is weird because it's lifecycle model depends on the parameters you pass to `init`.
Modifications:
Always validate the lifecycle (in debug mode).
Result:
API makes more sense.
* add NIO event loop as an argument for execute
* review fix: add to np-delegate method as well
* Resolve confict
* add missing linux test
* fix formatting
* missing self
* review fix: add event loop preference argument instead of eventloop
* formatting
* review fix: spelling
* fix compilation error
* review fixes: make preference argument not explicit and add precondition that EL must be part of ELG
* Ignore uncleanShutdown error when state is head or body
* Add ignoreNIOSSLUncleanShutdownError to Configuration
* Revert old HTTPClient.init founctions
* Run generate_linux_tests.rb
* Rename ignoreNIOSSLUncleanShutdownError to ignoreUncleanSSLShutdown
* Make tests compatible with swift 5.0
motivation: safer handling of request validation and mutation
changes:
* drop request version
* made request method and url immutable
* made request scheme and host internal
* fix scheme logic to be non-case sensitive
* adjusted redirect handler implementation to stricter request immutabllity
* adjust and add tests
motivaiton: unified format
changes:
* fix outstanding formatting issues
* add a call to swiftformat as part of sanity script
* fix sanity script language check debugging statements
Johannes WeissandGitHub
Artem RedkinandGitHub
Iain SmithandGitHub
Fabian FettandGitHub
TrevörandGitHub
Marcin KrzyzanowskiandJohannes Weiss
Andreas KostuchandJohannes Weiss
tomer doronandGitHub
vkillandJohannes Weiss
Ludovic DewaillyandArtem Redkin
TobiasandJohannes Weiss