This commit reverts most of dea940c7 and d89c7024. The "fix" is to run
attach fairings on a new thread. If a runtime is already running, it is
used. Otherwise, the future is executed in a single-threaded executor.
This commit completely overhauls Rocket's configuration systems, basing
it on the new Figment library. It includes many breaking changes
pertaining to configuration. They are:
* "Environments" are replaced by "profiles".
* 'ROCKET_PROFILE' takes the place of 'ROCKET_ENV'.
* Profile names are now arbitrary, but 'debug' and 'release' are given
special treatment as default profiles for the debug and release
compilation profiles.
* A 'default' profile now sits along-side the meta 'global' profile.
* The concept of "extras" is no longer present; users can extract any
values they want from the configured 'Figment'.
* The 'Poolable' trait takes an '&Config'.
* The 'secrets' feature is disabled by default.
* It is a hard error if 'secrets' is enabled under the 'release'
profile and no 'secret_key' is configured.
* 'ConfigBuilder' no longer exists: all fields of 'Config' are public
with public constructors for each type.
* 'keep_alive' is disabled with '0', not 'false' or 'off'.
* Inlined error variants into the 'Error' structure.
* 'LoggingLevel' is now 'LogLevel'.
* Limits can now be specified in SI units: "1 MiB".
The summary of other changes are:
* The default config file can be configured with 'ROCKET_CONFIG'.
* HTTP/1 and HTTP/2 keep-alive configuration is restored.
* 'ctrlc' is now a recognized config option.
* 'serde' is now a core dependency.
* TLS misconfiguration errors are improved.
* Several example use '_' as the return type of '#[launch]' fns.
* 'AdHoc::config()' was added for simple config extraction.
* Added more documentation for using 'Limits'.
* Launch information is no longer treated specially.
* The configuration guide was rewritten.
Resolves#852.
Resolves#209.
Closes#1404.
Closes#652.
In brief, this commit:
* Updates to the latest upstream 'cookie', fixing a memory leak.
* Make changes to 'CookieJar' observable only through 'pending()'.
* Deprecates 'Client::new()' in favor of 'Client::tracked()'.
* Makes 'dispatch()' on tracked 'Client's synchronize on cookies.
* Makes 'Client::untracked()' actually untracked.
This commit updates to the latest 'cookie' which removes support for
'Sync' cookie jars. Instead of relying on 'cookie', this commit
implements an op-log based 'CookieJar' which internally keeps track of
changes. The API is such that changes are only observable through
specialized '_pending()' methods.
Prior to this commit, it was impossible to 'use' a route from a separate
namespace for use in a 'routes!' macro. Naturally, this was a common
source of confusion amongst users. This commit obviates this deficiency
by generating a "proxy" structure that can be imported and converted
into a 'Route'/'Catcher' or their static variants.
This change is largely backwards compatible but can break existing code
when routes are named identically to other types in the namespace.
The user-facing changes effected by this commit are:
* The 'http::Cookies<'_>' guard is now '&http::CookieJar<'_>'.
* The "one-at-a-time" jar restriction is no longer imposed.
* 'CookieJar' retrieval methods return 'http::CookieCrumb'.
* The 'private-cookies' feature is now called 'secrets'.
* Docs flag private cookie methods with feature cfg.
* Local, async request dispatching is never serialized.
* 'Client::cookies()' returns the tracked 'CookieJar'.
* 'LocalResponse::cookies()' returns a 'CookieJar'.
* 'Response::cookies()' returns an 'impl Iterator'.
* A path of '/' is set by default on all cookies.
* 'SameSite=strict' is set by default on all cookies.
* 'LocalRequest::cookies()' accepts any 'Cookie' iterator.
* The 'Debug' impl for 'Request' prints the cookie jar.
Resolves#1332.
This commit aims to make it impossible to modify a 'Route' structure in
a way that violates expectations of a code-generated 'Route'. It removes
'Route::set_uri()' in favor of 'Route::map_base()', which allows for
safe modifications of the route's base.
In a similar vain, this commit also includes the following changes:
* 'Route::path()' was added to safely retrieve the route's 'path'.
* The base of a 'Route' is underlined during launch printing.
* 'Origin::into_normalized()' replaces 'Origin::to_normalized()'.
Fixes#1262.
Prior to this commit, codegen used 'Span' information to generate a
unique id for a given route. This commit changes the id generation to
instead use 1) the route's name and path, 2) a an per-generation
atomically increasing ID, and 3) the ids of the process/thread the
proc-macro is running in. Together, these values should provide a unique
id for a given route, even in the face of the reused processes and
threads, while also removing the dependence on unstable Span features.
Fixes#1373.
This commit adds the 'local::blocking' module and moves the existing
asynchronous testing to 'local::asynchronous'. It also includes several
changes to improve the local API, bringing it to parity (and beyond)
with master. These changes are:
* 'LocalRequest' implements 'Clone'.
* 'LocalResponse' doesn't implement 'DerefMut<Target=Response>'.
Instead, direct methods on the type, such as 'into_string()', can
be used to read the 'Response'.
* 'Response::body()' returns an '&ResponseBody' as opposed to '&mut
ResponseBody', which is returned by a new 'Response::body_mut()'.
* '&ResponseBody' implements 'known_size()` to retrieve a body's size,
if it is known.
Co-authored-by: Jeb Rosen <jeb@jebrosen.com>
This is largely an internal change. Prior to this commit, the 'Manifest'
type, now replaced with the 'Cargo' type, robbed responsibility from the
core 'Rocket' type. This new construction restores the previous
responsibility and makes it clear that 'Cargo' is _only_ for freezing,
and representing the stability of, Rocket's internal state.
In summary, this commit modifies 'Responder' so that:
* ..it is no longer 'async'. To accommodate, the 'sized_body' methods
in 'Response' and 'ResponseBuilder' are no longer 'async' and accept
an optional size directly. If none is supplied, Rocket will attempt
to compute the size, by seeking, before writing out the response.
The 'Body' type was also changed to differentiate between its sized
'Seek' and chunked body variants.
* ..'&Request' gains a lifetime: 'r, and the returned 'Response' is
parameterized by a new 'o: 'r. This allows responders to return
references from the request or those that live longer.
Also:
* Remove 'response::ResultFuture'.
* Re-export 'tokio' and 'futures' from the crate root.
* Make 'ResponseBuilder::sized_body()' and 'async fn'.
* Remove the 'Future' implementation for 'ResponseBuilder'.
* Add 'ResponseBuilder::finalize()' for finalizing the builder.
In order to avoid making 'ResponseBuilder::sized_body' an asynchronous
function, the seeking is deferred until finalization. 'finalize()' is
replaced with '.await', and 'ResponseBuilder::ok()' is an 'async fn'.
* Update 'tokio', 'tokio-rustls', and 'hyper'.
* Remove unused dependencies on some `futures-*` crates.
* Rework 'spawn_on', which is now 'serve'.
* Simplify Ctrl-C handling.
Use I/O traits and types from 'tokio-io' as much as possible.
A few adapters only exist in futures-io-preview and use
futures-tokio-compat as a bridge for now.
* body_string_wait and body_bytes_wait are removed; use `.await` instead
* `dispatch()` is now an async fn and must be .await-ed
* Add `#[rocket::async_test]` macro, similar in purpose to `tokio::test`
* Tests now use either `rocket::async_test(async { })` or
`#[rocket::async_test]` in order to `.await` the futures returned
from `dispatch()` and `body_{string,bytes}()`
* Update 'test.sh' to reflect the tests that should be passing.
Broken:
* Cloned dispatch and mut_dispatch() with a live previous response now both fail, due to a (partial) check for mutable aliasing in LocalRequest.
* Some tests are still failing and need example-specific changes.
This is required to be able to do anything useful with the body in the
outgoing response. Request fairings do not appear to need to be async
as everything on Data that returns a future moves self and on_request only
gets &Data, but the same change in this commit should work for on_request
if desired.
The existing implementation of 'LocalRequest::clone()' mistakenly copied
the internal 'Request' pointer from the existing 'LocalRequest' to the
cloned 'LocalRequest'. This resulted in an aliased '*mut Request'
pointer, a clear soundness issue. The fix in this commit is to clone the
internal 'Request', replacing the internal pointer with the newly cloned
'Request' when producing the cloned 'LocalRequest'. A fix that removes
all 'unsafe' code should be explored.
Fixes#1312.
Every code example is now fully runnable and testable. As a result, all
examples are now tested and include imports. Relevant imports are shown
by default. Code examples can be expanded to show all imports.
Fixes#432.
Sergio Benitez
lewis
Jeb Rosen
Michael Howell
Paolo Barbolini
Jacob Pratt
lu4nm3
Linus Unnebäck