This introduces the `denort` binary - a slim version of deno without
tooling. The binary is used as the default for `deno compile`.
Improves `deno compile` final size by ~2.5x (141 MB -> 61 MB) on Linux
x86_64.
This implementation heavily depends on there being a lockfile, meaning
JSR specifiers will always diagnose as uncached unless it's there. In
practice this affects cases where a `deno.json` isn't being used. Our
NPM specifier support isn't subject to this.
The reason for this is that the version constraint solving code is
currently buried in `deno_graph` and not usable from the LSP, so the
only way to reuse that logic is the solved-version map in the lockfile's
`packages.specifiers`.
This looks like a massive PR, but it's only a move from cli/tests ->
tests, and updates of relative paths for files.
This is the first step towards aggregate all of the integration test
files under tests/, which will lead to a set of integration tests that
can run without the CLI binary being built.
While we could leave these tests under `cli`, it would require us to
keep a more complex directory structure for the various test runners. In
addition, we have a lot of complexity to ignore various test files in
the `cli` project itself (cargo publish exclusion rules, autotests =
false, etc).
And finally, the `tests/` folder will eventually house the `test_ffi`,
`test_napi` and other testing code, reducing the size of the root repo
directory.
For easier review, the extremely large and noisy "move" is in the first
commit (with no changes -- just a move), while the remainder of the
changes to actual files is in the second commit.
This PR separates integration tests from CLI tests into a new project
named `cli_tests`. This is a prerequisite for an integration test runner
that can work with either the CLI binary in the current project, or one
that is built ahead of time.
## Background
Rust does not have the concept of artifact dependencies yet
(https://github.com/rust-lang/cargo/issues/9096). Because of this, the
only way we can ensure a binary is built before running associated tests
is by hanging tests off the crate with the binary itself.
Unfortunately this means that to run those tests, you _must_ build the
binary and in the case of the deno executable that might be a 10 minute
wait in release mode.
## Implementation
To allow for tests to run with and without the requirement that the
binary is up-to-date, we split the integration tests into a project of
their own. As these tests would not require the binary to build itself
before being run as-is, we add a stub integration `[[test]]` target in
the `cli` project that invokes these tests using `cargo test`.
The stub test runner we add has `harness = false` so that we can get
access to a `main` function. This `main` function's sole job is to
`execvp` the command `cargo test -p deno_cli`, effectively "calling"
another cargo target.
This ensures that the deno executable is always correctly rebuilt before
running the stub test runner from `cli`, and gets us closer to be able
to run the entire integration test suite on arbitrary deno executables
(and therefore split the build into multiple phases).
The new `cli_tests` project lives within `cli` to avoid a large PR. In
later PRs, the test data will be split from the `cli` project. As there
are a few thousand files, it'll be better to do this as a completely
separate PR to avoid noise.
Fixes https://github.com/denoland/deno/issues/21840
The problem was hard to reproduce as its a race condition. I've added a
test that reproduces the problem 1/10 tries. We should move the
idleTimeout handling to Rust (maybe even built into fastwebsocket).
Main change is that:
- "hyper" has been renamed to "hyper_v014" to signal that it's legacy
- "hyper1" has been renamed to "hyper" and should be the default
This commit adds support for a new `kv.watch()` method that allows
watching for changes to a key-value pair. This is useful for cases
where you want to be notified when a key-value pair changes, but
don't want to have to poll for changes.
---------
Co-authored-by: losfair <zhy20000919@hotmail.com>
Issue was main does canary builds, which broke this test because it
didn't handle searching for a canary release. Tested by building as
canary locally.
This is an attempt to fix this flakiness:
```
---- integration::repl::assign_underscore stdout ----
deno_exe path /home/runner/work/deno/deno/target/release/deno
command /home/runner/work/deno/deno/target/release/deno repl
command cwd /tmp/deno-cli-testK3YASC
------ Start Full Text ------
""
------- End Full Text -------
Next text: ""
thread 'integration::repl::assign_underscore' panicked at test_util/src/pty.rs:41:11:
Timed out.
stack backtrace:
```
We only want one zlib dependency.
Zlib dependencies are reorganized so they use a hidden
`__vendored_zlib_ng` flag in cli that enables zlib-ng for both libz-sys
(used by ext/node) and flate2 (used by deno_web).
Remove tokio-rustls as a direct dependency of Deno and refactor
test_server to reduce code duplication.
All tcp and tls listener paths go through the same streams now, with the
exception of the simpler Hyper http-only handlers (those can be done in
a later follow-up).
Minor bugs fixed:
- gRPC server should only serve h2
- WebSocket over http/2 had a port overlap
- Restored missing eye-catchers for some servers (still missing on Hyper
ones)
Implements `WebSocket` over http/2. This requires a conformant http/2
server supporting the extended connect protocol.
Passes approximately 100 new WPT tests (mostly `?wpt_flags=h2` versions
of existing websockets APIs).
This is implemented as a fallback when http/1.1 fails, so a server that
supports both h1 and h2 WebSockets will still end up on the http/1.1
upgrade path.
The patch also cleas up the websockets handshake to split it up into
http, https+http1 and https+http2, making it a little less intertwined.
This uncovered a likely bug in the WPT test server:
https://github.com/web-platform-tests/wpt/issues/42896
This commit updates the ext/kv module to use the denokv_* crates for
the protocol and the sqlite backend. This also fixes a couple of bugs in
the sqlite backend, and updates versionstamps to be updated less
linearly.
