* Initial agent
* fix: Use buffered reader in peer to fix ShortBuffer
This prevents a io.ErrShortBuffer from occurring when the byte
slice being read is smaller than the chunks sent from the opposite
pipe.
This makes sense for unordered connections, where transmission is
not guarunteed, but does not make sense for TCP-like connections.
We use a bufio.Reader when ordered to ensure data isn't lost.
* SSH server works!
* Start Windows support
* Something works
* Refactor pty package to support Windows spawn
* SSH server now works on Windows
* Fix non-Windows
* Fix Linux PTY render
* FIx linux build tests
* Remove agent and wintest
* Add test for Windows resize
* Fix linting errors
* Add Windows environment variables
* Add strings import
* Add comment for attrs
* Add goleak
* Add require import
Fixes#304
Unblocks #298
After logging in, the login flow should redirect to a whatever path is specified by the `?redirect` query parameter. This is important for cases like #298 - where we need to set `?redirect=%2Fcli_auth`, but also really any case where the user is linked and might have to go back to the login screen.
The fix is simple - just check if the `redirect` query parameter is set, and if it is, use that as the path to redirect to on success. Also adds a test case - we had one checking that we redirect to the default (root `/`) url, but not one of the `?redirect` param
This prevents a io.ErrShortBuffer from occurring when the byte
slice being read is smaller than the chunks sent from the opposite
pipe.
This makes sense for unordered connections, where transmission is
not guaranteed, but does not make sense for TCP-like connections.
We use a bufio.Reader when ordered to ensure data isn't lost.
Fixes#291 - renames the `expect` go package to `console`, and changes the api from `expect.NewTestConsole` to `console.New`, and a few other small changes to support the linter (ie, `ConsoleOpts` -> `Opts`)
In v1, we had a quick and easy way to leave creation forms - an escape button (and key handler) in the top right corner.
This was especially helpful in cases where the form was long enough that the 'Cancel' button was off-screen.
This ports over that component into v2 and hooks up into our two existing forms:
<img width="977" alt="Screen Shot 2022-02-09 at 5 48 03 PM" src="https://user-images.githubusercontent.com/88213859/153321503-af957f2b-d674-4ebf-9ac5-97939cb9153f.png">
In addition, this adds test cases + a storybook story for it.
This brings together a bunch of random, partially implemented packages for support of the new(ish) Windows [`conpty`](https://devblogs.microsoft.com/commandline/windows-command-line-introducing-the-windows-pseudo-console-conpty/) API - such that we can leverage the `expect` style of CLI tests, but in a way that works in Linux/OSX `pty`s and Windows `conpty`.
These include:
- Vendoring the `go-expect` library from Netflix w/ some tweaks to work cross-platform
- Vendoring the `pty` cross-platform implementation from [waypoint-plugin-sdk](b55c787a65/internal/pkg/pty)
- Vendoring the `conpty` Windows-specific implementation from [waypoint-plugin-sdk](b55c787a65/internal/pkg/conpty)
- Adjusting the `pty` interface to work with `go-expect` + the cross-plat version
There were several limitations with the current packages:
- `go-expect` requires the same `os.File` (TTY) for input / output, but `conhost` requires separate file handles
- `conpty` does not handle input, only output
- The cross-platform `pty` didn't expose the full set of primitives needed for `console`
Therefore, the following changes were made:
- Handling of `stdin` was added to the `conpty` interface
- We weren't using the full extent of the `go-expect` interface, so some portions were removed (ie, exec'ing a process) to simplify our implementation and make it easier to extend cross-platform
- Instead of `console` exposing just a `Tty`, it exposes an `InTty` and `OutTty`, to help encapsulate the difference on Windows (on Linux, these point to the same pipe)
Future improvements:
- The `isatty` implementation doesn't support accurate detection of `conhost` pty's without an associated process. In lieu of a more robust check, I've added a `--force-tty` flag intended for test case use - that forces the CLI to run in tty mode.
- It seems the windows implementation doesn't support setting a deadline. This is needed for the expect.Timeout API, but isn't used by us yet.
Fixes#241
Fixes#244
- Adds `jest-junit`
- Configures `jest-junit` to output `junit.xml` in `site/test_results`
- Uploads the emitted `junit.xml` to datadog as part of the `test/js` workflow
* Refactor parameter parsing to return nil values if none computed
* Refactor parameter to allow for hiding redisplay
* Refactor parameters to enable schema matching
* Refactor provisionerd to dynamically update parameter schemas
* Refactor job update for provisionerd
* Handle multiple states correctly when provisioning a project
* Add project import job resource table
* Basic creation flow works!
* Create project fully works!!!
* Only show job status if completed
* Add create workspace support
* Replace Netflix/go-expect with ActiveState
* Fix linting errors
* Use forked chzyer/readline
* Add create workspace CLI
* Add CLI test
* Move jobs to their own APIs
* Remove go-expect
* Fix requested changes
* Skip workspacecreate test on windows
This adds a `make install` target that copies the CLI and other coder binaries to `$GOPATH/bin` Intended to be helpful for developers who aren't familiar with `go` or the directory structure, in running `coder` CLI locally.
I noticed when showing the development flow to a few engineers, there was one awkward step - you have to `cd site` and run `yarn install`, and then come back to root to `make build` or `./develop.sh`
This just adds `yarn install` as part of `make build` and `./develop.sh` - so whichever one is run first will start work.
While working on the projects page, I noticed the `/projects` endpoint would return 'null' instead of an empty array. An empty array would simplify the behavior on the client page.
There were already tests in place for this, but they only validated that the item's length is 0... and it turns out `len(nil)` is also `0`.
So this does a couple things:
- Updates the empty-project tests to check for `NotNil` as well
- Update the project endpoints to return an empty array if no rows are returned
- Remove the hack in `/projects` page for populating data, as it is no longer needed
Since we're using GH issues to track work (thanks @misskniss and @tjcran for setting this up!) - we probably don't want the stale bot to close untouched issues every month 😬
I considered using a separate stale-duration (like, 365 days for issues) - but just turned it off now. Easy for us to change if we want to adjust the process.
An issue came up last week... our `embed.go` strategy doesn't handle dynamic NextJS-style routes! This is a blocker, because I'm aiming to set up CD on Monday, and the v2 UI makes heavy use of dynamic routing.
As a potential solution, this implements a go pkg `nextrouter` that serves `html` files, but respecting the dynamic routing behavior of NextJS:
- Files that have square brackets - ie `[providers]` provide a single-level dynamic route
- Files that have `[[...` prefix - ie `[[...any]]` - are catch-all routes.
- Files should be preferred over folders (ie, `providers.html` is preferred over `/providers`)
- Fixes the trailing-slash bug we hit in the previous `embed` strategy
This also integrates with `slog.Logger` for tracing, and handles injecting template parameters - a feature we need in v1 and v2 to be able to inject stuff like CSRF tokens.
This implements testing by using an in-memory file-system, so that we can exercise all of these cases.
In addition, this adjust V2's `embed.go` strategy to use `nextrouter`, which simplifies that file considerably. I'm tempted to factor out the `secureheaders` logic into a separate package, too.
If this works OK, it could be used for V1 too (although that scenario is more complex due to our hybrid-routing strategy). Based on our FE variety meeting, there's always a chance we could move away from NextJS in v1 - if that's the case, this router will still work and be more tested than our previous strategy (it just won't make use of dynamic routing). So I figured this was worth doing to make sure we can make forward progress in V2.
* Nest jobs under an organization
* Rename project parameter to parameter schema
* Update references when computing project parameters
* Add files endpoint
* Allow one-off project import jobs
* Allow variables to be injected that are not defined by the schema
* Update API to use jobs first
* Fix CLI tests
* Fix linting
* Fix hex length for files table
* Reduce memory allocation for windows
* refactor: Rename ProjectParameter to ProjectVersionParameter
This was confusing with ParameterValue before. It still is a bit,
but this should help distinguish scope.
* Add project version resources table
* Allow project parameters to optionally have user and workspace
* Add dry run for provisioners
* Add resource detection on project import
This just adds a very simple dockerfile for deploying `coderd` (and later `provisionerd`).
This adds a `deploy` directory at the root, and a `make docker/build` command to the makefile.
Thanks @jawnsy for the all the help 😄
* refactor: Generalize log ownership to allow for scratch jobs
Importing may fail when creating a project. We don't want to lose this output,
but we don't want to allow users to create a failing project.
This generalizes logs to soon enable one-off situations where a user can upload
their archive, create a project, and watch the output parse to completion.
* Improve file table schema by using hash
* Fix racey test by allowing logs before
* Add debug logging for PostgreSQL insert
* ci: Replace DataDog CI with custom upload script
This will reduce CI time by ~6 minutes across all of
our runners. It's a bit janky, but I believe worth
the slight maintainance burden.
* Fix test race when job would complete too early
* Fix job cancelation override
* Fix race where provisioner job is inserted before project version
