This is a key cause of not being able to produce interpreted results, so
logging it helps us debug a lack of interpreted results.
Also make the database metadata check async
The PR contains the initial implementing of QL Test support in CodeQL for Visual Studio Code. Because QL Test support isn't quite ready in the CLI yet, this PR uses `odasa` to run the tests for now. As CLI support comes online, it should be straightforward to swap out the implementation to use the CLI.
The treeview UI for the tests is implemented via the `hbenl.vscode-test-explorer` extension. This extension is open source, and appears to be actively maintained. It's used by a couple dozen existing extensions for tests for various languages. The extension doesn't really do anything on its own, so taking it as a dependency isn't introducing any unwanted UI clutter. Note that I did have to remove the `--disable-extensions` argument from `launch.json`, because otherwise the test explorer extension gets disabled, preventing our own extension from loading.
The UI will display a root node for each QL pack that contains tests, with the actual test directories and files as descendants of that root node. We consider only those QL packs in the workspace; QL packs on the default CodeQL search path are ignored. We use `codeql resolve qlpacks` to find the packs, and then watch all `qlpack.yml` files in the workspace for changes in order to refresh the pack discovery when necessary. Ideally, we'd have the CLI return a set of path patterns to watch, but for now the current implementation works fine.
To discover the tests within a given pack, we walk the pack's directory tree manually for now, until the relevant CLI command is available. Because we do not yet have a mechanism in `qlpack.yml` to specify whether or not the pack contains tests, we assume that any pack whose name ends with "-tests" to contain nothing but tests, and any other pack to contain no tests. This is sufficient for the tests in the QL repo. As with QL pack discovery, we watch the file system for changes in `.ql` and `.qlref` files in order to refresh the tree of tests if anything changes.
To actually run the tests, we just invoke `odasa qltest` with the appropriate arguments. This code is pretty much a straight copy-and-paste from the repo where I've had a private version of QL Test support for several months. Once we can run tests via the CLI, this will all be deleted.
The `test-ui.ts` file implements a couple of additional commands for the context menu of the test treeview. You can accept the output of a failing test (copying the `.actual` file to the `.expected` file), and you can bring up a diff view of the `.expected` and `.actual` files).
This PR includes a couple of related utility classes. `UIService` makes it a little easier to implement a service that handles VS Code commands. `Discovery` is a base class that handles most of the work that is shared between the different kinds of discovery that we do, like avoiding running multiple discovery operations simultaneously if we get a storm of file change notifications.
This was a simple case of passing entirely the wrong data as an
argument to the `upgradeDatabase` call in `queries.ts`. We were
passing the list of search paths for upgrades, when we should have
been passing the list of directories directly containing upgrade
scripts to be applied.
