When re-importing a test database, if the source archive for that
database is not in the workspace, then old source code will be seen when
inspected.
To reproduce:
1. Run a ql test file with a failure (I'm using a javascript DB).
2. Right click and import on the db that sticks around.
3. Change the JS source file for the test.
4. Re-run and still have a failure.
5. Re-import the database
6. Run the query under test
7. Click on a result item
8. **BUG:** The source code is old
The problem is that the source archive cache is not being flushed in
this case. I added a case to flush the source archive when the archive
was imported into the workspace as a folder, but not when the archive is
external.
The fix is to listen for database remove events in the archive
filesystem provider and flush the associated database source
archive.
There is a complication:
The database remove event didn't emit the removed database. This is
because the only place that consumed the event didn't need it.
To get around this, I changed the structure of the events. I added a
new `fullRefresh` boolean. If true, then the original database change
handler will ensure the entire tree is refreshed. If false, only the
selected database.
Since we're only supporting the new query server, we can remove the
`QueryRunner` abstraction and just use the `NewQueryRunner` as a
concrete `QueryRunner` without an abstract base class. This simplifies
the code of the query server and removes some unnecessary indirection.
This removes the use of the CLI `codeql resolve library-path` command to
detect the language of a query pack. Instead, it uses the `qlpack.yml`
file to determine the language. This is slightly less correct since it
only works for `codeql/${language}-all` dependencies, but it is much
faster and more reliable. It also doesn't result in the CLI server
restarting for invalid query packs (such as in the `github/codeql`
repository or in any workspaces containing it).
This will avoid showing a query-server popup on hovers. When someone
does CTRL+hover on source code in a database archive, the definitions
provider is triggered. The first time this is run, there will be a query that
is invoked in order to find definitions. This can be a long process.
Previously, this query brought up a popup window that could be cancelled.
However, this is confusing users since the query is automatically cancelled
when the mouse hovers away from the element.
The downside of this PR is that even when find definitions is explicitly invoked,
(using F12), there will still be no hover.
I think this is an improvement, but I am happy to discuss if others disagree.
This will change the behavior of the "Create new query" command to
create the new query in the same folder as the first selected item in
the queries panel. If no items are selected, the behavior is the same
as before.
I've used events to communicate the selection from the queries panel to
the local queries module. This is some more code and some extra
complexity, but it ensures that we don't have a dependency from the
local queries module to the queries panel module. This makes testing
easier.
This will change the "Create new query" command to use the selected
language when creating a new query. If no language is selected, it will
still prompt the user to pick a language.
When calling for example `showAndLogExceptionWithTelemetry`, the stack
trace would be sent to Application Insights, but there was no way to
see the stack trace from within VS Code. This will add the stack trace
to the log by returning it from `fullMessageWithStack` and using it in
the appropriate places.
This change sets the `fullMessage` of the
`showAndLogExceptionWithTelemetry` to include the stack trace. This
makes it possible to find the source of the error rather than only
knowing that a specific error occurred. If the error does not have a
stack trace (which should be rare) the message will be the same as
before.
