Pass the user data dir to the browser environment service then derive
all the paths we can based off that path like the global storage path
which the vim extension uses to store history (otherwise it gets stored
in the working directory from when code-server was spawned).
Arguably the better solution is to use the userdata scheme but that
won't work because the vim extension ignores the VS Code API.
Fixes#1551.
See #1532 for more context.
- Errored JSON requests will get back the error in JSON instead of using
the status text. This seems better to me because it seems more correct
to utilize the response body over hijacking the status text. The
caller is expecting JSON anyway. Worst of all I never actually set the
status text like I thought I did so it wasn't working to begin with.
- Allow the update error to propagate for JSON update requests. It was
caught to show the error inline instead of an error page when using
the update page but for JSON requests it meant there was no error and
no error code so it looked like it succeeded.
- Make errors for failed requests to GitHub less incomprehensible.
Previously they would just be the code which is no context at all.
The raw value is now passed back to VS Code so it can do the parsing
with its own URI class rather than trying to parse using Node's url
module first since that has no guarantee of working the same way. It
also lets us keep the vscode-remote bit internal to VS Code.
Removed the logic that keeps trying paths until it finds a valid one
because it seems confusing to open a path and silently get some other
path instead of an error for the one you tried to open. Now it'll just
use exactly what you specified or fail trying.
Fixes#1488. The problem here was that url.parse was encoding the spaces
then the validation failed looking for a literal %20.
