mikavilpas
commented
9 years ago We already discussed this in a private conversation, but I'll restate this here: there is now a branch that's dedicated to adding support for the new omnisharp-roslyn features here.
That branch so far has the build ported for omnisharp-roslyn integration testing, and it has implemented text changes for renames. I've started working on stdio support so I can drop the dependency to curl for good. This should get rid of weird errors like this and this.
bsuh
abourget
juergenhoetzel
kirill-gerasimenko
k-ode
razzmatazz
OmniSharp-Roslyn has some features that aren't in the old server that emacs doesn't support yet. I'll list them here and will add to this issue as new stuff is added.
The first big change is the notion of TextChanges. The server can return these whenever change(s) need to be made to one or more documents. A TextChange is a range of text as well as the new text that should go in that range. It could be to delete some text, add some new text or change some text at a specified location.
So the idea is that instead of returning a complete document with all the changes, you can specify that you want only the changes to a document. We currently support this for code refactorings and format document. It is far more efficient to work in this way and you don't get the jarring effect when a document is replaced. It also has the effect of moving the cursor to the correct new location - so if you added a using statement, the cursor would automatically drop down a line. In Atom and Sublime we also use these changes to format a block on keystroke ( '}', ';' ) etc. so that code is automatically indented as you type. You can see how we apply these in Atom here - https://github.com/OmniSharp/omnisharp-atom/blob/2950bbca17af9a434fbdd7be8b45675a7a6516df/lib/omnisharp-atom/features/lib/apply-changes.ts I would suggest trying out omnisharp-atom just so that you can see how this works in practice - it's very smooth. If you do decide to add this to emacs, you'll need to make it conditional as the old server doesn't support TextChanges (and never will)
The next big change is using stdin/stdout to communicate with the server. The server now supports stdio in addition to the web interface.
It works similarly to the http request/response pattern. The format of the requests and responses doesn't change - only the transport changes and we add some extra information to the packets.
The main reason that we did this was so that the server can push information to the editor without the editor requesting any data. For DNX apps, the server loads projects asynchronously and sends events to the editor as each project is added. Another benefit of this is that you don't need to worry about finding free ports etc when working with multiple solutions simultaneously. We use the V8 debugger protocol for sending the data. Each request gets a sequence ID and the corresponding response for t)at request will have a matching ID so that you can pair them up. In the future, we may be able to use this protocol to support debugging too (run until breakpoint etc). We will also probably also send errors and warnings as events that occur in the background.
To start the server in stdio mode, just add
--stdioto the command line. Again, take a look at Atom as Atom is using this transport protocol. You will also see that Atom can load multiple solutions completely transparently.Having said all that, we won't break compatibility with the way emacs does things now, so you should be able to make these changes in your own time (if you want to!)