Hacker News Reader

Comment by DannyBee | Hacker News Reader

So, just because there seems to be confusion:

Clangd already exists. It already supports LSP. It is already part of the llvm project. It was written by !Apple, and used by a wide number of people.

(I want to make sure these people still get credit, i already see people in this thread thanking apple for supporting LSP, which is not actually what happened).

What apple has proposed contributing is: Making clangd work with xcode, and extensions to LSP.

This is great (as long as the second ends up standardized over time).

But the benefits people listed here so far already exist in clangd. The main benefit to work is "xcode support" and "better refactoring" which is awesome :)

View on HN

The proposal is to split the LSP implementation into a transport layer and a logical layer. The current LSP implementation is directly tied to JSON-RPC.

Apple's end goal is, as you say, to get clangd to work with Xcode. However, having a replaceable transport layer for LSP could benefit lots of people other than Apple.

Sure, it's nice, but again, being pragmatic, nobody else has desired such a thing that much.

Ie so far the "lots of people" have not materialized, even outside clangd.

I can certainly think of use cases for such a thing, but most people seem happy enough with json-rpc at the moment.

Part of that is likely that trying to speak LSP over a non local connection may be a fools errand anyway. It's a chatty protocol no matter what the transport.

Just because Apple is the first doesn't mean there won't be others. Maybe others have not materialized because the upfront cost of updating clangd was judged too high.

JSON-RPC has several alternatives that very large software companies have vested interests in: gRPC and XPC.

You seem more knowledgeable of these technologies than I am, though. It just seems to me a good thing that 1) Apple, and consequently, all iOS and macOS developers will be using clangd, and 2) Apple is putting significant effort into improving clangd. I'm not sure why any of that news or optimism needs to be quenched.

My point was precisely that plenty of languages have LSP implementations at this point, an f even in those, nobody has cared much to try to find an alternative to json rpc.

As for 2, all I'll say is based on history in llvm projects, I'm not anywhere near as optimistic as you. The history here is simply not one of all great things. The PR is, but the reality is much more mixed (some really great things, some really not so great). So while I certainly am hopeful, it's definitely not the unbridled optimism everyone displays here.

You don't think apple's support of llvm and clang has been important to the project? That's REALLY pushing it.

That's not what i said at all. I said some things have been really great, some really not so great. This isn't even a controversial view among long-time llvmers.

Heck, last time i asked him (a few months ago), even Chris Lattner agreed with my view, and he used to run his stuff at Apple, so i don't think it's unreasonable.

I think it's perfectly okay to be cautiously optimistic instead of fawningly happy.

Do you actually have data or a real argument you'd like to use to suggest otherwise (IE that it's been all overwhelmingly positive?)

While I agree that not everything is always great, I'd say the same for our own contributions ;) I've definitely let patches unreviewed for way too long due to internal priorities, and have gotten my share of shit at conferences for that...

In this case, the collaboration with Apple and Ericsson on the underlying architecture has been going on for a while, and while not everybody always agrees on priorities and design details, I'm thrilled to see Apple double down on their involvement here!

The [follow up](http://lists.llvm.org/pipermail/cfe-dev/2018-April/057672.ht...) message suggests that Google already internally using a non-JSPN-RPC transport layer.

Yeah, but that is likely to actually go away, not happen more. It's not a desired thing in that case, it was just easiest at the time for various reasons. We didn't want the public project to have to support our special snowflakism, hence it was done a different way.

clangd already has a layered architecture, where you can either use the JSON-RPC mechanism or more directly control clangd and implement your own transport. There are no plans for this to go away, as generally large companies often have environments in which they want better control of features than a standard protocol like LSP can give you, and we don't discourage that.

We also just talked with Apple folks at Euro LLVM about this, and while the plan for the near future is for them to have a transport conversion to JSON-RPC, I expect them to want the more fine grained control in the long term.

Really? I desire a protocol which is not coupled with JSON-RPC.

Reply on news.ycombinator.com