Need to work out protocol to determine "workspace".

[kaby76]

A client starts out an LSP session with the server with the Initialize call. This remote procedure call contains a "rootUri" string that specifies a directory for a "workspace". Unfortunately, the problem with this idea is that a directory is insufficient. The concept of what constitutes a program could be a solution file (.sln), a project file (.csproj), a Maven file (.pom), a CMake CMakeLists.txt file, or even a good ol' fashion Makefile. These files specify exactly what makes up a "workspace". I don't know why the designers decided that a directory is sufficient, or even relevant, to what is a program, what the components of a program are, and where to find them. But, this is what it is.

Right now, the code is implemented with the following assumptions:

• "." contains all of the program source code;
• the source is C#;
• any Antlr "import" statements assume "." for finding the imported grammar;
• any lexer grammar is in ".".

The problem with this is that doesn't work as well as it should. The code may not find the start symbol in the C# code, because I only do a Roslyn parse, not a full query.

I need to add to Antlrvsix a protocol to query from the client the program definition here. It is impossible for me to know what grammars to read, and it is only guessing otherwise.

[kaby76]

According to https://github.com/microsoft/language-server-protocol/issues/703, it looks like what LSP designers want is "to use the workspace.getConfiguration() request" to help resolve opening a "workspace"--whatever that means, if anything at all. The rootUri is just a directory. But, since a directory can contain thousands of files that are not part of a particular program, the rootUri is basically useless non-information. I've seen a directory contain a .pom, .sln, and .csproj all in one lovely directory. Antlr is Antlr, but it could be targeted for any language, and it can contain target-specific action blocks. Believe me, Antlr breaks everything in LSP. So, when initializing, send a query to the client to figure out what the user is doing here, what the "workspace" really is, i.e., a .pom, a CMakeLists.txt, a makefile, a .sln, .csproj, whatever. Otherwise, you can't compile jack, you cannot find the Antlr generated Listeners or Visitors, can you determine the start symbol for the grammar because it's a method call in your damn program.

Once I have this worked out, I can then compile the program and determine the target language for the Antlr generator. I can then compile and search properly for all start symbols; I can perform tagging of action code blocks; I can change the name of visitors and listeners when renaming an grammar symbol.