amilajack / node-embed Goto Github PK

The goal of this codebase is to help anchor a discussion of what an embeddable interface to Node.js would look like with running code.

• Embedding node.js today is not only a desirable goal, it is actively being done by a number of projects despite the lack of support.
• Lack of support starts with lack of documentation.
• The one stable interface node::Start() is a misnomer as it not only starts node, but it also executes, and terminates node. This rarely is what an embedder wants.
• Embedding node today requires extracting key bits from the source files, and knowledge of the build system, platforms, and dependencies (v8, openssl, etc).

• To start with, one needs to clone and build node from GitHub. This code has been tested on Mac OS/X and Ubuntu, and is likely to work on other Unix like operating systems.
• The next step is to run the provided mkmf.js and specify the directory for the built node tree. This builds a standalone Makefile for your platform.
• The final step is to run make. This will build an executable named ./node_main which you can run.

• The mkmf.js is pretty self explanatory.
• node.cc is an updated version of src/node.cc from the node repository with the following changes:
  • First, it is a subset so as to not cause duplicate symbols when linked with libnode.a. This part is done for demo purposes only, the intent is that this will be abandoned, and the results of the next two steps will be folded into node.js itself.
  • Second, node::Start has been split into node:Setup, node::ExecuteString and node::Teardown.
  • Third, a C interface to invoking these functions is provided.
• node_embed.h contains the C API
• node_main.c is a demo program that calls nodeSetup, successively calls nodeExecuteString and then finally nodeTeardown.

• I believe that creating a standalone Makefile and application should be a part of the CI for node.js. This demo shows how it could be done for discussion purposes before it is implemented "for real".

• Should we retire node::Start or implement it by calling these new wrappers?
• Is argc/argv the right interface for an API? Should the defaults be different (example: should no arguments trigger a REPL) when called by an API?
• Splitting node::Start into three steps means that some data is allocated on the heap instead of the stack, and some things (most noteabley locks) are obtained and released multiple times. My intuition is that this is negligible when compared to the overhead of starting a complete node executable, but that needs to be measured.
• Currently, nodeExecuteString is synchronous is that it waits until the event loop is emptied before returning. Should an asynchronous version be created, and if so, what is the desired semantics?

• This demo only shows incrementally updating a single execution context. It should be expanded to include creating multiple contexts (either concurrently or serially).
• The interface should be NAPIized, and make to work on Microsoft Windows. Help from the ChakraCore team would be greatly appreciated.
• While I suspect that only the C interface needs to be officially maintained, providing examples in Go, Java, Python, TCL, Ruby, and others would be valuable and straightforward.