Research Discussion: "Smash-ons" vs add-ons to simplify access to native libraries #9494
Description
Add-ons are a cool way to let node .js code access native code, be-it OS api's or other native libraries. However, deploying add-ons means that either A) the target machine must have a minimal build tool chain (which may not always be the case on a production server), or that B) the add-on must be prebuilt for all possible platforms and then downloaded (yuck). And building add-ons means being adept in C/C++ and quite familiar with V8's api's and internals.
I know other VM's offer marshalling services, like .NET's P/Invoke, that lets the VM get access to native stuff without first requiring some code getting compiled to native form.
It seems to me most add-ons today are just mostly marshallers. Now I'm sure a full-on native marshaller built as an add-on can provide the best marshalling performance, but I know from my own experience in other runtime worlds there's a bunch of use cases, like just wanting to a call a couple OS methods, that could be handled just fine, performance wise, with a built-in marshalling service.
I also know that in some cases some types are expensive to marshall, and if it's a 'chatty' api, using a built-in marshalling service wouldn't be ideal, performance wise, but would still be adequate and a reasonable trade off to the aforementioned constraints of building native add-ons.
Has anyone thought of or looked at the possibility of including a marshalling service directly in node? Besides me?
I've named a node module that calls into a native marshaller a "smash-on" because it's funny sounding, but also kinda hints at, that while easier to make, might be a tiny bit less performant in some cases.
I have ideas on how to add a marshalling service to node, and would enjoy doing so I think. Just trying to collect the thoughts and opinions of others in the know.