Javascript parser swallows errors thrown in command callback

[dainis]

Looks like error is here : https://github.com/mranney/node_redis/blob/master/lib/parser/javascript.js#L213

This try/catch catches also errors thrown from callbacks, but error isn't thrown/emited to client

in case of pubsub message same message will go on for ever

var redis = require('redis'),
    r = redis.createClient(3052, '127.0.0.1'),
    r2 = redis.createClient(3052, '127.0.0.1');

r.on('ready', function() {
    r.subscribe('channel');

    r.on('error', function(e){
        console.log(e);
    });

    r.on('message', function(channel, data) {
        console.log(channel, data);
        throw new Error('test');
    });
});

r2.on('ready', function() {
    var i = 0;
    console.log('r2');
    (function p() {
        r2.publish('channel', ++i);
        setTimeout(p, 1000);
    })()
});

output:

channel 2
channel 2
channel 2
...

Ready to merge checklist

• test(s) in test.js
• tests will fail without the PR, but succeed once applied
• docs, if applicable
• merges cleanly
• coding style (4-space indents, looks similar to other code)

[cxreg]

Note, this addresses the same issue as #357.

[brycebaril]

I think this is on the right track, though there is another Error the parser can through on incomplete read frames, the line https://github.com/dainis/node_redis/blob/master/lib/parser/javascript.js#L276

            throw new Error("didn't see LF after NL reading multi bulk count (" + offset + " => " + this._buffer.length + ", " + this._offset + ")");

I'm pretty sure that is another flavor of the 'too far' type Error here.

In terms of naming, given that these 'too far' Errors are fairly expected, maybe name the Exception something like IncompleteReadBuffer or something that better indicates what is going on. (I.e. the TCP packet was broken up in the middle of the reply it is parsing, so it needs the next packet to append to the buffer before it can continue)

This is definitely a very large issue with the current parser and one of the first things we'd like to address in the next version of node_redis.

[brycebaril]

This work was done as part of #382