Topic equalities: 'test/topic/' and 'test/topic' should be the same

[mcollina]

At the moment they are not. It is also valid for '/test/topic' and 'test/topic'.

I am not so sure that we should fix this here or in Ascoltatori.
The main reason for handling it in Mosca is to 'fix' the packet, so even in the persistance phase should not be considered again. The other way is I think we should also fix it in Ascoltatori.

@davedoesdev, what do you think?

[mcollina]

@davedoesdev here is the "dummy" implementation with a sequence of split, filter, join.
Do you think a regexp-based implementation can be better? Could you give it a go?

Should I merge this in and adress the "performance" part of this if the need arise?

[davedoesdev]

Also need to consider duplicate topic separators:

http://mqtt.org/wiki/doku.php/topic_format

This is tricky in that it seems a shame to incur the performance penalty for 'malformed' topics.

[davedoesdev]

Advantage of normalizing the topic in mosca is each ascoltatore doesn't have to do it.
Disadvantage is each ascoltatore might be able to do it more efficiently

Reluctantly I'd say from an architectural point of view its probably mosca's job to clean up the topic before passing it on. What do you think @mcollina ?

Another way to ask the question is what should ascoltatori specify as the topic format - i.e. what should it do about trailing and duplicate separators? Once we know that then we know the answer to this issue.

[mcollina]

I think Ascoltatori should handle this, but Mosca too.
Ascoltatori needs it because it may be used outside Mosca.
Mosca needs it because it should 'fix' a packet's topic.

The current 'fix' solves it, but it is highly inefficient, as it split and join a string.

One solution might be using an lru cache for caching the fixes, but it's just cpu vs memory.

[davedoesdev]

Ascoltatori could specify that it doesn't do anything special about duplicate or trailing separators. That seems the most efficient thing for it to do.
Mosca, however, has an external spec to conform with.

[mcollina]

Seems fine: it should be handled here, and Ascoltatori should state it in the readme.

How could we do it here in a fast way? We need a single regexp that:

1. remove any / at the beginning
2. remove any / at the end
3. remove any duplicate /, such as //

[davedoesdev]

First cut:

> "//a///b/c//d///".replace(/(([^/])\/+$)|(^\/+([^/]))|(\/+(\/))/g, "$2$4$6")
'a/b/c/d'

Not very concise but the best I could come up with in limited time.

[mcollina]

Using your regexp it requires 6300ms to do 10 millions iterations, compared to 15000ms of the split/join approach! :)

[davedoesdev]

Cool! One thing I thought of is just do a match first and then only replace if the topic matches.
The match should be quicker than a replace.
This would mean 'well-formed' topics should be quicker but 'malformed' ones will be slower.
Depends whether you want to bother with that. Speed increase would only show with mix of well-formed and malformed topics I guess.

[mcollina]

In fact, it requires around 20% more time to do the replacement.
However, matching and replacing requires 2x the time.

I think I will leave it as it is, with the regexp replacing every topic, it seems the better strategy.

[mcollina]

Do you think this is a 'patch' release or a minor one?

[davedoesdev]

Unfortunately a patch I think - might change the behaviour if anyone's using a topic with duplicate separators for example.