you have to multiply current object matrices and transformation in opposite direction. ObjectPos*=Transformation give transformation respectively to global basis, while ObjectPos=Transformation*ObjectPos give transformation respectively to object's local basis.

also it required to store current object transformation in separate matrix and multiply pvm to it before pass to shader for each object that you render  . Object's world position is independent from view an projection,  so have to be stored and manipulated on animation/simulation stage separately for each object,and combined with view and projection only on rendering stage.

http://jsfiddle.net/nmjuptsg/

enjoy

Off topic: Please don't be sorry to ask in JS. In fact, it's even more convienient than says C++ or C# as it's easier to run :).

On 10/1/2018 at 3:05 PM, mr_tawan said:

more convienient

Really JS have no any modern core language tools unlike C# and especially C++, and have a huge disadvantages like a dynamic typing, dynamic object signatures and GC,  but for browser-based solutions it just no other widely used langueges yet.

11 hours ago, Fulcrum.013 said:

Really JS have no any modern core language tools unlike C# and especially C++, and have a huge disadvantages like a dynamic typing, dynamic object signatures and GC,  but for browser-based solutions it just no other widely used langueges yet.

I'm in a place with no compilers at all (and cannot install one). In this case I'd say it's more convenient. ... well may be at least for me I guess. 

1 hour ago, fleabay said:

This doesn't work correctly for me. I've tried it on 2 different computers with 2 different browsers.

Looks like it have some problems with drag detection so don't stop rotation when mouse released. But  rotation made respectively to local axes. Verify your code to make sure dX and dY is zero after drag finished .

http://jsfiddle.net/z85smbxy/

fixed.