Multithreading and Loggers

[TheFlyingFiddle]

On Friday, 31 January 2014 at 13:48:09 UTC, Nicolas F. wrote:
> Hello,
>
> I'm having some problems wrapping my head around shared classes 
> and whatnot.
> So my application has a Logger class, which basically just 
> writes formatted messages to a target, of which debugGLLogger 
> inherits. What debugGLLogger does is simple in theory: it 
> supplies a callback function to OpenGL, and said callback 
> function uses a logger to write debug messages coming from 
> OpenGL to.
>
> First off, I can't control the fact that it's a callback 
> function, or when it is called, as this is done by the driver, 
> windowing system, and so on.
>
> Hence, to know which logger I need to write to, I used a static 
> variable which gets set to my main logger for OpenGL.
>
> In theory, this should work, as I'm currently (or so I thought) 
> not using concurrency, which means that the static variable 
> should exist in all cases.
>
> However, apparently the callback function is possibly called 
> from a different thread which I have no control over, as I've 
> had to painfully discover. My mainLogger static variable is 
> useless, and everything falls apart. I'm not sure where this 
> thread comes from, and maybe I'm just imagining things, but 
> since I see no other way the variable could suddenly become 
> null, I'm guessing that's the issue.
>
> So in some way I've got to bring the "shared" keyword into 
> play, so I have a reference to a logger instance which I can 
> use across threads. However, apparently for that, the class 
> also needs to be shared, and I'm not sure if that's even the 
> "right" way to do things.
>
> I've also thought that maybe I should utilise a seperate thread 
> to run all loggers in, and communicate with the loggers through 
> message-passing.
>
> I'm not an experienced programmer, so I'd love to hear some 
> advice on how I could solve this.

You could always use __gshared instead of shared. Then it becomes 
a global static variable and you don't have to deal with the 
current shared mess. It comes with the canveat that it's not 
safe. (as in not data-race safe)