[erlang-questions] Heads-up: The cost of get_stacktrace()

[Richard Carlsson]

On 2013-11-10 13:39, Anthony Ramine wrote:
> Could your system use erlang:raise(error, {foo,{Class,Term}}, erlang:get_stacktrace()) instead of returning things?

Probably, but the (somewhat convoluted) logic needs to be rewritten to 
catch that exception further up instead of expecting an error tuple as 
the return value. We'll either be doing something along those lines, or 
avoid the need for an intermediate catch altogether.

> The call to erlang:get_stacktrace() could then be optimized away by the compiler and no informations would be lost ever.

Interesting idea, and might be worth doing. You'd have to be sure that 
the un-expanded stack dump only gets passed into the third arg of 
raise/3, and doesn't escape to any other Erlang code.

Just note that re-throwing with raise/3 should be used very carefully. 
If you substitute the exception class and term while keeping the stack 
trace pointing to where the original exception occurred, someone trying 
to debug the crash might then look at the source code at that point, 
realize there's no way it could result in that kind of exception, and be 
greatly confused. It's generally preferable to re-throw only with the 
exact same class/term.

    /Richard