Exported convenience function to get a callstack

[amitmurthy]

Currently, in case of errors, the callstack is only printed and displayed by the REPL.

It will be good to have an exported function(s) that can a) print the callstack when called and b) return a stringified version of the same.

See https://groups.google.com/d/topic/julia-users/S485_5jG2Nw/discussion

[Keno]

Base.show_backtrace(STDOUT,backtrace())
and

buf = IOBuffer()
Base.show_backtrace(buf,backtrace())
takebuf_string(buf)

?

[JeffBezanson]

Also backtrace and catch_backtrace for getting the traces.

[amitmurthy]

backtrace and catch_backtrace are already exported.

Exported helper functions can be

How about callstack() = Base.show_backtrace(STDOUT, catch_backtrace()) and scallstack for the string variant?

For a user point of view, typical use would be in a catch block to either log to a file, or display it.

[Keno]

How about this: We export show_backtrace make both arguments optional (defaulting to STDOUT and backtrace()) respectively. Then you can get the string variant using:

sprint(show_backtrace)

[amitmurthy]

We will need catch_backtrace() in a catch block, right?

[Keno]

if you want the backtrace of the exception caught, yes.

[amitmurthy]

Exporting show_backtrace with both arguments optional is good. However, I would still prefer a less verbose, convenience function for the typical user code scenario (server side code) - i.e., logging caught exceptions.

[Keno]

Well, I'm open to suggestions for names, but I think sprint(show_backtrace,catch_backtrace()) isn't all that bad. scallstack seems ambiguous since you don't know which one to print (the caught one or the current one).

[amitmurthy]

How about just callstack() which only returns the stringified catch trace. This will cover the typical scenario.
A typical debug statement to stdout will be println("Caught Exception $e at xxxx, (state: $var1, $var2) : ", callstack() ) in user code

For other scenarios, people can use sprint(show_backtrace,backtrace()) or show_backtrace(backtrace())

[Keno]

Yeah, I guess that's fine with me. I still don't think that name is ideal, but oh well. @JeffBezanson ?

[Keno]

Bump.

[ssfrr]

What if backtrace() and catch_backtrace() returned a BackTrace type, rather than just an array of Ptr{None}?

That way we can define the usual show(bt::BackTrace), etc. It seems like the existence of Base.show_backtrace is a good example of the "underscores in function names are a code smell" guideline.

[ivarne]

@ssfrr +1 for that idea.
There are probably not many uses of the raw pointer array anyway, so a simple wrapping immutable is probably not a hugely breaking change. BackTrace could also implement methods in order to behave like the original Array{Ptr{None}} (getindex, length, start).

[spurll]

Happened upon this while debugging this PR. Seems relevant: #14469

[KristofferC]

I think this is fixed with the stacktrace() function that gets printed as:

julia> stacktrace()
4-element Array{StackFrame,1}:
 eval(::Module, ::Expr) at REPL.jl:3                                             
 eval_user_input(::Any, ::Base.REPL.REPLBackend) at REPL.jl:69                   
 macro expansion at REPL.jl:100 [inlined]                                        
 (::getfield(Base.REPL, Symbol("##1#2")){Base.REPL.REPLBackend})() at event.jl:9