Had another 48hr game jam this weekend...

[Walter Bright]

On 9/1/2013 5:14 PM, H. S. Teoh wrote:
> But I will say that debugging can and must be improved. Currently, about
> the only thing usable for dmd -g is to get a stacktrace of a program
> crash. Nothing else seems to be properly supported (I use gdb). Stepping
> through statements and setting breakpoints more-or-less works, but I
> can't get at most variables (keeps complains about being unable to
> reference 'this' or something similar), sometimes variable values are
> outright wrong or completely unrelated to the actual value, sometimes
> variables shown right on the source line being debugged don't exist in
> the debugger ('no such symbol'). Unable to look into nested structs
> without hitting odd behaviour. Doesn't understand D naming conventions
> (or does so poorly).

I've tried to figure this out many times, and have found the Dwarf format to be 
completely impenetrable. As far as I can tell, dmd generates correct Dwarf 
symbolic debug info according to the Dwarf spec.

gdb relies on very undocumented idiosyncratic combinations of things to work. It 
appears to be wrapped tightly around exactly what gcc generates.

I suppose the only way to figure it out is to read the gdb source code. Sigh.

BTW,

     dumpobj -p

on linux will pretty-print the Dwarf symbolic debug info. If anyone wants to 
take a stab at it, please do so!