Re: Loading dynamic docstrings and bytecode more lazily

[Helmut Eller]

On Sat, Feb 09 2013, Stefan Monnier wrote:

> So, my current decision is to keep this patch in my "treasure chest"
> because it's simply not worth the trouble.  But if someone can make
> a good case for it, I could change my mind.

I think such a lazily loaded section would be quite useful for debugging
information.  E.g. the compiler could emit tables to associate the names
of lexically scoped variables with stack offsets and also to map byte
offsets in compiled code back to the position in the source file.

Such information is necessary to make compiled code easier to debug and
potentially quite big, but of course it's not needed for normal
execution.  Therefore not loading such information before it's actually
needed would be an obvious optimization.

Teaching the compiler to emit those tables and the debugger to use them
is the real work but knowing that additional debugging information costs
"nothing" as long as it is not used, could lower the barrier the get
such improvements accepted.

Helmut