Re: Kudos to jwe on exception handling

[John W. Eaton]

On 10/06/2015 04:07 PM, Rik wrote:
> All,
>
> I wanted to give a shout out to jwe who has been doing exceptional (pun
> intended) work transitioning the core away from using the global variable
> error_state.  The implementation began in Darmstadt just 2 weeks ago and if
> I grep through libinterp for "error_state" from that time I find 1671
> instances.  Today, the number is down to 496.  70% of the instances have
> been removed, and most of them it seems in just the last day.  Hip Hip Hurray!

Thanks, and sorry I stalled out for the last three weeks.

I noticed a few problems along the way.  The first, having to do with 
wrong-type-argument errors, was already discussed here in the "Using 
exceptions in the core" thread.  Another is that for code like this

  function f () error ("foobar"); end
  function g () 1:f:f; end
  g ()

you will see the following errors:

  error: foobar
  error: called from
      f at line 1 column 20
      g at line 1 column 15
  error: invalid limit value in colon expression
  error: called from
      g at line 1 column 15

Note the duplicate stack trace info.  This is also a problem in 4.0.0, 
so it is not related to the latest exception handling changes.  I 
introduced this behavior when I made the error function display the 
stack trace immediately.  I've just checked in a change to correct this 
problem:

  http://hg.savannah.gnu.org/hgweb/octave/rev/e0e2c2ce7e94

The idea is to store the stack trace info in the exception object that 
is thrown by error, and then display it just before displaying the next 
prompt.  I don't really like the structure yet, but this change solves 
the immediate problem.  I'd prefer to store information about the stack 
in the exception object instead of just storing a string.  But for now I 
think this works.

It would probably also be good to try to design a good hierarchy of 
classes to use for exception handling, possibly derived from the 
standard library exceptions.

Next I'll keep working on removing more uses of error_state and fixing 
problems that I've introduced (for example, the problems with glob).

jwe