Re: error with d28016d16e9a (change to jsonencode.cc)

[John W. Eaton]

On 10/1/20 1:24 PM, Rik wrote:

> In one sense, this is easy to resolve.  Instead of using the keyword 
> "auto" just specify the exact type of the underlying variable. However, 
> the compiler really should be able to handle this construct.  Could you 
> try 'g++ --version'?  And is it really GNU g++ or is this a soft link to 
> clang?

It looks like accepting auto in this context is a C++14 feature.

Also, I noticed we can simplify some things.  The unwind_action allows 
writing things like

  int val;  // Will be captured by the unwind_action objects below.

  unwind_action act ([val] () { use_val (val); })

or

  unwind_action act ([] (int val) { use_val (val); }, val);

in the first case, VAL is captured directly by the lambda expression and 
is not passed as an argument to the function it creates.  In the second, 
the unwind_action object captures the value and passes it as an argument 
to the function that the lambda expression creates.

Both forms have the same ultimate effect, but the second is useful when 
you want to capture a value that is not directly available as a variable 
and can't be captured by the lambda.  For example, in

  octave::unwind_action restore_warning_state
    ([] (const octave_value_list& old_warning_state)
     {
       set_warning_state (old_warning_state);
     }, set_warning_state ("Octave:classdef-to-struct", "off"));

the warning state is only accessible through function calls in this 
context, so we can't capture it directly in the [] part of the lambda 
expression, but in other places, we could write

  octave::unwind_action close_outfile
    ([outfile] () { std::fclose (outfile); });

instead of

  octave::unwind_action close_outfile
    ([] (const auto file_ptr) { std::fclose (file_ptr); }, outfile);

Using either form should work but maybe we should use the direct lambda 
capture form when the values we need to capture are variables accessible 
in the current scope?

jwe