>>>>> "Ulrich" == Ulrich Weigand <uweigand@de.ibm.com> writes:

Ulrich> My choice would be to just go with what you have now.

Ok.

Tom> * print_scalar_formatted probably needs an options argument as well.
Tom> What do you think?  This worries me a little since it is another 47
Tom> call sites; so it could make the patch even bigger.  Maybe this
Tom> ought to be a follow-up.

Ulrich> This looks more critical, and could in fact even be a bug: for example,
Ulrich> in the cases where print_scalar_formatted recurses back to val_print,
Ulrich> you now lose the inspect_it setting ...

Good point.  I implemented this.  It turned out to be pretty easy;
most of the call sites already had an option structure available.

Ulrich> One other issue I noticed: the deref_ref flag seems to be handled
Ulrich> incorrectly with your current patch.  It is always 0 in the structs
Ulrich> returned by the get_... routines, and never set to any nonzero value
Ulrich> manually either, as far as I can see.

Thanks.  The appended tries to fix this.  I took a behavior-preserving
approach.  I looked at all the val_print-like calls.  If the code
before the patch passed 'deref_ref' to such a call, I ignored it.
Otherwise I made a copy of the options and set deref_ref
appropriately.

In some cases I suspect the previous code is really a "don't care" and
we could just pass through the options.  But, I think that would be
better done by someone more familiar with each call site than I am.

I'll start testing this and writing the ChangeLog entry.

Tom