Re: [PATCH v11 05/13] ref-filter: implement an `align` atom

[Junio C Hamano <gitster@pobox.com>]

Karthik Nayak <karthik.188@gmail.com> writes:

> On Mon, Aug 17, 2015 at 7:37 AM, Eric Sunshine <sunshine@sunshineco.com> wrote:
> ...
>> Second, I realize that Junio suggested the 'return_to' idea, but it
>> seems like it could become overly painful since each handler of this
>> sort is going to have to perform the same manipulation to append its
>> collected output to its parent state's output. What if you instead
>> make it the responsibility of pop_state() to append the 'output' from
>> the state being popped to the "prev" state's 'output'? This way, it
>> happens automatically, thus reducing code in each individual handler,
>> and reducing the burden of having to keep writing the same code.
>
> Good question, what if we don't want to append to strbuf at all?
> For e.g., We were discussing an "%(if).....%(then)......%(end)"
> atom structure, here if the if condition isn't met we wouldn't want to
> append to the prev strbuf, hence I thought it's better if the handler
> decided whether or not to append to prev strbuf.

I'd imagine the implementation of these to be along the lines of
(thinking aloud):

 - "%(if:[nonempty|empty|...])" pushes a new stack, and sets its
   attend/at_end/end_scope function to signal a syntax error.  It
   also records what condition (e.g. "nonempty") to use in the new
   stack.

 - "%(then)" inspects the top-of-stack output and uses the condition
   recorded by the %(if) that created the stack to decide true or
   false.  The stack element pushed by %(if) is then removed.
   Notice that the end_scope function prepared by %(if) is never
   called.

   Then (no pun intended):

   - If true, that means we would want the (expansion of) text up to
     "%(end)" or "%(else)", whichever comes first, appended to the
     surrounding output.  Push a new stack and set its end_scope
     function to the one that appends the top-of-stack output to the
     surrounding output, expecting %(end) will follow without
     %(else).

   - If false, that means we would want the (expansion of) text up
     to "%(end)" or "%(else)", whichever comes first, discarded.
     Push a new stack and set its end_scope function to the one that
     discards the top-of-stack output, expecting %(end) will follow
     without %(else).

 - "%(else)" inspects the top of the stack, and if it is not left by
   "%(then)", signal a syntax error.

   Else (no pun intended), it runs the end_scope function left by
   "%(then)" on the top-of-stack output (e.g. if "%(then)" found
   that the condition holds true, the accumulated output at this
   point should be appended to the surrounding output, and it was
   expected to be done by "%(end)" if this "%(else)" weren't
   present.  We do it here before waiting for "%(end)").

   Truncate the top-of-stack output, flip the end_scope function to
   the one opposite from the one left by "%(then)".  And let "%(end)"
   take care of it.

 - "%(end)" just unconditionally runs the end_scope function on the
   top of the stack output.

Eric's suggestion is let the caller of the end_scope function to
always append the output of the top-of-stack, and I think it makes
sense.  It makes a common "%(atom)" implementation simpler.  Things
like %(then) and %(else) above need to make sure that they reset the
top-of-stack output to empty as necessary, but that is not such a
huge implementation burden---their operation is already unusual and
needs to be more complex than the plain-vanilla %(atom)s anyway.