[Webmacro-devel] Re: [Melati-developers] WebMacro Test Case: TestSharedOutputStream

[Tim J. <ti...@ho...>]

Lane,

I have only spent 5 mins on this, so much of this is from memory, but here
is my 2p.

> The problem, as presented by Keats, is that in 2.0, WM renders a
> template to a backing store using a FastWriter impl instance, not to the
> outputstream. I kinda knew this to be the case but could not verify it.

it always did this (buffered the output stream).

> Therefore, prior to 2.0, Melati exploited its knowledge of the behavior
> of FW. With 2.0, this knowledge of the WM implementation is no longer
> correct. WM writes to a backing store in 2.0, not directly to the output
> stream.

ok.

> The FastWriter instance used internally to write out a template is not
> visible to the template in 2.0. In 2.0, you write to an output stream or
> to a string. In 2.0, as has been noted, the reference to the FW instance
> used to render the template internally is not available.

> Personally, I think it is bad form that melati exploited internal
> knowledge of the FW mechanics. This broke the encapsulation contract and
> now melati is broken. The melati template variable, $ml, instead of
> returning strings for method calls, has an internal operation which
> writes to the output stream. Since the template is buffered, this write
> op ends up at the front of the output.

i would not cast blame here.  the outputstream was (and still is) exposed
therefor it is not unreasonable to reference it elsewhere.  without an
explicit unit test the contract is ambiguous, and melati made the assumption
that it was allowed to make direct write calls to the outputstream.
similarly, changing the implementation of FastWriter is also totally
reasonable.

> My suggestion at this point is to back out the exploited knowledge. $ml,
> a context variable and object, should return renderable text, not
> attempt to write to the output stream. This would decouple it from
> internal knowledge of the templating frameworks and standardize its
> behavior in a normal manner for all template engines, not just WM.

why can't we just flush() before we write our stuff.

timj