[bv-dev] Invoking Configuration#buildValidatorFactory() several times

[Hardy Ferentschik]

On Mar 26, 2012, at 10:48 PM, Matt Benson wrote:

> On Mon, Mar 26, 2012 at 3:47 PM, Matt Benson <gudnabrsam at gmail.com> wrote:
>> On Mon, Mar 26, 2012 at 3:41 PM, Gunnar Morling
>> <gunnar.morling at googlemail.com> wrote:
>>> Hi,
>>> 
>>>>  In this case it would still be up to the implementation to support > 1 calls to Configuration#buildValidatorFactory()
>>> 
>>> In HV, we're currently thinking about using InputStream#mark() and
>>> reset() to re-read the mapping streams. That the stream supports
>>> mark/reset, would have to be ensured by the user, or alternatively we
>>> might wrap the given streams ourselves into resettable ones. It might
>>> make sense that the spec mandates such a behavior.
> 
> Good point, Gunnar.  For some reason I failed to think about the much
> less expensive idea of simply calling #isMarkSupported() and wrapping
> to a mark-supporting version, though e.g. using BufferedInputStream
> this still requires that the length of the stream not exceed the mark
> limit... :\

I think using mark and reset is the best solution. I don't think we have to worry
about the length of the stream exceeding the mark limit. We are talking configuration
files after all. Via InputStream#mark(int readlimit) we can even set the limit to something 
like Integer.MAX_VALUE.

The question is whether we should wrap an input stream which does not support mark/reset 
into a BufferedInputStream. I am tempted to say that it is the responsibility of the client to provide
a buffered stream in case it wants to reuse the configuration. Of course the documentation has to
be clear about this.

--Hardy