Useful to know this as there's implications with threads and the static
state variable used for random numbrs.

Keep a cache of any records which needed converting to the ASCII
locale used for searching.

These changes would have to be broadened if somebody wanted to
search in a more complex locale; currently the only characters
allowable are [a-z0-9]. It would need a greater understanding of
how SDL handles the user input.

It's valid to call free() on a NULL pointer, so use this.

We have limited options for text rendering, so must do a conversion
from the system locale to UTF-8 where necessary.

Deliberately not using TTF_RenderUNICODE and wchar_t conversion,
despite it being simpler;  because the system wchar_t is 32-bit,
but SDL's idea of unicode is 16 bit.

Leave iconv out of the path of the rendering that happens every
frame as these are known non-unicode strings.

Increase the LFSR sequence for a longer cycle time, and reduce the number
of bits driving the randomness so that it takes even longer to cycle.

Thanks to Stefan Berg-Johansen for a high quality bug report.

> The type declaration on line 79 in`player.c` should read `static
> unsigned short x = 0xbabe;`. In the current master branch, `x` is a
> static _signed_ short, which results in it taking on negative values
> outside of the intended range. Another side effect is that the
> returned sequence happens to converge after about 15 calls to a
> small negative number, so effectively the current dither function
> only imposes a small constant offset and nothing else.

This is the first time we've done this twice on one release; shows
how long some of these patches have been waiting for.

Thanks to Alessio Treglia <alessio@debian.org> for providing the basis
of this patch.

Message-ID: <700845d3-c785-2e79-5ce0-f637fdae448e@64studio.com>
Requested-by: Daniel James <daniel@64studio.com>

Because grep returns non-zero error code if no matches were found,
the script considers this an error condition.

Switch to using "find -regex". I chose the posix-egrep syntax as this
seems to be one of the standard formats which gives the most concise
syntax for the regex itself.

When this code was originally put in place there was some debate, and
I wrote on the mailing list (29 Jan 2012):

> I merged the changes just now, but I did it using grep instead of
> iregex. And found that grep does not do a full-string match by default.
>
> I chose grep -E for a couple of minor reasons:
>
> * more widely available
>
> * matching in a separate thread to the 'find' could give marginally
>   better performance
>
> Nothing seriously wrong with the other method though :-) Anyway, it's in
> the beta now, so everyone can give it a test.

In light of the information on error conditions, switch to the regex
case. This breaks compatibility with non-GNU find implementations.

As well as clarity to the implementation there is some flexibilty here
if we might wish to control other video flags.

We get the following warning on build:

In file included from /usr/include/assert.h:35:
/usr/include/features.h:148:3: warning: "_BSD_SOURCE and _SVID_SOURCE are
      deprecated, use _DEFAULT_SOURCE" [-W#warnings]

This is a FAQ in the linux-man project [*]

  You are compiling the code on a system that has glibc 2.20 or later
  installed. In glibc 2.20, the _BSD_SOURCE and _SVID_SOURCE feature
  test macros were deprecated

[*] http://man7.org/tlpi/code/faq.html

For consistency with the system manual pages. The formatting
and readability of the source will have to wait.

* deck:
  A dummy deck for testing
  Refactor deck creation
  Make deck responsible for importer
  Make deck responsible for the protect flag
  Make deck responsible for timecode init
  Rename local variables

The dicer flag is now in the 'long' format.

Make it easier to create decks in other parts of the code.

First step in making the code a bit more logical here, but we are
going to have to pass a lot more variables into the 'deck'.

Avoid confusion with the global 'deck[]', with which we are about to
do some refactoring.

Check if we used an optimal LFSR or not.

As static memory is implicitly zero'd, the defaults are 0 and false.
This is rather useful.

Previously we were setting the maximum buffer size. For some soundcards
this could allow the device to selecta much smaller buffer, leading to
overruns.

Be more specific by using _near, which should match the user's expectations
and perhaps will fix problems with some soundcards.