Re: proposal: possibility to read dumped table's name from file

I went and had another look at this.  The patch has been around for 18
commitfests and is widely considered to add a good feature, so it seems about
time to get reach closure.

As I've mentioned in the past I'm not a big fan of the parser, but the thread
has overruled on that.  Another thing I think is a bit overcomplicated is the
layered error handling for printing log messages, and bubbling up of errors to
get around not being able to call exit_nicely.

In the attached version I've boiled down the error logging into a single new
function pg_log_filter_error() which takes a variable format string.  This
removes a fair bit of the extra calls and makes logging easier.  I've also
added a function pointer to the FilterStateData for passing the exit function
via filter_init.  This allows the filtering code to exit gracefully regardless
of which application is using it.  Finally, I've also reimplemented the logic
for checking the parsed tokens into switch statements without defaults in order
to get the compilerwarning on a missed case.  It's easy to miss adding code to
handle a state, especially when adding new ones, and this should help highlight
that.

Overall, this does shave a bit off the patch in size for what IMHO is better
readability and maintainability.  (I've also made a pgindent pass over it of
course).

What are your thoughts on this version?  It's not in a committable state as it
needs a bit more comments here and there and a triplecheck that nothing was
missed in changing this, but I prefer to get your thoughts before spending the
extra time.

--
Daniel Gustafsson