GitLab

We never have a chain of the transitions. We only do start and zero or
one handle on the returned one. Simplify handle to return void, not
NULL.

Besides simplification, it allows testing without fake states (TBD) - we
had no real state to test the multiple transitions.

Conflicts (from cherry-pick):
	src/lib/dns/tests/master_lexer_unittest.cc

The copy is shallow anyway, so it wouldn't survive for long.

If there's an error that manifests at the end of a file, like an
unbalanced parenthesis or a quoted string that does not end, report the
error. But also, return EOF as the next token, not an exception.

It was deemed unneeded and is removed therefore, to make the code
smaller.

A scoped pointer, in this case. It probably isn't safer, as the ->handle
couldn't throw in this case, but just for the habit.

It was not safe, because it may refer to some data elsewhere. Luckily,
it seems it is not needed anyway, since the only way that a token can
get out of the lexer is through the getNextToken, which would get a new
one, and through a testing method not meant for public use.

Move the ReadError exception into the MasterLexer. That way, it is in
public namespace, so people can expect to see it.

This is a debug test. It checks the state machine produced a token.
There are no tests for this test, since such condition would be a
programmer error, not something expected.

No error handling, no special cases, no returning.

No fancy stuff yet, like ungetting tokens, changes of parentheses count,
etc. Just the basic stuff.

It is for tests, but in the main code for technical reasons.

With doxygen description of what it should do and empty method bodies,
so the tests in following commits will at least compile.

Also fix some issues from the review.

note I renumbered existing entry.

I expect we'll need the string object version of token pretty often
for rrtype and TTL, and this version will help make it a bit more
efficient.