The color_paragraphs method of DocumentClass attempts to iterate through all the defined paragraph_type methods for each paragraph.  If a paragraph_type method returns anything but None, the colorization has been completed by the paragraph_type method, and we move on.  However, when a paragraph_type cannot be found to colorize a paragraph, we make a copy of the paragraph, and attempt to colorize its children with the color_paragraphs function.  In this case, the DOM-style attributes attached to the original paragraph were lost.  I modified the color_paragraphs method to retain the original paragraph's DOM-style attributes when a colorizer for the paragraph was not found.

be ignored.  This simplifies the logic a bit.

Be more careful about importing setpath.

expressionCompiler is passed in.  Import it when needed.

us any more, but we may need it).

Don't import Products.ParsedXML.

Removed assignment to wordMap that called lexicon.items() in getEntryForObject.  It was a nonsensical call that must have been detritus from the various overhauls done over time.  Also changed two comments to two methods so they properly say what they're doing.  Thanks to Erik Enge for the bugreports.

StructuredTextNG's HTMLClass ordered and unordered list output (for numbered and bulleted items respectively) differed unnecessarily from ClassicStructuredText's HTML output by not inserting a surrounding the list text with a paragraph tag. Both now do surround list text with a paragraph tag.

Specifically,

_loadSerialEx(): New method which is similar to the API method
loadSerial() except that it returns both the pickle and the version
string for the specified oid+serial pair.  This is not a public
method.

loadSerial(): Reimplement in terms of _loadSerialEx() to reduce code
duplication.

iterator(): Public method to return a "transactions iterator".

_nexttxn(): Helper method to return the transaction metadata for the
transaction following the given in the argument.  Raises IndexError if
there is no next transaction.

_alltxnoids(): Helper method to return a list of all the oids modified
in the given transaction.  Note that the Full Berkeley storage has the
implied semantics that if an object is modified more than once in a
transaction, only the last such modification is retained.

Classes _TransactionsIterator, _RecordsIterator, _Record are private
helper classes to support the iterator interface.

This is the compatibility module to make StructuredTextNG "look like" Classic Structured Text.  I was stupid earlier before and used "ClassicDocumentClass" to represent a Document.  This was failing in some cases when attemtping to use a "ClassicDocument" as input to a StructuredTextNG HTML renderer.  To fix: Use DocumentClass instead of ClassicDocumentClass.

- Optimized slightly.

- Clarified by using more variable names.

Add another optimization stage to detect longer sequences in the bytecode.

New opcode:  rawtextBeginScope.  This is equivalent to:

[endScope] rawtextColumn setPosition beginScope

These two sequences are very common in some of the slowest test documents
I've been using.

Lots more micro-optimization of the same sorts as we've been doing.

Updated to reflect the re-ordering of scope boundaries and text chunks.

New optimization of the instruction stream:  beginScope/endScope
instructions are always pushed in front of text chunks; the affect of
this is that many text chunks previously separated by scope boundaries
are now joined together, causing fewer instructions to be generated.

Do not call sys.exc_info() to initialize some locals we don't use.

SafeMapping.has_get():
    Simplify in order to reduce the number of Python bytecodes.  There
    is no longer a branch, which isn't strictly necessary.

Context.evaluateStructure(), .evaluateMacro():
    Since these methods are essentially synonyms for evaluate(), avoid
    the extra method lookup & call by simply making them aliases.

Context.evaluate():
    Re-arrange the try/except/except to avoid some of the exception
    catching by moving part of the try clause into an else clause.

When operating on raw strings which had DOS-style linefeeds (e.g. "\r\n"), StructuredText would neglect to strip the trailing garbage off the end of a paragraph.  Thus, the test for "examples" and other features that depended on the last characters in a paragraph to not be whitespace for proper operation were failing, causing problems mainly for people who use Windows to author STX content.  This is now fixed.