(minidom in particular); it was using PyDOM which is now obsolete.
Only write the output file on success -- this avoids updating the timestamp
on the file on failure, which confuses "make".
implements a SAX XMLReader interface instead of the old Builder interface
used with PyDOM (now obsolete).
This only depends on the standard library, not PyXML.
join_adjacent_elements(): Hack to merge adjacent instances of
<option>; the source \programopt with GNU-style long options
created problems with LaTeX2HTML; this removes the evil
workaround, which should never be necessary from structured
documents(!).
Do adopt DocBook's <command>, <option>, <envar> names for \program,
\programopt, \envvar.
Add \filenq --> <filename quote="no"> (it has to go somewhere, at
least for now).
(must be named in the conversion spec.).
TableParser.start_entityref(): New method. Use to support creating
entity references via the conversion specification.
<description> elements. Add another sanity
check to make sure a special case only becomes
involved for <opcodedesc> and <opcodedescni>
elements.
fixup_paras_helper(): Clean up control flow.
Everywhere that checks node.nodeType==ELEMENT and node.tagName now
just check node.nodeName (nodeName is guaranteed for all node types
and will be special values that don't conflict with GIs for
non-ELEMENT nodes). Simplifies a number of tests and improves
readability in a lot of places.
engine in latex2esis.py can do, so discard it.
paragraph detection & markup: Handle some of the list-like
environments properly, picking up paragraphs inside
list items. (Only changed a few element-sets used for
recognition.)
fixup_refmodindexes_chunk(): Use the right attribute to determine the
module a <ref*modindex/> element is referring to.
convert(): Don't call remap_element_names(); it was removed.
same as sys.stderr.write(), but bwrite() will make it
bold if stderr is a tty (for visibility).
Use these everywhere instead of accessing
sys.stderr.write() directly.
find_all_child_elements(): Find all elements by type name that are
immediate children of another.
cleanup_synopses(): Receive both the document and fragment objects as
parameters; if only the document is passed, synopsis
information is never found since it's in the fragment!
build_para(): Add a chunk of code to ensure that a newline always
follows the </para> end-tag. Not really important, but
makes presentation consistent for authoring/
maintenance.
fixup_bifuncindexes_chunk(): Use find_all_child_elements() instead of
find_all_elements() so we don't work with too many
levels at once; that could cause DOM exceptions with
the <elem>.removeChild() method.
Removed some obsolete commented-out code.
Merged BaseConversion into Conversion since there's only one
conversion process now.
Remove push/pop of output stream; it's no longer needed.
Separate the Conversion class into a base and a subclass; the subclass
is pretty minimal but the separation is useful for....
NewConversion: New class that implements a somewhat different
approach to the conversion. This uses a table of
instances (rather than tuples) that have more
information than the tuples used for the older
conversion procedure. This allows a lot more control
over the conversion, and it seems to be pretty
stable.
TableEntry,
Parameter: New classes that are used to build the conversion
specification used by NewConversion.
TableParser: xmllib.XMLParser subclass that builds a conversion
specification from an XML document.
load_table(): Convenience function that loads a table from a file.
main(): Added flags --new and --old; these select which conversion is
used. The default is --new.
Several fixes have been made in the old conversion as well; these were
done before writing & switching to the new conversion, and should be
archived.
The next checkin of this file will discard the old conversion; is is
kept in this checkin to allow it to be retrieved if needed, and to
avoid lossing the bugfixes that have been made to it in the interim.
Use a temporary file to pass results from latex2esis.py to
docfixer.py; this makes for better error detection by make (if the
first pipeline stage fails, make wasn't catching it).
Fred Drake
Guido van Rossum