diff options
Diffstat (limited to 'Doc/tools/sgmlconv/docfixer.py')
-rwxr-xr-x | Doc/tools/sgmlconv/docfixer.py | 81 |
1 files changed, 36 insertions, 45 deletions
diff --git a/Doc/tools/sgmlconv/docfixer.py b/Doc/tools/sgmlconv/docfixer.py index 43959e2..a56ac6e 100755 --- a/Doc/tools/sgmlconv/docfixer.py +++ b/Doc/tools/sgmlconv/docfixer.py @@ -74,7 +74,7 @@ xml.dom.core.Document.get_childNodes = get_childNodes def get_first_element(doc, gi): for n in doc.childNodes: - if n.nodeType == ELEMENT and n.tagName == gi: + if n.nodeName == gi: return n def extract_first_element(doc, gi): @@ -86,7 +86,7 @@ def extract_first_element(doc, gi): def find_all_elements(doc, gi): nodes = [] - if doc.nodeType == ELEMENT and doc.tagName == gi: + if doc.nodeName == gi: nodes.append(doc) for child in doc.childNodes: if child.nodeType == ELEMENT: @@ -99,16 +99,15 @@ def find_all_elements(doc, gi): def find_all_child_elements(doc, gi): nodes = [] for child in doc.childNodes: - if child.nodeType == ELEMENT: - if child.tagName == gi: - nodes.append(child) + if child.nodeName == gi: + nodes.append(child) return nodes def find_all_elements_from_set(doc, gi_set): return __find_all_elements_from_set(doc, gi_set, []) def __find_all_elements_from_set(doc, gi_set, nodes): - if doc.nodeType == ELEMENT and doc.tagName in gi_set: + if doc.nodeName in gi_set: nodes.append(doc) for child in doc.childNodes: if child.nodeType == ELEMENT: @@ -156,7 +155,7 @@ def cleanup_root_text(doc): skip = 0 if n.nodeType == TEXT and not prevskip: discards.append(n) - elif n.nodeType == ELEMENT and n.tagName == "COMMENT": + elif n.nodeName == "COMMENT": skip = 1 for node in discards: doc.removeChild(node) @@ -191,7 +190,8 @@ def rewrite_descriptor(doc, descriptor): # Do these things: # 1. Add an "index='no'" attribute to the element if the tagName # ends in 'ni', removing the 'ni' from the name. - # 2. Create a <signature> from the name attribute and <args>. + # 2. Create a <signature> from the name attribute + # 2a.Create an <args> if it appears to be available. # 3. Create additional <signature>s from <*line{,ni}> elements, # if found. # 4. If a <versionadded> is found, move it to an attribute on the @@ -199,6 +199,7 @@ def rewrite_descriptor(doc, descriptor): # 5. Move remaining child nodes to a <description> element. # 6. Put it back together. # + # 1. descname = descriptor.tagName index = 1 if descname[-2:] == "ni": @@ -217,7 +218,11 @@ def rewrite_descriptor(doc, descriptor): signature.appendChild(name) name.appendChild(doc.createTextNode(descriptor.getAttribute("name"))) descriptor.removeAttribute("name") + # 2a. if descriptor.attributes.has_key("var"): + if descname != "opcodedesc": + raise RuntimeError, \ + "got 'var' attribute on descriptor other than opcodedesc" variable = descriptor.getAttribute("var") if variable: args = doc.createElement("args") @@ -227,10 +232,12 @@ def rewrite_descriptor(doc, descriptor): descriptor.removeAttribute("var") newchildren = [signature] children = descriptor.childNodes - pos = skip_leading_nodes(children, 0) + pos = skip_leading_nodes(children) if pos < len(children): child = children[pos] - if child.nodeType == ELEMENT and child.tagName == "args": + if child.nodeName == "args": +## bwrite("found <args> in descriptor, moving to <signature>\n") +## ewrite(descriptor.toxml() + "\n---\n") # create an <args> in <signature>: args = doc.createElement("args") argchildren = [] @@ -242,10 +249,9 @@ def rewrite_descriptor(doc, descriptor): signature.appendChild(args) signature.appendChild(doc.createTextNode("\n ")) # 3, 4. - pos = skip_leading_nodes(children, pos + 1) + pos = skip_leading_nodes(children, pos) while pos < len(children) \ - and children[pos].nodeType == ELEMENT \ - and children[pos].tagName in (linename, "versionadded"): + and children[pos].nodeName in (linename, "versionadded"): if children[pos].tagName == linename: # this is really a supplemental signature, create <signature> sig = methodline_to_signature(doc, children[pos]) @@ -351,8 +357,7 @@ def fixup_trailing_whitespace(doc, wsmap): while queue: node = queue[0] del queue[0] - if node.nodeType == ELEMENT \ - and wsmap.has_key(node.tagName): + if wsmap.has_key(node.nodeName): ws = wsmap[node.tagName] children = node.childNodes children.reverse() @@ -467,8 +472,7 @@ def create_module_info(doc, section): if title: children = title.childNodes if len(children) >= 2 \ - and children[0].nodeType == ELEMENT \ - and children[0].tagName == "module" \ + and children[0].nodeName == "module" \ and children[0].childNodes[0].data == name: # this is it; morph the <title> into <short-synopsis> first_data = children[1] @@ -512,8 +516,7 @@ def create_module_info(doc, section): children = section.childNodes for i in range(len(children)): node = children[i] - if node.nodeType == ELEMENT \ - and node.tagName == "moduleinfo": + if node.nodeName == "moduleinfo": nextnode = children[i+1] if nextnode.nodeType == TEXT: data = nextnode.data @@ -595,7 +598,7 @@ def fixup_row(doc, row): def move_elements_by_name(doc, source, dest, name, sep=None): nodes = [] for child in source.childNodes: - if child.nodeType == ELEMENT and child.tagName == name: + if child.nodeName == name: nodes.append(child) for node in nodes: source.removeChild(node) @@ -635,9 +638,7 @@ PARA_LEVEL_PRECEEDERS = ( def fixup_paras(doc, fragment): for child in fragment.childNodes: - if child.nodeType == ELEMENT \ - and child.tagName in RECURSE_INTO_PARA_CONTAINERS: - # + if child.nodeName in RECURSE_INTO_PARA_CONTAINERS: fixup_paras_helper(doc, child) descriptions = find_all_elements(fragment, "description") for description in descriptions: @@ -647,26 +648,17 @@ def fixup_paras(doc, fragment): def fixup_paras_helper(doc, container, depth=0): # document is already normalized children = container.childNodes - start = 0 + start = skip_leading_nodes(children) while len(children) > start: - start = skip_leading_nodes(children, start) - if start >= len(children): - break - # - # Either paragraph material or something to recurse into: - # - if (children[start].nodeType == ELEMENT) \ - and (children[start].tagName in RECURSE_INTO_PARA_CONTAINERS): + if children[start].nodeName in RECURSE_INTO_PARA_CONTAINERS: + # Something to recurse into: fixup_paras_helper(doc, children[start]) - start = skip_leading_nodes(children, start + 1) - continue - # - # paragraph material: - # - build_para(doc, container, start, len(children)) - if DEBUG_PARA_FIXER and depth == 10: - sys.exit(1) - start = start + 1 + else: + # Paragraph material: + build_para(doc, container, start, len(children)) + if DEBUG_PARA_FIXER and depth == 10: + sys.exit(1) + start = skip_leading_nodes(children, start + 1) def build_para(doc, parent, start, i): @@ -731,7 +723,7 @@ def build_para(doc, parent, start, i): return start + 1 -def skip_leading_nodes(children, start): +def skip_leading_nodes(children, start=0): """Return index into children of a node at which paragraph building should begin or a recursive call to fixup_paras_helper() should be made (for subsections, etc.). @@ -785,8 +777,7 @@ def fixup_signatures(doc, fragment): def fixup_args(doc, arglist): for child in arglist.childNodes: - if child.nodeType == ELEMENT \ - and child.tagName == "optional": + if child.nodeName == "optional": # found it; fix and return arglist.insertBefore(doc.createTextNode("["), child) optkids = child.childNodes @@ -809,7 +800,7 @@ def fixup_sectionauthors(doc, fragment): sectauth.removeAttribute("name") after = section.childNodes[2] title = section.childNodes[1] - if title.nodeType == ELEMENT and title.tagName != "title": + if title.nodeName != "title": after = section.childNodes[0] section.insertBefore(doc.createTextNode("\n "), after) section.insertBefore(sectauth, after) |