############################################################################## # Name: misc/scripts/update_doc_baseclasses.py # Purpose: Warns about missing classes in the "Derived from" # sections in the doc files # Created: 2007-07-28 # RCS-ID: $Id$ # Copyright: (c) 2007 Francesco Montorsi # Licence: wxWindows licence ############################################################################## from update_doc_utils import scanTexFiles # classes whose docs cannot be fixed automatically # because of: # 1) multiple inheritance # 2) other strange things specific of these classes EXCEPTIONS=['wxNotebook','wxChoicebook','wxListbook','wxToolbook','wxTreebook','wxURLDataObject','wxHScrolledWindow','wxVScrolledWindow','wxVarHVScrollHelper','wxHVScrolledWindow','wxFileStream'] def myCallback(classname, texFileName, content, i): # now search the base classes for this class baseclasses = [] for j in range(i,len(content)): line = content[j] if line.startswith("\wxheading{Derived from}"): # take all lines contained between this \wxheading and the next one # as base classes for k in range(j+1,len(content)): line = content[k] if "\wxheading" in line: break elif "\helpref" in line: baseclasses.append(line) break # we've already processed the 'derived from' section for this class if baseclasses == []: print " no base classes declared for class %s" % classname return True # keep going on with next \class tags # polish baseclasses list for i in range(len(baseclasses)): s = baseclasses[i] baseclasses[i] = s[s.find("{")+1:s.find("}")] # store the baseclasses tree[classname] = baseclasses treetex[classname] = texFileName treepos[classname] = j+1 print " class '%s' has %d base class(es): %s" % (classname, len(baseclasses), ','.join(baseclasses)) return True tree = dict() treetex = dict() treepos = dict() count = scanTexFiles(myCallback) print "\nProcessed %d files." % count print "\nNow starting to compare the base class lists.\n" def getFullListOfBaseClasses(classname): if classname not in tree or classname=='': return [] # only use the first base class of info taken from .tex files # i.e. we assume that at least the first class declared as base class is always correct baseclass = tree[classname][0] return [baseclass] + getFullListOfBaseClasses(baseclass) # now compare the theorical list of base classes with the list of base # classes taken from the .tex files fixed=0 tofix=set() for classname in tree: theorical=getFullListOfBaseClasses(classname) real=tree[classname] # compare them if real!=theorical: print "* for class '%s' documented in '%s'" % (classname,treetex[classname]) print " theorical list: %s" % theorical print " declared list: %s" % real if classname in EXCEPTIONS: tofix.add(treetex[classname]) print " cannot fix automatically (blacklisted class!)\n" continue # fix it! file = open(treetex[classname], "r") content = file.readlines() #print "old content is:" #print ''.join(content) # remove old \helpref lines startidx = treepos[classname] count = 0 while count < len(tree[classname]): # we want to remove n \helpref lines, where 'n' is the # number of base classes declared if content[startidx].startswith('\helpref'): del content[startidx] count = count+1 else: startidx = startidx+1 # probably an empty line # insert new ones if len(theorical)>1: for j in range(len(theorical)-1): c = theorical[j] content.insert(startidx+j, "\helpref{%s}{%s}\\\\\n" % (c, c.lower())) else: j=-1 c = theorical[j+1] content.insert(startidx+j+1, "\helpref{%s}{%s}\n" % (c, c.lower())) #print "new content is:" #print ''.join(content) # save the file file = open(treetex[classname], "w") file.write(''.join(content)) file.flush() print " fixed the .tex file\n" fixed=fixed+1 print "Total number of errors reported: %d" % fixed print "There are %d files to fix manually:\n%s" % (len(tofix), '\n'.join(tofix))