import yaptu
import re, os, sys, copy, errno, traceback

from StringIO import StringIO
import matplotlib
from hthelpers import modules

rex=re.compile('@([^@]+)@')
rbe=re.compile('\s*\+')
ren=re.compile('\s*-')
rco=re.compile('\s*= ')


def exception_to_str(s = None):
   sh = StringIO()
   if s is not None: print >>sh, s
   traceback.print_exc(file=sh)
   return sh.getvalue()

class NewsBox:
    def __init__(self, body, title='News Flash', bgcolor='red'):
        self.body = body
        self.title = title
        self.bgcolor = bgcolor
        
    def format_header(self):
        return """
    <tr><td  bgcolor="%s" align="left">
        <font class="tableheading">
                <b>%s</b>
        </font>
    </td></tr>
    """%(self.bgcolor, self.title)

    def __repr__(self):
        s =  '<table width=100% border=1 cellpadding=1 ' +\
               'cellspacing=1>\n'
        s += self.format_header()
        s += '<tr><td valign="top" bgcolor=#efefef>\n'
        s += '<font color="black">%s</font>'%self.body
        s += '</td></tr>\n'
        s += '</table>\n'
        return s

class LinkBox:
    def __init__(self, header, links):
        self.header = header
        self.links = links

    def format_header(self):
        return """
    <tr><td  bgcolor=#bfbfbf align="left">
        <font class="tableheading">
                <b>%s</b>
        </font>
    </td></tr>
    """ % self.header
            

    def __repr__(self):
        s =  '<table width=100% border=1 cellpadding=1 ' +\
               'cellspacing=1>\n'
        s += self.format_header()
        s += '  <tr><td valign="top" bgcolor=#efefef>\n'
        for text, link in self.links:
            s += '    <a href=%s>%s</a><br>\n' % (link, text)
            
        s += '</td></tr>\n'
        s += '</table>\n'
        return s

class FormatGoals:
    """Reads simple text file of goals and formats as html table"""
    
    def __init__(self, filename):
        self.table = []
        f = open(filename)
        self.lines = f.readlines()
        self.nline = 0
        self.format_table()
    def nextline(self):
        self.nline += 1
    def getline(self):
        if self.nline >= len(self.lines):
            return None
        else:
            return self.lines[self.nline].strip()
            
    def format_table(self):
        while 1:
            line = self.getline()
            if line is None: # all done!
                return
            if isBlank(line) or isComment(line):  # comment or empty, ignore
                pass
            elif onlyContains(line, '='): # Toplevel table heading section
                self.nextline()
                self.table.append("<tr><td colspan=3 bgcolor=#c0c0c0><b>")
                self.table.append(self.getline())
                self.table.append("</b></td></tr>")
            elif onlyContains(line, '+'): # Second level table heading section
                self.nextline()
                self.table.append("<tr><td colspan=3 bgcolor=#dddddd><b>")
                self.table.append(self.getline())
                self.table.append("</b></td></tr>")
                
            elif onlyContains(line, '*'): # Regular table entry
                # suck up any intervening blank lines
                self.nextline()
                while isBlank(self.getline()):
                    self.nextline()
                self.parse_entry()
                continue
            else:
                print "WARNING: goals text file is malformed at or around line", \
                    self.nline
            self.nextline()                      
                                  
    def parse_entry(self):
        """Deal with the different components of a normal goal row"""
        ncol = 0
        cols = [[], [], []]
        while 1:
            line = self.getline()
            if line is None or isNewEntry(line):   # End of file or new entry
                self.addrow(cols)                  # only way out of this loop
                return
            if isComment(line):
                pass
            elif not isBlank(line):
                cols[ncol].append(line)
                self.nextline()
            else: # suck up any following blank lines
                while isBlank(line):
                    self.nextline()
                    line = self.getline()
                    if line is None:
                        break
                ncol += 1
                if ncol > 2:
                    ncol = 2
               
    def addrow(self, cols):
        """Format the entry for a row"""
        ncol = 0
        self.table.append("<tr>")
        for col in cols:
            self.table.append("<td>")
            for line in col:
                self.table.append(line)
            if not len(col):
                self.table.append("&nbsp;")
                #self.table.append("<font color=#ffffff>.</font>") # need something in cell to format well
            self.table.append("</td>")
            ncol += 1
        self.table.append("</tr>")      
                    
    def __repr__(self):
        return "\n".join(self.table)

# helper functions for FormatGoals
def onlyContains(line, char):
    """Does line only contain one or more instances of given character?
    
    (aside from leading or trailing whitespace)"""

    tline = line.strip()
    if len(tline) and len(tline)*char == tline:
        return 1
    else:
        return 0

def isComment(line):
    return line.strip() and (line.strip()[0] == '#')
    
def isBlank(line):  
    return not line.strip()
    
            
def isNewEntry(line):
    if (onlyContains(line, '=') or 
        onlyContains(line, '+') or 
        onlyContains(line, '*')):
        return 1
    else:
        return 0

# re-add news to the params dict 'tables' entry below for news

news1= NewsBox("""We have recetly merged significant changes in the internal matplotlib transformation infrastructure into the main matplotlib codebase.  While this will not affect typical pylab users, some users of the API and transformation code will  to make some code changes to work with the new transformations.  Please see <a href=MIGRATION.txt>migration</a> and <a href=API_CHANGES>API_CHANGES</a> for a summary of the migration path and changes in the API, and let us know on the <a href=http://sourceforge.net/mail/?group_id=80706>mailing lists</a> if you are experiecing problems""", title='Major changes in matplotlib svn')


news2 = NewsBox("""Help support matplotlib development by <a href=http://sourceforge.net/project/project_donations.php?group_id=80706>donating</a> to fund developer sprints and other matplotlib development costs.""", title='Donate')


news3 = NewsBox("""matplotlib <a
href=http://sourceforge.net/project/platformdownload.php?group_id=80706>0.91.2</a>
is out.  See <a href=whats_new.html>what's new</a> for a summary of new features.  """, title='New release')


table1 =  LinkBox(header='Matplotlib', links=(
    ('Home', 'http://matplotlib.sourceforge.net'),
    ("What's&nbsp;New", 'whats_new.html'),
    ('Download', 'http://sourceforge.net/projects/matplotlib'),
    ('Installing', 'installing.html'),
    ('Screenshots', 'screenshots.html'),
    ('Examples&nbsp;(zip)', 'matplotlib_examples_%s.zip'%matplotlib.__version__),        
    ('Mailing lists', 'http://sourceforge.net/mail/?group_id=80706'),
    ('Donate', 'http://sourceforge.net/project/project_donations.php?group_id=80706'),        
    ))

table2 =  LinkBox(header='Documentation', links=(
    ('Tutorial', 'tutorial.html'),
    ("User's&nbsp;Guide&nbsp;(pdf)", 'users_guide_%s.pdf'%matplotlib.__version__),
    ("API&nbsp;(pdf)", 'api.pdf'),
    ('FAQ', 'faq.html'),
    ('Cookbook&nbsp;/&nbsp;wiki', 'http://www.scipy.org/wikis/topical_software/MatplotlibCookbook'),        
    ('pylab&nbsp;interface', 'pylab_commands.html'),
    ('Class&nbsp;library', 'classdocs.html'),
    ('Backends', 'backends.html'),
    ('Toolkits', 'toolkits.html'),
    ('Fonts', 'fonts.html'),
    ('Interactive', 'interactive.html'),
    ('Goals', 'goals.html'),
    ))

table3 =  LinkBox(header='Other', links=(
    ('Credits', 'credits.html'),
    ('License', 'license.html'),
    ))


params = {
    'myemail' : '<a href=mailto:jdhunter@ace.bsd.uchicago.edu> (jdhunter@ace.bsd.uchicago.edu)</a>',
    'tables' : (news1, news2, news3, table1, table2, table3),
    'default_table' :  'border=1 cellpadding=3 cellspacing=2', 
          }

headerBuffer = StringIO()
cop = yaptu.copier(rex, params, rbe, ren, rco, ouf=headerBuffer)
lines = file('header.html.template').readlines()
cop.copy(lines)
params['header'] = headerBuffer.getvalue()

footerBuffer = StringIO()
cop = yaptu.copier(rex, params, rbe, ren, rco, ouf=footerBuffer)
lines = file('footer.html.template').readlines()
cop.copy(lines)
params['footer'] = footerBuffer.getvalue()

docs = [modname + '.html.template' for modname in modules]
         

files = [
    'backends.html.template',
    'classdocs.html.template',
    'toolkits.html.template',
    'credits.html.template',
    'faq.html.template',
    'fonts.html.template',
    'goals.html.template',
    'index.html.template',
    'installing.html.template',
    'interactive.html.template',
    'license.html.template',
    'pylab_commands.html.template', 
    'screenshots.html.template',
    'tutorial.html.template',
    'whats_new.html.template', 
         ]
files.extend(docs)
         
         

#print params

keysOrig = {}
for key in locals().keys():
    keysOrig[key] = 1

for inFile in files:
    print '\tConverting', inFile
    if not os.path.exists(inFile):
       print 'Could not find', inFile
       continue
    fh = file(inFile, 'r')
    s = ''

    while 1:
        line = fh.readline()
        if line.find('@header@')==0:
            break
        s += line
    fileParams = copy.copy(params)
    if len(s)>0: exec(s)
    for key, val in locals().items():
        if not keysOrig.has_key(key):
            fileParams[key] = val

    outFile, ext = os.path.splitext(inFile)
    try:
        cop = yaptu.copier(rex, fileParams, rbe, ren, rco, ouf=file(outFile, 'w'))
        lines = ['@header@']
        lines.extend(fh.readlines())
        cop.copy(lines)
    except:
        print exception_to_str('Failed converting: %s'%inFile)
        continue

        
