676 lines
21 KiB
Python
676 lines
21 KiB
Python
# -*- mode: python -*-
|
|
# -*- coding: utf-8 -*-
|
|
|
|
##############################################################################
|
|
#
|
|
# Gestion scolarite IUT
|
|
#
|
|
# Copyright (c) 1999 - 2021 Emmanuel Viennet. All rights reserved.
|
|
#
|
|
# This program is free software; you can redistribute it and/or modify
|
|
# it under the terms of the GNU General Public License as published by
|
|
# the Free Software Foundation; either version 2 of the License, or
|
|
# (at your option) any later version.
|
|
#
|
|
# This program is distributed in the hope that it will be useful,
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
# GNU General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU General Public License
|
|
# along with this program; if not, write to the Free Software
|
|
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
#
|
|
# Emmanuel Viennet emmanuel.viennet@viennet.net
|
|
#
|
|
##############################################################################
|
|
|
|
|
|
""" Excel file handling
|
|
"""
|
|
|
|
from pyExcelerator import *
|
|
|
|
from notes_log import log
|
|
from scolog import logdb
|
|
|
|
from sco_exceptions import ScoValueError
|
|
import sco_utils as scu
|
|
import notesdb
|
|
|
|
import time, datetime
|
|
from types import StringType, IntType, FloatType, LongType
|
|
|
|
# colors, voir exemple format.py
|
|
COLOR_CODES = {
|
|
"black": 0,
|
|
"red": 0x0A,
|
|
"mauve": 0x19,
|
|
"marron": 0x3C,
|
|
"blue": 0x4,
|
|
"orange": 0x34,
|
|
"lightyellow": 0x2B,
|
|
}
|
|
|
|
|
|
def sendExcelFile(REQUEST, data, filename):
|
|
"""publication fichier.
|
|
(on ne doit rien avoir émis avant, car ici sont générés les entetes)
|
|
"""
|
|
filename = (
|
|
scu.unescape_html(scu.suppress_accents(filename))
|
|
.replace("&", "")
|
|
.replace(" ", "_")
|
|
)
|
|
REQUEST.RESPONSE.setHeader("content-type", scu.XLS_MIMETYPE)
|
|
REQUEST.RESPONSE.setHeader(
|
|
"content-disposition", 'attachment; filename="%s"' % filename
|
|
)
|
|
return data
|
|
|
|
|
|
## (stolen from xlrd)
|
|
# Convert an Excel number (presumed to represent a date, a datetime or a time) into
|
|
# a Python datetime.datetime
|
|
# @param xldate The Excel number
|
|
# @param datemode 0: 1900-based, 1: 1904-based.
|
|
# @return a datetime.datetime object, to the nearest_second.
|
|
# <br>Special case: if 0.0 <= xldate < 1.0, it is assumed to represent a time;
|
|
# a datetime.time object will be returned.
|
|
# <br>Note: 1904-01-01 is not regarded as a valid date in the datemode 1 system; its "serial number"
|
|
# is zero.
|
|
|
|
_XLDAYS_TOO_LARGE = (2958466, 2958466 - 1462) # This is equivalent to 10000-01-01
|
|
|
|
|
|
def xldate_as_datetime(xldate, datemode=0):
|
|
if datemode not in (0, 1):
|
|
raise ValueError("invalid mode %s" % datemode)
|
|
if xldate == 0.00:
|
|
return datetime.time(0, 0, 0)
|
|
if xldate < 0.00:
|
|
raise ValueError("invalid date code %s" % xldate)
|
|
xldays = int(xldate)
|
|
frac = xldate - xldays
|
|
seconds = int(round(frac * 86400.0))
|
|
assert 0 <= seconds <= 86400
|
|
if seconds == 86400:
|
|
seconds = 0
|
|
xldays += 1
|
|
if xldays >= _XLDAYS_TOO_LARGE[datemode]:
|
|
raise ValueError("date too large %s" % xldate)
|
|
|
|
if xldays == 0:
|
|
# second = seconds % 60; minutes = seconds // 60
|
|
minutes, second = divmod(seconds, 60)
|
|
# minute = minutes % 60; hour = minutes // 60
|
|
hour, minute = divmod(minutes, 60)
|
|
return datetime.time(hour, minute, second)
|
|
|
|
if xldays < 61 and datemode == 0:
|
|
raise ValueError("ambiguous date %s" % xldate)
|
|
|
|
return datetime.datetime.fromordinal(
|
|
xldays + 693594 + 1462 * datemode
|
|
) + datetime.timedelta(seconds=seconds)
|
|
|
|
|
|
# Sous-classes pour ajouter methode savetostr()
|
|
# (generation de fichiers en memoire)
|
|
# XXX ne marche pas car accès a methodes privees (__xxx)
|
|
# -> on utilise version modifiee par nous meme de pyExcelerator
|
|
#
|
|
# class XlsDocWithSave(CompoundDoc.XlsDoc):
|
|
# def savetostr(self, stream):
|
|
# #added by Emmanuel: save method, but returns a string
|
|
# # 1. Align stream on 0x1000 boundary (and therefore on sector boundary)
|
|
# padding = '\x00' * (0x1000 - (len(stream) % 0x1000))
|
|
# self.book_stream_len = len(stream) + len(padding)
|
|
|
|
# self.__build_directory()
|
|
# self.__build_sat()
|
|
# self.__build_header()
|
|
|
|
# return self.header+self.packed_MSAT_1st+stream+padding+self.packed_MSAT_2nd+self.packed_SAT+self.dir_stream
|
|
|
|
# class WorkbookWithSave(Workbook):
|
|
# def savetostr(self):
|
|
# doc = XlsDocWithSave()
|
|
# return doc.savetostr(self.get_biff_data())
|
|
|
|
|
|
def Excel_MakeStyle(
|
|
bold=False, italic=False, color="black", bgcolor=None, halign=None, valign=None
|
|
):
|
|
style = XFStyle()
|
|
font = Font()
|
|
if bold:
|
|
font.bold = bold
|
|
if italic:
|
|
font.italic = italic
|
|
font.name = "Arial"
|
|
colour_index = COLOR_CODES.get(color, None)
|
|
if colour_index:
|
|
font.colour_index = colour_index
|
|
if bgcolor:
|
|
style.pattern = Pattern()
|
|
style.pattern.pattern = Pattern.SOLID_PATTERN
|
|
style.pattern.pattern_fore_colour = COLOR_CODES.get(bgcolor, None)
|
|
al = None
|
|
if halign:
|
|
al = Alignment()
|
|
al.horz = {
|
|
"left": Alignment.HORZ_LEFT,
|
|
"right": Alignment.HORZ_RIGHT,
|
|
"center": Alignment.HORZ_CENTER,
|
|
}[halign]
|
|
if valign:
|
|
if not al:
|
|
al = Alignment()
|
|
al.vert = {
|
|
"top": Alignment.VERT_TOP,
|
|
"bottom": VERT_BOTTOM,
|
|
"center": VERT_CENTER,
|
|
}[valign]
|
|
if al:
|
|
style.alignment = al
|
|
style.font = font
|
|
return style
|
|
|
|
|
|
class ScoExcelSheet:
|
|
def __init__(self, sheet_name="feuille", default_style=None):
|
|
self.sheet_name = sheet_name
|
|
self.cells = [] # list of list
|
|
self.cells_styles_lico = {} # { (li,co) : style }
|
|
self.cells_styles_li = {} # { li : style }
|
|
self.cells_styles_co = {} # { co : style }
|
|
if not default_style:
|
|
default_style = Excel_MakeStyle()
|
|
self.default_style = default_style
|
|
|
|
def set_style(self, style=None, li=None, co=None):
|
|
if li != None and co != None:
|
|
self.cells_styles_lico[(li, co)] = style
|
|
elif li != None:
|
|
self.cells_styles_li[li] = style
|
|
elif co != None:
|
|
self.cells_styles_co[co] = style
|
|
|
|
def append(self, l):
|
|
"""Append a line of cells"""
|
|
self.cells.append(l)
|
|
|
|
def get_cell_style(self, li, co):
|
|
"""Get style for specified cell"""
|
|
return (
|
|
self.cells_styles_lico.get((li, co), None)
|
|
or self.cells_styles_li.get(li, None)
|
|
or self.cells_styles_co.get(co, None)
|
|
or self.default_style
|
|
)
|
|
|
|
def gen_workbook(self, wb=None):
|
|
"""Generates and returns a workbook from stored data.
|
|
If wb, add a sheet (tab) to the existing workbook (in this case, returns None).
|
|
"""
|
|
if wb == None:
|
|
wb = Workbook() # Création du fichier
|
|
sauvegarde = True
|
|
else:
|
|
sauvegarde = False
|
|
ws0 = wb.add_sheet(self.sheet_name.decode(scu.SCO_ENCODING))
|
|
li = 0
|
|
for l in self.cells:
|
|
co = 0
|
|
for c in l:
|
|
# safety net: allow only str, int and float
|
|
if type(c) == LongType:
|
|
c = int(c) # assume all ScoDoc longs fits in int !
|
|
elif type(c) not in (IntType, FloatType):
|
|
c = str(c).decode(scu.SCO_ENCODING)
|
|
|
|
ws0.write(li, co, c, self.get_cell_style(li, co))
|
|
co += 1
|
|
li += 1
|
|
if sauvegarde == True:
|
|
return wb.savetostr()
|
|
else:
|
|
return None
|
|
|
|
|
|
def Excel_SimpleTable(titles=[], lines=[[]], SheetName="feuille", titlesStyles=[]):
|
|
"""Export simple type 'CSV': 1ere ligne en gras, le reste tel quel"""
|
|
# XXX devrait maintenant utiliser ScoExcelSheet
|
|
wb = Workbook()
|
|
ws0 = wb.add_sheet(SheetName.decode(scu.SCO_ENCODING))
|
|
if not titlesStyles:
|
|
style = Excel_MakeStyle(bold=True)
|
|
titlesStyles = [style] * len(titles)
|
|
# ligne de titres
|
|
col = 0
|
|
for it in titles:
|
|
ws0.write(0, col, it.decode(scu.SCO_ENCODING), titlesStyles[col])
|
|
col += 1
|
|
# suite
|
|
default_style = Excel_MakeStyle()
|
|
text_style = Excel_MakeStyle()
|
|
text_style.num_format_str = "@"
|
|
li = 1
|
|
for l in lines:
|
|
col = 0
|
|
for it in l:
|
|
cell_style = default_style
|
|
# safety net: allow only str, int and float
|
|
if type(it) == LongType:
|
|
it = int(it) # assume all ScoDoc longs fits in int !
|
|
elif type(it) not in (IntType, FloatType):
|
|
it = str(it).decode(scu.SCO_ENCODING)
|
|
cell_style = text_style
|
|
ws0.write(li, col, it, cell_style)
|
|
col += 1
|
|
li += 1
|
|
#
|
|
return wb.savetostr()
|
|
|
|
|
|
def Excel_feuille_saisie(E, titreannee, description, lines):
|
|
"""Genere feuille excel pour saisie des notes.
|
|
E: evaluation (dict)
|
|
lines: liste de tuples
|
|
(etudid, nom, prenom, etat, groupe, val, explanation)
|
|
"""
|
|
SheetName = "Saisie notes"
|
|
wb = Workbook()
|
|
ws0 = wb.add_sheet(SheetName.decode(scu.SCO_ENCODING))
|
|
# ajuste largeurs colonnes (unite inconnue, empirique)
|
|
ws0.col(0).width = 400 # codes
|
|
ws0.col(1).width = 6000 # noms
|
|
ws0.col(2).width = 4000 # prenoms
|
|
ws0.col(3).width = 6000 # groupes
|
|
ws0.col(4).width = 3000 # notes
|
|
ws0.col(5).width = 13000 # remarques
|
|
# styles
|
|
style_titres = XFStyle()
|
|
font0 = Font()
|
|
font0.bold = True
|
|
font0.name = "Arial"
|
|
font0.bold = True
|
|
font0.height = 14 * 0x14
|
|
style_titres.font = font0
|
|
|
|
style_expl = XFStyle()
|
|
font_expl = Font()
|
|
font_expl.name = "Arial"
|
|
font_expl.italic = True
|
|
font0.height = 12 * 0x14
|
|
font_expl.colour_index = 0x0A # rouge, voir exemple format.py
|
|
style_expl.font = font_expl
|
|
|
|
topborders = Borders()
|
|
topborders.top = 1
|
|
topleftborders = Borders()
|
|
topleftborders.top = 1
|
|
topleftborders.left = 1
|
|
rightborder = Borders()
|
|
rightborder.right = 1
|
|
|
|
style_ro = XFStyle() # cells read-only
|
|
font_ro = Font()
|
|
font_ro.name = "Arial"
|
|
font_ro.colour_index = COLOR_CODES["mauve"]
|
|
style_ro.font = font_ro
|
|
style_ro.borders = rightborder
|
|
|
|
style_dem = XFStyle() # cells read-only
|
|
font_dem = Font()
|
|
font_dem.name = "Arial"
|
|
font_dem.colour_index = COLOR_CODES["marron"]
|
|
style_dem.font = font_dem
|
|
style_dem.borders = topborders
|
|
|
|
style = XFStyle()
|
|
font1 = Font()
|
|
font1.name = "Arial"
|
|
font1.height = 12 * 0x14
|
|
style.font = font1
|
|
|
|
style_nom = XFStyle() # style pour nom, prenom, groupe
|
|
style_nom.font = font1
|
|
style_nom.borders = topborders
|
|
|
|
style_notes = XFStyle()
|
|
font2 = Font()
|
|
font2.name = "Arial"
|
|
font2.bold = True
|
|
style_notes.font = font2
|
|
style_notes.num_format_str = "general"
|
|
style_notes.pattern = Pattern() # fond jaune
|
|
style_notes.pattern.pattern = Pattern.SOLID_PATTERN
|
|
style_notes.pattern.pattern_fore_colour = COLOR_CODES["lightyellow"]
|
|
style_notes.borders = topborders
|
|
|
|
style_comment = XFStyle()
|
|
font_comment = Font()
|
|
font_comment.name = "Arial"
|
|
font_comment.height = 9 * 0x14
|
|
font_comment.colour_index = COLOR_CODES["blue"]
|
|
style_comment.font = font_comment
|
|
style_comment.borders = topborders
|
|
|
|
# ligne de titres
|
|
li = 0
|
|
ws0.write(
|
|
li, 0, u"Feuille saisie note (à enregistrer au format excel)", style_titres
|
|
)
|
|
li += 1
|
|
ws0.write(li, 0, u"Saisir les notes dans la colonne E (cases jaunes)", style_expl)
|
|
li += 1
|
|
ws0.write(li, 0, u"Ne pas modifier les cases en mauve !", style_expl)
|
|
li += 1
|
|
# Nom du semestre
|
|
ws0.write(
|
|
li, 0, scu.unescape_html(titreannee).decode(scu.SCO_ENCODING), style_titres
|
|
)
|
|
li += 1
|
|
# description evaluation
|
|
ws0.write(
|
|
li, 0, scu.unescape_html(description).decode(scu.SCO_ENCODING), style_titres
|
|
)
|
|
li += 1
|
|
ws0.write(
|
|
li, 0, u"Evaluation du %s (coef. %g)" % (E["jour"], E["coefficient"]), style
|
|
)
|
|
li += 1
|
|
li += 1 # ligne blanche
|
|
# code et titres colonnes
|
|
ws0.write(li, 0, u"!%s" % E["evaluation_id"], style_ro)
|
|
ws0.write(li, 1, u"Nom", style_titres)
|
|
ws0.write(li, 2, u"Prénom", style_titres)
|
|
ws0.write(li, 3, u"Groupe", style_titres)
|
|
ws0.write(li, 4, u"Note sur %g" % E["note_max"], style_titres)
|
|
ws0.write(li, 5, u"Remarque", style_titres)
|
|
# etudiants
|
|
for line in lines:
|
|
li += 1
|
|
st = style_nom
|
|
ws0.write(li, 0, ("!" + line[0]).decode(scu.SCO_ENCODING), style_ro) # code
|
|
if line[3] != "I":
|
|
st = style_dem
|
|
if line[3] == "D": # demissionnaire
|
|
s = "DEM"
|
|
else:
|
|
s = line[3] # etat autre
|
|
else:
|
|
s = line[4] # groupes TD/TP/...
|
|
ws0.write(li, 1, line[1].decode(scu.SCO_ENCODING), st)
|
|
ws0.write(li, 2, line[2].decode(scu.SCO_ENCODING), st)
|
|
ws0.write(li, 3, s.decode(scu.SCO_ENCODING), st)
|
|
try:
|
|
val = float(line[5])
|
|
except:
|
|
val = line[5].decode(scu.SCO_ENCODING)
|
|
ws0.write(li, 4, val, style_notes) # note
|
|
ws0.write(li, 5, line[6].decode(scu.SCO_ENCODING), style_comment) # comment
|
|
# explication en bas
|
|
li += 2
|
|
ws0.write(li, 1, u"Code notes", style_titres)
|
|
ws0.write(li + 1, 1, u"ABS", style_expl)
|
|
ws0.write(li + 1, 2, u"absent (0)", style_expl)
|
|
ws0.write(li + 2, 1, u"EXC", style_expl)
|
|
ws0.write(li + 2, 2, u"pas prise en compte", style_expl)
|
|
ws0.write(li + 3, 1, u"ATT", style_expl)
|
|
ws0.write(li + 3, 2, u"en attente", style_expl)
|
|
ws0.write(li + 4, 1, u"SUPR", style_expl)
|
|
ws0.write(li + 4, 2, u"pour supprimer note déjà entrée", style_expl)
|
|
ws0.write(li + 5, 1, u"", style_expl)
|
|
ws0.write(li + 5, 2, u"cellule vide -> note non modifiée", style_expl)
|
|
return wb.savetostr()
|
|
|
|
|
|
def Excel_to_list(data, convert_to_string=str): # we may need 'encoding' argument ?
|
|
"""returns list of list
|
|
convert_to_string is a conversion function applied to all non-string values (ie numbers)
|
|
"""
|
|
try:
|
|
P = parse_xls("", scu.SCO_ENCODING, doc=data)
|
|
except:
|
|
log("Excel_to_list: failure to import document")
|
|
open("/tmp/last_scodoc_import_failure.xls", "w").write(data)
|
|
raise ScoValueError(
|
|
"Fichier illisible: assurez-vous qu'il s'agit bien d'un document Excel !"
|
|
)
|
|
|
|
diag = [] # liste de chaines pour former message d'erreur
|
|
# n'utilise que la première feuille
|
|
if len(P) < 1:
|
|
diag.append("Aucune feuille trouvée dans le classeur !")
|
|
return diag, None
|
|
if len(P) > 1:
|
|
diag.append("Attention: n'utilise que la première feuille du classeur !")
|
|
# fill matrix
|
|
sheet_name, values = P[0]
|
|
sheet_name = sheet_name.encode(scu.SCO_ENCODING, "backslashreplace")
|
|
if not values:
|
|
diag.append("Aucune valeur trouvée dans le classeur !")
|
|
return diag, None
|
|
indexes = values.keys()
|
|
# search numbers of rows and cols
|
|
rows = [x[0] for x in indexes]
|
|
cols = [x[1] for x in indexes]
|
|
nbcols = max(cols) + 1
|
|
nbrows = max(rows) + 1
|
|
M = []
|
|
for _ in range(nbrows):
|
|
M.append([""] * nbcols)
|
|
|
|
for row_idx, col_idx in indexes:
|
|
v = values[(row_idx, col_idx)]
|
|
if isinstance(v, unicode):
|
|
v = v.encode(scu.SCO_ENCODING, "backslashreplace")
|
|
elif convert_to_string:
|
|
v = convert_to_string(v)
|
|
M[row_idx][col_idx] = v
|
|
diag.append('Feuille "%s", %d lignes' % (sheet_name, len(M)))
|
|
# diag.append(str(M))
|
|
#
|
|
return diag, M
|
|
|
|
|
|
#
|
|
def Excel_feuille_listeappel(
|
|
context,
|
|
sem,
|
|
groupname,
|
|
lines,
|
|
partitions=[], # partitions a montrer (colonnes)
|
|
with_codes=False, # indique codes etuds
|
|
with_paiement=False, # indique si etudiant a paye inscription
|
|
server_name=None,
|
|
):
|
|
"generation feuille appel"
|
|
formsemestre_id = sem["formsemestre_id"]
|
|
SheetName = "Liste " + groupname
|
|
wb = Workbook()
|
|
ws0 = wb.add_sheet(SheetName.decode(scu.SCO_ENCODING))
|
|
|
|
font1 = Font()
|
|
font1.name = "Arial"
|
|
font1.height = 10 * 0x14
|
|
|
|
font1i = Font()
|
|
font1i.name = "Arial"
|
|
font1i.height = 10 * 0x14
|
|
font1i.italic = True
|
|
|
|
style1i = XFStyle()
|
|
style1i.font = font1i
|
|
|
|
style1b = XFStyle()
|
|
style1b.font = font1
|
|
borders = Borders()
|
|
borders.left = 1
|
|
borders.top = 1
|
|
borders.bottom = 1
|
|
style1b.borders = borders
|
|
|
|
style2 = XFStyle()
|
|
font2 = Font()
|
|
font2.name = "Arial"
|
|
font2.height = 14 * 0x14
|
|
style2.font = font2
|
|
|
|
style2b = XFStyle()
|
|
style2b.font = font1i
|
|
borders = Borders()
|
|
borders.left = 1
|
|
borders.top = 1
|
|
borders.bottom = 1
|
|
borders.right = 1
|
|
style2b.borders = borders
|
|
|
|
style2tb = XFStyle()
|
|
borders = Borders()
|
|
borders.top = 1
|
|
borders.bottom = 1
|
|
style2tb.borders = borders
|
|
style2tb.font = Font()
|
|
style2tb.font.height = 16 * 0x14 # -> ligne hautes
|
|
|
|
style2t3 = XFStyle()
|
|
borders = Borders()
|
|
borders.top = 1
|
|
borders.bottom = 1
|
|
borders.left = 1
|
|
style2t3.borders = borders
|
|
|
|
style2t3bold = XFStyle()
|
|
borders = Borders()
|
|
borders.top = 1
|
|
borders.bottom = 1
|
|
borders.left = 1
|
|
style2t3bold.borders = borders
|
|
fontb = Font()
|
|
fontb.bold = True
|
|
style2t3bold.font = fontb
|
|
|
|
style3 = XFStyle()
|
|
font3 = Font()
|
|
font3.name = "Arial"
|
|
font3.bold = True
|
|
font3.height = 14 * 0x14
|
|
style3.font = font3
|
|
|
|
NbWeeks = 4 # nombre de colonnes pour remplir absences
|
|
|
|
# ligne 1
|
|
li = 0
|
|
ws0.write(
|
|
li,
|
|
1,
|
|
(
|
|
"%s %s (%s - %s)"
|
|
% (
|
|
context.get_preference("DeptName", formsemestre_id),
|
|
notesdb.unquote(sem["titre_num"]),
|
|
sem["date_debut"],
|
|
sem["date_fin"],
|
|
)
|
|
).decode(scu.SCO_ENCODING),
|
|
style2,
|
|
)
|
|
# ligne 2
|
|
li += 1
|
|
ws0.write(li, 1, u"Discipline :", style2)
|
|
# ligne 3
|
|
li += 1
|
|
ws0.write(li, 1, u"Enseignant :", style2)
|
|
ws0.write(li, 5, ("Groupe %s" % groupname).decode(scu.SCO_ENCODING), style3)
|
|
# Avertissement pour ne pas confondre avec listes notes
|
|
ws0.write(
|
|
li + 1, 2, u"Ne pas utiliser cette feuille pour saisir les notes !", style1i
|
|
)
|
|
#
|
|
li += 2
|
|
li += 1
|
|
ws0.write(li, 1, u"Nom", style3)
|
|
co = 2
|
|
for partition in partitions:
|
|
if partition["partition_name"]:
|
|
ws0.write(
|
|
li, co, partition["partition_name"].decode(scu.SCO_ENCODING), style3
|
|
)
|
|
co += 1
|
|
if with_codes:
|
|
coc = co
|
|
ws0.write(li, coc, u"etudid", style3)
|
|
ws0.write(li, coc + 1, u"code_nip", style3)
|
|
ws0.write(li, coc + 2, u"code_ine", style3)
|
|
co += 3
|
|
|
|
for i in range(NbWeeks):
|
|
ws0.write(li, co + i, "", style2b)
|
|
n = 0
|
|
for t in lines:
|
|
n += 1
|
|
li += 1
|
|
ws0.write(li, 0, n, style1b)
|
|
nomprenom = (
|
|
t["civilite_str"]
|
|
+ " "
|
|
+ t["nom"]
|
|
+ " "
|
|
+ scu.strcapitalize(scu.strlower(t["prenom"]))
|
|
)
|
|
style_nom = style2t3
|
|
if with_paiement:
|
|
paie = t.get("paiementinscription", None)
|
|
if paie is None:
|
|
nomprenom += " (inscription ?)"
|
|
style_nom = style2t3bold
|
|
elif not paie:
|
|
nomprenom += " (non paiement)"
|
|
style_nom = style2t3bold
|
|
ws0.write(li, 1, nomprenom.decode(scu.SCO_ENCODING), style_nom)
|
|
co = 2
|
|
for partition in partitions:
|
|
if partition["partition_name"]:
|
|
ws0.write(
|
|
li,
|
|
co,
|
|
t.get(partition["partition_id"], "").decode(scu.SCO_ENCODING),
|
|
style2t3,
|
|
)
|
|
co += 1
|
|
if with_codes:
|
|
ws0.write(li, coc, t["etudid"].decode(scu.SCO_ENCODING), style2t3)
|
|
if t["code_nip"]:
|
|
code_nip = t["code_nip"].decode(scu.SCO_ENCODING)
|
|
else:
|
|
code_nip = u""
|
|
ws0.write(li, coc + 1, code_nip, style2t3)
|
|
if t["code_ine"]:
|
|
code_ine = t["code_ine"].decode(scu.SCO_ENCODING)
|
|
else:
|
|
code_ine = u""
|
|
ws0.write(li, coc + 2, code_ine, style2t3)
|
|
if t["etath"]:
|
|
etath = t["etath"].decode(scu.SCO_ENCODING)
|
|
else:
|
|
etath = u""
|
|
ws0.write(li, co, etath, style2b) # etat
|
|
for i in range(1, NbWeeks):
|
|
ws0.write(li, co + i, u"", style2b) # cellules vides
|
|
ws0.row(li).height = 850 # sans effet ?
|
|
#
|
|
li += 2
|
|
dt = time.strftime("%d/%m/%Y à %Hh%M")
|
|
if server_name:
|
|
dt += " sur " + server_name
|
|
ws0.write(li, 1, ("Liste éditée le " + dt).decode(scu.SCO_ENCODING), style1i)
|
|
#
|
|
ws0.col(0).width = 850
|
|
ws0.col(1).width = 9000
|
|
|
|
return wb.savetostr()
|