TestsScoDoc7API/sco_trombino.py

737 lines
25 KiB
Python
Raw Normal View History

2020-09-26 16:19:37 +02:00
# -*- mode: python -*-
# -*- coding: utf-8 -*-
##############################################################################
#
# Gestion scolarite IUT
#
2021-01-01 17:51:08 +01:00
# Copyright (c) 1999 - 2021 Emmanuel Viennet. All rights reserved.
2020-09-26 16:19:37 +02:00
#
# 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
#
##############################################################################
"""Photos: trombinoscopes
"""
try:
from cStringIO import StringIO
except:
from StringIO import StringIO
from zipfile import ZipFile, BadZipfile
import xml
import tempfile
from notes_log import log
from sco_utils import *
import scolars
import sco_photos
import sco_formsemestre
import sco_groups
import sco_groups_view
import sco_portal_apogee
from sco_formsemestre_status import makeMenu
from sco_pdf import *
import ImportScolars
import sco_excel
from reportlab.lib import colors
def trombino(
context,
REQUEST=None,
group_ids=[], # liste des groupes à afficher
formsemestre_id=None, # utilisé si pas de groupes selectionné
etat=None,
format="html",
dialog_confirmed=False,
):
"""Trombinoscope"""
if not etat:
etat = None # may be passed as ''
# Informations sur les groupes à afficher:
groups_infos = sco_groups_view.DisplayedGroupsInfos(
context, group_ids, formsemestre_id=formsemestre_id, etat=etat, REQUEST=REQUEST
)
#
if format != "html" and not dialog_confirmed:
ok, dialog = check_local_photos_availability(
context, groups_infos, REQUEST, format=format
)
if not ok:
return dialog
if format == "zip":
return _trombino_zip(context, groups_infos, REQUEST)
elif format == "pdf":
return _trombino_pdf(context, groups_infos, REQUEST)
elif format == "pdflist":
return _listeappel_photos_pdf(context, groups_infos, REQUEST)
else:
raise Exception("invalid format")
# return _trombino_html_header(context, REQUEST) + trombino_html(context, group, members, REQUEST=REQUEST) + context.sco_footer(REQUEST)
def _trombino_html_header(context, REQUEST):
return context.sco_header(REQUEST, javascripts=["js/trombino.js"])
def trombino_html(context, groups_infos, REQUEST=None):
"HTML snippet for trombino (with title and menu)"
args = groups_infos.groups_query_args
menuTrombi = [
{"title": "Charger des photos...", "url": "photos_import_files_form?%s" % args},
{
"title": "Obtenir archive Zip des photos",
"url": "trombino?%s&format=zip" % args,
},
{
"title": "Recopier les photos depuis le portail",
"url": "trombino_copy_photos?%s" % args,
},
]
if groups_infos.members:
if groups_infos.tous_les_etuds_du_sem:
ng = "Tous les étudiants"
else:
ng = "Groupe %s" % groups_infos.groups_titles
else:
ng = "Aucun étudiant inscrit dans ce groupe !"
H = [
'<table style="padding-top: 10px; padding-bottom: 10px;"><tr><td><span style="font-style: bold; font-size: 150%%; padding-right: 20px;">%s</span></td>'
% (ng)
]
if groups_infos.members:
H.append(
"<td>" + makeMenu("Gérer les photos", menuTrombi, alone=True) + "</td>"
)
H.append("</tr></table>")
H.append("<div>")
i = 0
for t in groups_infos.members:
H.append(
'<span class="trombi_box"><span class="trombi-photo" id="trombi-%s">'
% t["etudid"]
)
if sco_photos.etud_photo_is_local(context, t, size="small"):
foto = sco_photos.etud_photo_html(context, t, title="", REQUEST=REQUEST)
else: # la photo n'est pas immédiatement dispo
foto = (
'<span class="unloaded_img" id="%s"><img border="0" height="90" alt="en cours" src="/ScoDoc/static/icons/loading.jpg"/></span>'
% t["etudid"]
)
H.append('<a href="ficheEtud?etudid=' + t["etudid"] + '">' + foto + "</a>")
H.append("</span>")
H.append(
'<span class="trombi_legend"><span class="trombi_prenom">'
+ scolars.format_prenom(t["prenom"])
+ '</span><span class="trombi_nom">'
+ scolars.format_nom(t["nom"])
)
H.append("</span></span></span>")
i += 1
H.append("</div>")
H.append(
'<div style="margin-bottom:15px;"><a class="stdlink" href="trombino?format=pdf&amp;%s">Version PDF</a></div>'
% args
)
return "\n".join(H)
def check_local_photos_availability(context, groups_infos, REQUEST, format=""):
"""Verifie que toutes les photos (des gropupes indiqués) sont copiées localement
2020-10-21 22:56:25 +02:00
dans ScoDoc (seules les photos dont nous disposons localement peuvent être exportées
2020-09-26 16:19:37 +02:00
en pdf ou en zip).
Si toutes ne sont pas dispo, retourne un dialogue d'avertissement pour l'utilisateur.
"""
nb_missing = 0
for t in groups_infos.members:
etudid = t["etudid"]
2020-10-21 22:56:25 +02:00
url = sco_photos.etud_photo_url(
context, t, REQUEST=REQUEST
) # -> copy distant files if needed
2020-09-26 16:19:37 +02:00
if not sco_photos.etud_photo_is_local(context, t):
nb_missing += 1
if nb_missing > 0:
parameters = {"group_ids": groups_infos.group_ids, "format": format}
return (
False,
context.confirmDialog(
"""<p>Attention: %d photos ne sont pas disponibles et ne peuvent pas être exportées.</p><p>Vous pouvez <a class="stdlink" href="%s">exporter seulement les photos existantes</a>"""
% (
nb_missing,
groups_infos.base_url
+ "&amp;dialog_confirmed=1&amp;format=%s" % format,
),
dest_url="trombino",
OK="Exporter seulement les photos existantes",
cancel_url="groups_view?curtab=tab-photos&amp;"
+ groups_infos.groups_query_args,
REQUEST=REQUEST,
parameters=parameters,
),
)
else:
return True, ""
def _trombino_zip(context, groups_infos, REQUEST):
"Send photos as zip archive"
data = StringIO()
Z = ZipFile(data, "w")
# assume we have the photos (or the user acknowledged the fact)
# Archive originals (not reduced) images, in JPEG
for t in groups_infos.members:
im_path = sco_photos.photo_pathname(context, t, size="orig")
if not im_path:
continue
img = open(im_path).read()
code_nip = t["code_nip"]
if code_nip:
filename = code_nip + ".jpg"
else:
filename = t["nom"] + "_" + t["prenom"] + "_" + t["etudid"] + ".jpg"
Z.writestr(filename, img)
Z.close()
size = data.tell()
log("trombino_zip: %d bytes" % size)
content_type = "application/zip"
REQUEST.RESPONSE.setHeader(
"content-disposition", 'attachement; filename="trombi.zip"'
)
REQUEST.RESPONSE.setHeader("content-type", content_type)
REQUEST.RESPONSE.setHeader("content-length", size)
return data.getvalue()
# Copy photos from portal to ScoDoc
def trombino_copy_photos(context, group_ids=[], REQUEST=None, dialog_confirmed=False):
"Copy photos from portal to ScoDoc (overwriting local copy)"
groups_infos = sco_groups_view.DisplayedGroupsInfos(
context, group_ids, REQUEST=REQUEST
)
back_url = "groups_view?%s&amp;curtab=tab-photos" % groups_infos.groups_query_args
portal_url = sco_portal_apogee.get_portal_url(context)
header = context.sco_header(REQUEST, page_title="Chargement des photos")
footer = context.sco_footer(REQUEST)
if not portal_url:
return (
header
+ '<p>portail non configuré</p><p><a href="%s">Retour au trombinoscope</a></p>'
% back_url
+ footer
)
if not dialog_confirmed:
return context.confirmDialog(
"""<h2>Copier les photos du portail vers ScoDoc ?</h2>
<p>Les photos du groupe %s présentes dans ScoDoc seront remplacées par celles du portail (si elles existent).</p>
<p>(les photos sont normalement automatiquement copiées lors de leur première utilisation, l'usage de cette fonction n'est nécessaire que si les photos du portail ont été modifiées)</p>
"""
% (groups_infos.groups_titles),
dest_url="",
REQUEST=REQUEST,
cancel_url=back_url,
parameters={"group_ids": group_ids},
)
msg = []
nok = 0
for etud in groups_infos.members:
path, diag = sco_photos.copy_portal_photo_to_fs(context, etud, REQUEST=REQUEST)
msg.append(diag)
if path:
nok += 1
msg.append("<b>%d photos correctement chargées</b>" % nok)
return (
header
+ "<h2>Chargement des photos depuis le portail</h2><ul><li>"
+ "</li><li>".join(msg)
+ "</li></ul>"
+ '<p><a href="%s">retour au trombinoscope</a>' % back_url
+ footer
)
def _get_etud_platypus_image(context, t, image_width=2 * cm):
2020-10-21 22:56:25 +02:00
"""Returns aplatypus object for the photo of student t"""
2020-09-26 16:19:37 +02:00
try:
path = sco_photos.photo_pathname(context, t, size="small")
if not path:
# log('> unknown')
path = sco_photos.UNKNOWN_IMAGE_PATH
im = PILImage.open(path)
w0, h0 = im.size[0], im.size[1]
if w0 > h0:
W = image_width
H = h0 * W / w0
else:
H = image_width
W = w0 * H / h0
return reportlab.platypus.Image(path, width=W, height=H)
except:
log(
"*** exception while processing photo of %s (%s) (path=%s)"
% (t["nom"], t["etudid"], path)
)
raise
def _trombino_pdf(context, groups_infos, REQUEST):
"Send photos as pdf page"
# Generate PDF page
filename = "trombino_%s" % groups_infos.groups_filename + ".pdf"
sem = groups_infos.formsemestre # suppose 1 seul semestre
PHOTOWIDTH = 3 * cm
COLWIDTH = 3.6 * cm
N_PER_ROW = 5 # XXX should be in ScoDoc preferences
StyleSheet = styles.getSampleStyleSheet()
report = StringIO() # in-memory document, no disk file
objects = [
Paragraph(
SU("Trombinoscope " + sem["titreannee"] + " " + groups_infos.groups_titles),
StyleSheet["Heading3"],
)
]
L = []
n = 0
currow = []
log("_trombino_pdf %d elements" % len(groups_infos.members))
for t in groups_infos.members:
img = _get_etud_platypus_image(context, t, image_width=PHOTOWIDTH)
elem = Table(
[
[img],
[
Paragraph(
SU(
scolars.format_sexe(t["sexe"])
+ " "
+ scolars.format_prenom(t["prenom"])
+ " "
+ scolars.format_nom(t["nom"])
),
StyleSheet["Normal"],
)
],
],
colWidths=[PHOTOWIDTH],
)
currow.append(elem)
if n == (N_PER_ROW - 1):
L.append(currow)
currow = []
n = (n + 1) % N_PER_ROW
if currow:
currow += [" "] * (N_PER_ROW - len(currow))
L.append(currow)
if not L:
table = Paragraph(SU("Aucune photo à exporter !"), StyleSheet["Normal"])
else:
table = Table(
L,
colWidths=[COLWIDTH] * N_PER_ROW,
style=TableStyle(
[
# ('RIGHTPADDING', (0,0), (-1,-1), -5*mm),
("VALIGN", (0, 0), (-1, -1), "TOP"),
("GRID", (0, 0), (-1, -1), 0.25, colors.grey),
]
),
)
objects.append(table)
# Build document
document = BaseDocTemplate(report)
document.addPageTemplates(
ScolarsPageTemplate(
2020-10-21 22:56:25 +02:00
document,
context,
preferences=context.get_preferences(sem["formsemestre_id"]),
2020-09-26 16:19:37 +02:00
)
)
document.build(objects)
data = report.getvalue()
return sendPDFFile(REQUEST, data, filename)
# --------------------- Sur une idée de l'IUT d'Orléans:
def _listeappel_photos_pdf(context, groups_infos, REQUEST):
"Doc pdf pour liste d'appel avec photos"
filename = "trombino_%s" % groups_infos.groups_filename + ".pdf"
sem = groups_infos.formsemestre # suppose 1 seul semestre
PHOTOWIDTH = 2 * cm
COLWIDTH = 3.6 * cm
ROWS_PER_PAGE = 26 # XXX should be in ScoDoc preferences
StyleSheet = styles.getSampleStyleSheet()
report = StringIO() # in-memory document, no disk file
objects = [
Paragraph(
SU(
sem["titreannee"]
+ " "
+ groups_infos.groups_titles
+ " (%d)" % len(groups_infos.members)
),
StyleSheet["Heading3"],
)
]
L = []
n = 0
currow = []
log("_listeappel_photos_pdf %d elements" % len(groups_infos.members))
n = len(groups_infos.members)
# npages = n / 2*ROWS_PER_PAGE + 1 # nb de pages papier
# for page in range(npages):
for i in range(n): # page*2*ROWS_PER_PAGE, (page+1)*2*ROWS_PER_PAGE):
t = groups_infos.members[i]
img = _get_etud_platypus_image(context, t, image_width=PHOTOWIDTH)
txt = Paragraph(
SU(
scolars.format_sexe(t["sexe"])
+ " "
+ scolars.format_prenom(t["prenom"])
+ " "
+ scolars.format_nom(t["nom"])
),
StyleSheet["Normal"],
)
if currow:
currow += [""]
currow += [img, txt, ""]
if i % 2:
L.append(currow)
currow = []
if currow:
currow += [" "] * 3
L.append(currow)
if not L:
table = Paragraph(SU("Aucune photo à exporter !"), StyleSheet["Normal"])
else:
table = Table(
L,
colWidths=[2 * cm, 4 * cm, 27 * mm, 5 * mm, 2 * cm, 4 * cm, 27 * mm],
style=TableStyle(
[
# ('RIGHTPADDING', (0,0), (-1,-1), -5*mm),
("VALIGN", (0, 0), (-1, -1), "TOP"),
("GRID", (0, 0), (2, -1), 0.25, colors.grey),
("GRID", (4, 0), (-1, -1), 0.25, colors.grey),
]
),
)
objects.append(table)
# Build document
document = BaseDocTemplate(report)
document.addPageTemplates(
ScolarsPageTemplate(
2020-10-21 22:56:25 +02:00
document,
context,
preferences=context.get_preferences(sem["formsemestre_id"]),
2020-09-26 16:19:37 +02:00
)
)
document.build(objects)
data = report.getvalue()
return sendPDFFile(REQUEST, data, filename)
objects = []
StyleSheet = styles.getSampleStyleSheet()
report = StringIO() # in-memory document, no disk file
filename = ("trombino-%s.pdf" % ng).replace(
" ", "_"
) # XXX should sanitize this filename
objects.append(
Paragraph(SU("Liste " + sem["titreannee"] + " " + ng), StyleSheet["Heading3"])
)
PHOTOWIDTH = 3 * cm
COLWIDTH = 3.6 * cm
L = [] # cells
n = 0
currow = []
for t in T:
n = n + 1
img = _get_etud_platypus_image(context, t, image_width=2 * cm)
currow += [
Paragraph(
SU(
scolars.format_sexe(t["sexe"])
+ " "
+ scolars.format_prenom(t["prenom"])
+ " "
+ scolars.format_nom(t["nom"])
),
StyleSheet["Normal"],
),
"", # empty cell (signature ou autre info a remplir sur papier)
img,
]
if not L:
table = Paragraph(SU("Aucune photo à exporter !"), StyleSheet["Normal"])
else:
table = Table(
L,
colWidths=[COLWIDTH] * 7,
style=TableStyle(
[
("VALIGN", (0, 0), (-1, -1), "TOP"),
("GRID", (0, 0), (2, -1), 0.25, colors.grey),
("GRID", (2, 0), (-1, -1), 0.25, colors.red), # <<<
]
),
)
objects.append(table)
# Réduit sur une page
objects = [KeepInFrame(0, 0, objects, mode="shrink")]
# --- Build document
document = BaseDocTemplate(report)
document.addPageTemplates(
ScolarsPageTemplate(
2020-10-21 22:56:25 +02:00
document,
context,
preferences=context.get_preferences(sem["formsemestre_id"]),
2020-09-26 16:19:37 +02:00
)
)
document.build(objects)
data = report.getvalue()
return sendPDFFile(REQUEST, data, filename)
# --------------------- Upload des photos de tout un groupe
def photos_generate_excel_sample(context, group_ids=[], REQUEST=None):
2020-10-21 22:56:25 +02:00
"""Feuille excel pour import fichiers photos"""
2020-09-26 16:19:37 +02:00
fmt = ImportScolars.sco_import_format()
data = ImportScolars.sco_import_generate_excel_sample(
fmt,
context=context,
group_ids=group_ids,
only_tables=["identite"],
exclude_cols=[
"date_naissance",
"lieu_naissance",
"nationalite",
"statut",
"photo_filename",
],
extra_cols=["fichier_photo"],
REQUEST=REQUEST,
)
return sco_excel.sendExcelFile(REQUEST, data, "ImportPhotos.xls")
def photos_import_files_form(context, group_ids=[], REQUEST=None):
2020-10-21 22:56:25 +02:00
"""Formulaire pour importation photos"""
2020-09-26 16:19:37 +02:00
groups_infos = sco_groups_view.DisplayedGroupsInfos(
context, group_ids, REQUEST=REQUEST
)
back_url = "groups_view?%s&amp;curtab=tab-photos" % groups_infos.groups_query_args
H = [
context.sco_header(REQUEST, page_title="Import des photos des étudiants"),
"""<h2 class="formsemestre">Téléchargement des photos des étudiants</h2>
<p><b>Vous pouvez aussi charger les photos individuellement via la fiche de chaque étudiant (menu "Etudiant" / "Changer la photo").</b></p>
<p class="help">Cette page permet de charger en une seule fois les photos de plusieurs étudiants.<br/>
Il faut d'abord remplir une feuille excel donnant les noms
des fichiers images (une image par étudiant).
</p>
<p class="help">Ensuite, réunir vos images dans un fichier zip, puis télécharger
simultanément le fichier excel et le fichier zip.
</p>
<ol>
<li><a class="stdlink" href="photos_generate_excel_sample?%s">
Obtenir la feuille excel à remplir</a>
</li>
<li style="padding-top: 2em;">
"""
% groups_infos.groups_query_args,
]
F = context.sco_footer(REQUEST)
REQUEST.form["group_ids"] = groups_infos.group_ids
tf = TrivialFormulator(
REQUEST.URL0,
REQUEST.form,
(
("xlsfile", {"title": "Fichier Excel:", "input_type": "file", "size": 40}),
("zipfile", {"title": "Fichier zip:", "input_type": "file", "size": 40}),
("group_ids", {"input_type": "hidden", "type": "list"}),
),
)
if tf[0] == 0:
return "\n".join(H) + tf[1] + "</li></ol>" + F
elif tf[0] == -1:
return REQUEST.RESPONSE.redirect(back_url)
else:
return photos_import_files(
context,
group_ids=tf[2]["group_ids"],
xlsfile=tf[2]["xlsfile"],
zipfile=tf[2]["zipfile"],
REQUEST=REQUEST,
)
def photos_import_files(
context, group_ids=[], xlsfile=None, zipfile=None, REQUEST=None
):
2020-10-21 22:56:25 +02:00
"""Importation des photos"""
2020-09-26 16:19:37 +02:00
groups_infos = sco_groups_view.DisplayedGroupsInfos(
context, group_ids, REQUEST=REQUEST
)
back_url = "groups_view?%s&curtab=tab-photos" % groups_infos.groups_query_args
filename_title = "fichier_photo"
page_title = "Téléchargement des photos des étudiants"
def callback(context, etud, data, filename, REQUEST):
sco_photos.store_photo(context, etud, data, REQUEST)
r = zip_excel_import_files(
context, xlsfile, zipfile, REQUEST, callback, filename_title, page_title
)
return REQUEST.RESPONSE.redirect(back_url + "&amp;head_message=photos%20 importees")
def zip_excel_import_files(
context,
xlsfile=None,
zipfile=None,
REQUEST=None,
callback=None,
filename_title="", # doit obligatoirement etre specifié
page_title="",
):
"""Importation de fichiers à partir d'un excel et d'un zip
La fonction
callback()
est appelé pour chaque fichier trouvé.
"""
# 1- build mapping etudid -> filename
exceldata = xlsfile.read()
if not exceldata:
raise ScoValueError("Fichier excel vide ou invalide")
diag, data = sco_excel.Excel_to_list(exceldata)
if not data: # probably a bug
raise ScoValueError("Fichier excel vide !")
# on doit avoir une colonne etudid et une colonne filename_title ('fichier_photo')
titles = data[0]
try:
etudid_idx = titles.index("etudid")
filename_idx = titles.index(filename_title)
except:
raise ScoValueError(
"Fichier excel incorrect (il faut une colonne etudid et une colonne %s) !"
% filename_title
)
def normfilename(fn, lowercase=True):
"normalisation used to match filenames"
fn = fn.replace("\\", "/") # not sure if this is necessary ?
fn = fn.strip()
if lowercase:
fn = strlower(fn)
fn = fn.split("/")[-1] # use only last component, not directories
return fn
Filename2Etud = {} # filename : etudid
for l in data[1:]:
filename = l[filename_idx].strip()
if filename:
Filename2Etud[normfilename(filename)] = l[etudid_idx]
# 2- Ouvre le zip et
try:
z = ZipFile(zipfile)
except BadZipfile:
raise ScoValueError("Fichier ZIP incorrect !")
ignored_zipfiles = []
stored = [] # [ (etud, filename) ]
for name in z.namelist():
if len(name) > 4 and name[-1] != "/" and "." in name:
data = z.read(name)
# match zip filename with name given in excel
normname = normfilename(name)
if normname in Filename2Etud:
etudid = Filename2Etud[normname]
# ok, store photo
try:
etud = context.getEtudInfo(etudid=etudid, filled=True)[0]
del Filename2Etud[normname]
except:
raise ScoValueError("ID étudiant invalide: %s" % etudid)
callback(
context,
etud,
data,
normfilename(name, lowercase=False),
REQUEST=REQUEST,
)
stored.append((etud, name))
else:
log("zip: zip name %s not in excel !" % name)
ignored_zipfiles.append(name)
else:
if name[-1] != "/":
ignored_zipfiles.append(name)
log("zip: ignoring %s" % name)
if Filename2Etud:
# lignes excel non traitées
unmatched_files = Filename2Etud.keys()
else:
unmatched_files = []
# 3- Result page
H = [
_trombino_html_header(context, REQUEST),
"""<h2 class="formsemestre">%s</h2>
<h3>Opération effectuée</h3>
"""
% page_title,
]
if ignored_zipfiles:
H.append("<h4>Fichiers ignorés dans le zip:</h4><ul>")
for name in ignored_zipfiles:
H.append("<li>%s</li>" % name)
H.append("</ul>")
if unmatched_files:
H.append(
"<h4>Fichiers indiqués dans feuille mais non trouvés dans le zip:</h4><ul>"
)
for name in unmatched_files:
H.append("<li>%s</li>" % name)
H.append("</ul>")
if stored:
H.append("<h4>Fichiers chargés:</h4><ul>")
for (etud, name) in stored:
H.append("<li>%s: <tt>%s</tt></li>" % (etud["nomprenom"], name))
H.append("</ul>")
return "\n".join(H)