ScoDocMM/app/scodoc/sco_recapcomplet.py

1072 lines
40 KiB
Python

# -*- mode: python -*-
# -*- coding: utf-8 -*-
##############################################################################
#
# Gestion scolarite IUT
#
# Copyright (c) 1999 - 2022 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
#
##############################################################################
"""Tableau récapitulatif des notes d'un semestre
"""
import datetime
import json
import time
from xml.etree import ElementTree
from flask import g, request
from flask import make_response, url_for
from app import log
from app.but import bulletin_but
from app.comp import res_sem
from app.comp.res_compat import NotesTableCompat
from app.models import FormSemestre
from app.models.etudiants import Identite
from app.models.evaluations import Evaluation
import app.scodoc.sco_utils as scu
from app.scodoc import html_sco_header
from app.scodoc import sco_bulletins_json
from app.scodoc import sco_bulletins_xml
from app.scodoc import sco_bulletins, sco_excel
from app.scodoc import sco_codes_parcours
from app.scodoc import sco_evaluations
from app.scodoc import sco_evaluation_db
from app.scodoc import sco_formations
from app.scodoc import sco_formsemestre
from app.scodoc import sco_formsemestre_status
from app.scodoc import sco_groups
from app.scodoc import sco_permissions_check
from app.scodoc import sco_preferences
from app.scodoc import sco_etud
from app.scodoc import sco_users
from app.scodoc import sco_xml
from app.scodoc.sco_codes_parcours import DEF, UE_SPORT
def formsemestre_recapcomplet(
formsemestre_id=None,
modejury=False, # affiche lien saisie decision jury
hidemodules=False, # cache colonnes notes modules
hidebac=False, # cache colonne Bac
tabformat="html",
sortcol=None,
xml_with_decisions=False, # XML avec decisions
rank_partition_id=None, # si None, calcul rang global
pref_override=True, # si vrai, les prefs ont la priorite sur le param hidebac
force_publishing=True, # publie les XML/JSON meme si bulletins non publiés
):
"""Page récapitulant les notes d'un semestre.
Grand tableau récapitulatif avec toutes les notes de modules
pour tous les étudiants, les moyennes par UE et générale,
trié par moyenne générale décroissante.
"""
sem = sco_formsemestre.get_formsemestre(formsemestre_id)
F = sco_formations.formation_list(args={"formation_id": sem["formation_id"]})[0]
parcours = sco_codes_parcours.get_parcours_from_code(F["type_parcours"])
formsemestre = FormSemestre.query.get_or_404(formsemestre_id)
# Pour APC (BUT): cache les modules par défaut car moyenne n'a pas de sens
if formsemestre.formation.is_apc():
hidemodules = True
# traduit du DTML
modejury = int(modejury)
hidemodules = (
int(hidemodules) or parcours.UE_IS_MODULE
) # cache les colonnes des modules
pref_override = int(pref_override)
if pref_override:
hidebac = int(sco_preferences.get_preference("recap_hidebac", formsemestre_id))
else:
hidebac = int(hidebac)
xml_with_decisions = int(xml_with_decisions)
force_publishing = int(force_publishing)
isFile = tabformat in ("csv", "xls", "xml", "xlsall", "json")
H = []
if not isFile:
H += [
html_sco_header.sco_header(
page_title="Récapitulatif",
no_side_bar=True,
init_qtip=True,
javascripts=["js/etud_info.js", "js/table_recap.js"],
),
sco_formsemestre_status.formsemestre_status_head(
formsemestre_id=formsemestre_id
),
]
if len(formsemestre.inscriptions) > 0:
H += [
'<form name="f" method="get" action="%s">' % request.base_url,
'<input type="hidden" name="formsemestre_id" value="%s"></input>'
% formsemestre_id,
'<input type="hidden" name="pref_override" value="0"></input>',
]
if modejury:
H.append(
'<input type="hidden" name="modejury" value="%s"></input>'
% modejury
)
H.append(
'<select name="tabformat" onchange="document.f.submit()" class="noprint">'
)
for (format, label) in (
("html", "HTML"),
("xls", "Fichier tableur (Excel)"),
("xlsall", "Fichier tableur avec toutes les évals"),
("csv", "Fichier tableur (CSV)"),
("xml", "Fichier XML"),
("json", "JSON"),
):
if format == tabformat:
selected = " selected"
else:
selected = ""
H.append('<option value="%s"%s>%s</option>' % (format, selected, label))
H.append("</select>")
H.append(
f"""&nbsp;(cliquer sur un nom pour afficher son bulletin ou <a class="stdlink"
href="{url_for('notes.formsemestre_bulletins_pdf',
scodoc_dept=g.scodoc_dept, formsemestre_id=formsemestre_id)}">
ici avoir le classeur papier</a>)
<div class="warning">Nouvelle version: export excel inachevés. Merci de signaler les problèmes.</div>
"""
)
data = do_formsemestre_recapcomplet(
formsemestre_id,
format=tabformat,
hidemodules=hidemodules,
hidebac=hidebac,
modejury=modejury,
sortcol=sortcol,
xml_with_decisions=xml_with_decisions,
rank_partition_id=rank_partition_id,
force_publishing=force_publishing,
)
if tabformat == "xml":
response = make_response(data)
response.headers["Content-Type"] = scu.XML_MIMETYPE
return response
H.append(data)
if not isFile:
if len(formsemestre.inscriptions) > 0:
H.append("</form>")
H.append(
"""<p><a class="stdlink" href="formsemestre_pvjury?formsemestre_id=%s">Voir les décisions du jury</a></p>"""
% formsemestre_id
)
if sco_permissions_check.can_validate_sem(formsemestre_id):
H.append("<p>")
if modejury:
H.append(
"""<a class="stdlink" href="formsemestre_validation_auto?formsemestre_id=%s">Calcul automatique des décisions du jury</a></p>"""
% (formsemestre_id,)
)
else:
H.append(
"""<a class="stdlink" href="formsemestre_recapcomplet?formsemestre_id=%s&modejury=1&hidemodules=1">Saisie des décisions du jury</a>"""
% formsemestre_id
)
H.append("</p>")
if sco_preferences.get_preference("use_ue_coefs", formsemestre_id):
H.append(
"""
<p class="infop">utilise les coefficients d'UE pour calculer la moyenne générale.</p>
"""
)
H.append(html_sco_header.sco_footer())
# HTML or binary data ?
if len(H) > 1:
return "".join(H)
elif len(H) == 1:
return H[0]
else:
return H
def do_formsemestre_recapcomplet(
formsemestre_id=None,
format="html", # html, xml, xls, xlsall, json
hidemodules=False, # ne pas montrer les modules (ignoré en XML)
hidebac=False, # pas de colonne Bac (ignoré en XML)
xml_nodate=False, # format XML sans dates (sert pour debug cache: comparaison de XML)
modejury=False, # saisie décisions jury
sortcol=None, # indice colonne a trier dans table T
xml_with_decisions=False,
disable_etudlink=False,
rank_partition_id=None, # si None, calcul rang global
force_publishing=True,
):
"""Calcule et renvoie le tableau récapitulatif."""
formsemestre = FormSemestre.query.get_or_404(formsemestre_id)
if format == "html" and not modejury:
res: NotesTableCompat = res_sem.load_formsemestre_results(formsemestre)
data, filename = gen_formsemestre_recapcomplet_html(formsemestre, res)
else:
data, filename, format = make_formsemestre_recapcomplet(
formsemestre_id=formsemestre_id,
format=format,
hidemodules=hidemodules,
hidebac=hidebac,
xml_nodate=xml_nodate,
modejury=modejury,
sortcol=sortcol,
xml_with_decisions=xml_with_decisions,
disable_etudlink=disable_etudlink,
rank_partition_id=rank_partition_id,
force_publishing=force_publishing,
)
# ---
if format == "xml" or format == "html":
return data
elif format == "csv":
return scu.send_file(data, filename=filename, mime=scu.CSV_MIMETYPE)
elif format.startswith("xls") or format.startswith("xlsx"):
return scu.send_file(data, filename=filename, mime=scu.XLSX_MIMETYPE)
elif format == "json":
js = json.dumps(data, indent=1, cls=scu.ScoDocJSONEncoder)
return scu.send_file(
js, filename=filename, suffix=scu.JSON_SUFFIX, mime=scu.JSON_MIMETYPE
)
else:
raise ValueError("unknown format %s" % format)
def make_formsemestre_recapcomplet(
formsemestre_id=None,
format="html", # html, xml, xls, xlsall, json
hidemodules=False, # ne pas montrer les modules (ignoré en XML)
hidebac=False, # pas de colonne Bac (ignoré en XML)
xml_nodate=False, # format XML sans dates (sert pour debug cache: comparaison de XML)
modejury=False, # saisie décisions jury
sortcol=None, # indice colonne a trier dans table T
xml_with_decisions=False,
disable_etudlink=False,
rank_partition_id=None, # si None, calcul rang global
force_publishing=True, # donne bulletins JSON/XML meme si non publiés
):
"""Grand tableau récapitulatif avec toutes les notes de modules
pour tous les étudiants, les moyennes par UE et générale,
trié par moyenne générale décroissante.
"""
civ_nom_prenom = False # 3 colonnes différentes ou une seule avec prénom abrégé ?
if format == "xml":
return _formsemestre_recapcomplet_xml(
formsemestre_id,
xml_nodate,
xml_with_decisions=xml_with_decisions,
force_publishing=force_publishing,
)
elif format == "json":
return _formsemestre_recapcomplet_json(
formsemestre_id,
xml_nodate=xml_nodate,
xml_with_decisions=xml_with_decisions,
force_publishing=force_publishing,
)
if format[:3] == "xls":
civ_nom_prenom = True # 3 cols: civilite, nom, prenom
keep_numeric = True # pas de conversion des notes en strings
else:
keep_numeric = False
if hidebac:
admission_extra_cols = []
else:
admission_extra_cols = [
"type_admission",
"classement",
"apb_groupe",
"apb_classement_gr",
]
formsemestre = FormSemestre.query.get_or_404(formsemestre_id)
# A ré-écrire XXX
sem = sco_formsemestre.do_formsemestre_list(
args={"formsemestre_id": formsemestre_id}
)[0]
parcours = formsemestre.formation.get_parcours()
nt: NotesTableCompat = res_sem.load_formsemestre_results(formsemestre)
modimpls = formsemestre.modimpls_sorted
ues = nt.get_ues_stat_dict() # incluant le(s) UE de sport
partitions, partitions_etud_groups = sco_groups.get_formsemestre_groups(
formsemestre_id
)
if rank_partition_id and format == "html":
# Calcul rang sur une partition et non sur l'ensemble
# seulement en format HTML (car colonnes rangs toujours presentes en xls)
rank_partition = sco_groups.get_partition(rank_partition_id)
rank_label = "Rg (%s)" % rank_partition["partition_name"]
else:
rank_partition = sco_groups.get_default_partition(formsemestre_id)
rank_label = "Rg"
T = nt.get_table_moyennes_triees()
if not T:
return "", "", format
# Construit une liste de listes de chaines: le champs du tableau resultat (HTML ou CSV)
F = []
h = [rank_label]
if civ_nom_prenom:
h += ["Civilité", "Nom", "Prénom"]
else:
h += ["Nom"]
if not hidebac:
h.append("Bac")
# Si CSV ou XLS, indique tous les groupes
if format[:3] == "xls" or format == "csv":
for partition in partitions:
h.append("%s" % partition["partition_name"])
else:
h.append("Gr")
h.append("Moy")
# Ajoute rangs dans groupe seulement si CSV ou XLS
if format[:3] == "xls" or format == "csv":
for partition in partitions:
h.append("rang_%s" % partition["partition_name"])
cod2mod = {} # code : moduleimpl
mod_evals = {} # moduleimpl_id : liste de toutes les evals de ce module
for ue in ues:
if ue["type"] != UE_SPORT:
h.append(ue["acronyme"])
else: # UE_SPORT:
# n'affiche pas la moyenne d'UE dans ce cas
# mais laisse col. vide si modules affichés (pour séparer les UE)
if not hidemodules:
h.append("")
pass
if not hidemodules and not ue["is_external"]:
for modimpl in modimpls:
if modimpl.module.ue_id == ue["ue_id"]:
code = modimpl.module.code
h.append(code)
cod2mod[code] = modimpl # pour fabriquer le lien
if format == "xlsall":
evals = nt.modimpls_results[
modimpl.id
].get_evaluations_completes(modimpl)
# evals = nt.get_mod_evaluation_etat_list(...
mod_evals[modimpl.id] = evals
h += _list_notes_evals_titles(code, evals)
h += admission_extra_cols
h += ["code_nip", "etudid"]
F.append(h)
def fmtnum(val): # conversion en nombre pour cellules excel
if keep_numeric:
try:
return float(val)
except:
return val
else:
return val
# Compte les decisions de jury
codes_nb = scu.DictDefault(defaultvalue=0)
#
is_dem = {} # etudid : bool
for t in T:
etudid = t[-1]
dec = nt.get_etud_decision_sem(etudid)
if dec:
codes_nb[dec["code"]] += 1
etud_etat = nt.get_etud_etat(etudid)
if etud_etat == "D":
gr_name = "Dém."
is_dem[etudid] = True
elif etud_etat == DEF:
gr_name = "Déf."
is_dem[etudid] = False
else:
group = sco_groups.get_etud_main_group(etudid, formsemestre_id)
gr_name = group["group_name"] or ""
is_dem[etudid] = False
if rank_partition_id:
rang_gr, _, rank_gr_name = sco_bulletins.get_etud_rangs_groups(
etudid, formsemestre_id, partitions, partitions_etud_groups, nt
)
if rank_gr_name[rank_partition_id]:
rank = "%s %s" % (
rank_gr_name[rank_partition_id],
rang_gr[rank_partition_id],
)
else:
rank = ""
else:
rank = nt.get_etud_rang(etudid)
e = nt.identdict[etudid]
if civ_nom_prenom:
sco_etud.format_etud_ident(e)
l = [rank, e["civilite_str"], e["nom_disp"], e["prenom"]] # civ, nom prenom
else:
l = [rank, nt.get_nom_short(etudid)] # rang, nom,
e["admission"] = {}
if not hidebac:
e["admission"] = nt.etuds_dict[etudid].admission.first()
if e["admission"]:
bac = nt.etuds_dict[etudid].admission[0].get_bac()
l.append(bac.abbrev())
else:
l.append("")
if format[:3] == "xls" or format == "csv": # tous les groupes
for partition in partitions:
group = partitions_etud_groups[partition["partition_id"]].get(
etudid, None
)
if group:
l.append(group["group_name"])
else:
l.append("")
else:
l.append(gr_name) # groupe
# Moyenne générale
l.append(fmtnum(scu.fmt_note(t[0], keep_numeric=keep_numeric)))
# Ajoute rangs dans groupes seulement si CSV ou XLS
if format[:3] == "xls" or format == "csv":
rang_gr, _, gr_name = sco_bulletins.get_etud_rangs_groups(
etudid, formsemestre_id, partitions, partitions_etud_groups, nt
)
for partition in partitions:
l.append(rang_gr[partition["partition_id"]])
# Nombre d'UE au dessus de 10
# t[i] est une chaine :-)
# nb_ue_ok = sum(
# [t[i] > 10 for i, ue in enumerate(ues, start=1) if ue["type"] != UE_SPORT]
# )
ue_index = [] # indices des moy UE dans l (pour appliquer style css)
for i, ue in enumerate(ues, start=1):
if ue["type"] != UE_SPORT:
l.append(
fmtnum(scu.fmt_note(t[i], keep_numeric=keep_numeric))
) # moyenne etud dans ue
else: # UE_SPORT:
# n'affiche pas la moyenne d'UE dans ce cas
if not hidemodules:
l.append("")
ue_index.append(len(l) - 1)
if not hidemodules and not ue["is_external"]:
j = 0
for modimpl in modimpls:
if modimpl.module.ue_id == ue["ue_id"]:
l.append(
fmtnum(
scu.fmt_note(
t[j + len(ues) + 1], keep_numeric=keep_numeric
)
)
) # moyenne etud dans module
if format == "xlsall":
l += _list_notes_evals(mod_evals[modimpl.id], etudid)
j += 1
if not hidebac:
for k in admission_extra_cols:
l.append(getattr(e["admission"], k, "") or "")
l.append(
nt.identdict[etudid]["code_nip"] or ""
) # avant-derniere colonne = code_nip
l.append(etudid) # derniere colonne = etudid
F.append(l)
# Dernière ligne: moyennes, min et max des UEs et modules
if not hidemodules: # moy/min/max dans chaque module
mods_stats = {} # moduleimpl_id : stats
for modimpl in modimpls:
mods_stats[modimpl.id] = nt.get_mod_stats(modimpl.id)
def add_bottom_stat(key, title, corner_value=""):
l = ["", title]
if civ_nom_prenom:
l += ["", ""]
if not hidebac:
l.append("")
if format[:3] == "xls" or format == "csv":
l += [""] * len(partitions)
else:
l += [""]
l.append(corner_value)
if format[:3] == "xls" or format == "csv":
for _ in partitions:
l += [""] # rangs dans les groupes
for ue in ues:
if ue["type"] != UE_SPORT:
if key == "nb_valid_evals":
l.append("")
elif key == "coef":
if sco_preferences.get_preference("use_ue_coefs", formsemestre_id):
l.append("%2.3f" % ue["coefficient"])
else:
l.append("")
else:
if key == "ects":
if keep_numeric:
l.append(ue[key])
else:
l.append(str(ue[key]))
else:
l.append(scu.fmt_note(ue[key], keep_numeric=keep_numeric))
else: # UE_SPORT:
# n'affiche pas la moyenne d'UE dans ce cas
if not hidemodules:
l.append("")
# ue_index.append(len(l) - 1)
if not hidemodules and not ue["is_external"]:
for modimpl in modimpls:
if modimpl.module.ue_id == ue["ue_id"]:
if key == "coef":
coef = modimpl.module.coefficient
if format[:3] != "xls":
coef = str(coef)
l.append(coef)
elif key == "ects":
l.append("") # ECTS module ?
else:
val = mods_stats[modimpl.id][key]
if key == "nb_valid_evals":
if (
format[:3] != "xls"
): # garde val numerique pour excel
val = str(val)
else: # moyenne du module
val = scu.fmt_note(val, keep_numeric=keep_numeric)
l.append(val)
if format == "xlsall":
l += _list_notes_evals_stats(mod_evals[modimpl.id], key)
if modejury:
l.append("") # case vide sur ligne "Moyennes"
l += [""] * len(admission_extra_cols) # infos admission vides ici
F.append(l + ["", ""]) # ajoute cellules code_nip et etudid inutilisees ici
add_bottom_stat(
"min", "Min", corner_value=scu.fmt_note(nt.moy_min, keep_numeric=keep_numeric)
)
add_bottom_stat(
"max", "Max", corner_value=scu.fmt_note(nt.moy_max, keep_numeric=keep_numeric)
)
add_bottom_stat(
"moy",
"Moyennes",
corner_value=scu.fmt_note(nt.moy_moy, keep_numeric=keep_numeric),
)
add_bottom_stat("coef", "Coef")
add_bottom_stat("nb_valid_evals", "Nb évals")
add_bottom_stat("ects", "ECTS")
# Génération de la table au format demandé
if format == "html":
# Table format HTML
H = [
"""
<script type="text/javascript">
function va_saisir(formsemestre_id, etudid) {
loc = 'formsemestre_validation_etud_form?formsemestre_id='+formsemestre_id+'&etudid='+etudid;
loc += '#etudid' + etudid;
document.location=loc;
}
</script>
<table class="notes_recapcomplet gt_table_searchable compact" id="recapcomplet">
"""
]
if sortcol: # sort table using JS sorttable
H.append(
"""<script type="text/javascript">
</script>
"""
% (int(sortcol))
)
ligne_titres_head = _ligne_titres(
ue_index, F, cod2mod, modejury, with_modules_links=False
)
ligne_titres_foot = _ligne_titres(
ue_index, F, cod2mod, modejury, with_modules_links=True
)
H.append("<thead>\n" + ligne_titres_head + "\n</thead>\n<tbody>\n")
if disable_etudlink:
etudlink = "%(name)s"
else:
etudlink = """<a
href="formsemestre_bulletinetud?formsemestre_id=%(formsemestre_id)s&etudid=%(etudid)s&version=selectedevals"
id="%(etudid)s" class="etudinfo">%(name)s</a>"""
ir = 0
nblines = len(F) - 1
for l in F[1:]:
etudid = l[-1]
if ir == nblines - 6:
H.append("</tbody>")
H.append("<tfoot>")
if ir >= nblines - 6:
# dernieres lignes:
el = l[1]
styl = (
"recap_row_min",
"recap_row_max",
"recap_row_moy",
"recap_row_coef",
"recap_row_nbeval",
"recap_row_ects",
)[ir - nblines + 6]
cells = '<tr class="%s sortbottom">' % styl
else:
el = etudlink % {
"formsemestre_id": formsemestre_id,
"etudid": etudid,
"name": l[1],
}
if ir % 2 == 0:
cells = '<tr class="recap_row_even" id="etudid%s">' % etudid
else:
cells = '<tr class="recap_row_odd" id="etudid%s">' % etudid
ir += 1
# XXX nsn = [ x.replace('NA', '-') for x in l[:-2] ]
# notes sans le NA:
nsn = l[:-2] # copy
for i in range(len(nsn)):
if nsn[i] == "NA":
nsn[i] = "-"
try:
order = int(nsn[0].split()[0])
except:
order = 99999
cells += (
f'<td class="recap_col" data-order="{order:05d}">{nsn[0]}</td>' # rang
)
cells += '<td class="recap_col">%s</td>' % el # nom etud (lien)
if not hidebac:
cells += '<td class="recap_col_bac">%s</td>' % nsn[2] # bac
idx_col_gr = 3
else:
idx_col_gr = 2
cells += '<td class="recap_col">%s</td>' % nsn[idx_col_gr] # group name
# Style si moyenne generale < barre
idx_col_moy = idx_col_gr + 1
cssclass = "recap_col_moy"
try:
if float(nsn[idx_col_moy]) < (parcours.BARRE_MOY - scu.NOTES_TOLERANCE):
cssclass = "recap_col_moy_inf"
except:
pass
cells += '<td class="%s">%s</td>' % (cssclass, nsn[idx_col_moy])
ue_number = 0
for i in range(idx_col_moy + 1, len(nsn)):
if i in ue_index:
cssclass = "recap_col_ue"
# grise si moy UE < barre
ue = ues[ue_number]
ue_number += 1
if (ir < (nblines - 4)) or (ir == nblines - 3):
try:
if float(nsn[i]) < parcours.get_barre_ue(
ue["type"]
): # NOTES_BARRE_UE
cssclass = "recap_col_ue_inf"
elif float(nsn[i]) >= parcours.NOTES_BARRE_VALID_UE:
cssclass = "recap_col_ue_val"
except:
pass
else:
cssclass = "recap_col"
if (
ir == nblines - 3
): # si moyenne generale module < barre ue, surligne:
try:
if float(nsn[i]) < parcours.get_barre_ue(ue["type"]):
cssclass = "recap_col_moy_inf"
except:
pass
cells += '<td class="%s">%s</td>' % (cssclass, nsn[i])
if modejury and etudid:
decision_sem = nt.get_etud_decision_sem(etudid)
if is_dem[etudid]:
code = "DEM"
act = ""
elif decision_sem:
code = decision_sem["code"]
act = "(modifier)"
else:
code = ""
act = "saisir"
cells += '<td class="decision">%s' % code
if act:
# cells += ' <a href="formsemestre_validation_etud_form?formsemestre_id=%s&etudid=%s">%s</a>' % (formsemestre_id, etudid, act)
cells += (
""" <a href="#" onclick="va_saisir('%s', '%s')">%s</a>"""
% (formsemestre_id, etudid, act)
)
cells += "</td>"
H.append(cells + "</tr>")
H.append(ligne_titres_foot)
H.append("</tfoot>")
H.append("</table>")
# Form pour choisir partition de classement:
if not modejury and partitions:
H.append("Afficher le rang des groupes de: ")
if not rank_partition_id:
checked = "checked"
else:
checked = ""
H.append(
'<input type="radio" name="rank_partition_id" value="" onchange="document.f.submit()" %s/>tous '
% (checked)
)
for p in partitions:
if p["partition_id"] == rank_partition_id:
checked = "checked"
else:
checked = ""
H.append(
'<input type="radio" name="rank_partition_id" value="%s" onchange="document.f.submit()" %s/>%s '
% (p["partition_id"], checked, p["partition_name"])
)
# recap des decisions jury (nombre dans chaque code):
if codes_nb:
H.append("<h4>Décisions du jury</h4><table>")
cods = list(codes_nb.keys())
cods.sort()
for cod in cods:
H.append("<tr><td>%s</td><td>%d</td></tr>" % (cod, codes_nb[cod]))
H.append("</table>")
# Avertissements
if formsemestre.formation.is_apc():
H.append(
"""<p class="help">Pour les formations par compétences (comme le BUT), la moyenne générale est purement indicative et ne devrait pas être communiquée aux étudiants.</p>"""
)
return "\n".join(H), "", "html"
elif format == "csv":
CSV = scu.CSV_LINESEP.join(
[scu.CSV_FIELDSEP.join([str(x) for x in l]) for l in F]
)
semname = sem["titre_num"].replace(" ", "_")
date = time.strftime("%d-%m-%Y")
filename = "notes_modules-%s-%s.csv" % (semname, date)
return CSV, filename, "csv"
elif format[:3] == "xls":
semname = sem["titre_num"].replace(" ", "_")
date = time.strftime("%d-%m-%Y")
if format == "xls":
filename = "notes_modules-%s-%s%s" % (semname, date, scu.XLSX_SUFFIX)
else:
filename = "notes_modules_evals-%s-%s%s" % (semname, date, scu.XLSX_SUFFIX)
sheet_name = "notes %s %s" % (semname, date)
if len(sheet_name) > 31:
sheet_name = "notes %s %s" % ("...", date)
xls = sco_excel.excel_simple_table(
titles=["etudid", "code_nip"] + F[0][:-2],
lines=[
[x[-1], x[-2]] + x[:-2] for x in F[1:]
], # reordonne cols (etudid et nip en 1er),
sheet_name=sheet_name,
)
return xls, filename, "xls"
else:
raise ValueError("unknown format %s" % format)
def _ligne_titres(ue_index, F, cod2mod, modejury, with_modules_links=True):
"""Cellules de la ligne de titre (haut ou bas)"""
cells = '<tr class="recap_row_tit sortbottom" id="recap_trtit">'
for i in range(len(F[0]) - 2):
if i in ue_index:
cls = "recap_tit_ue"
else:
cls = "recap_tit"
attr = f'class="{cls}"'
if i == 0 or F[0][i] == "classement": # Rang: force tri numerique
try:
order = int(F[0][i].split()[0])
except:
order = 99999
attr += f' data-order="{order:05d}"'
if F[0][i] in cod2mod: # lien vers etat module
modimpl = cod2mod[F[0][i]]
if with_modules_links:
href = url_for(
"notes.moduleimpl_status",
scodoc_dept=g.scodoc_dept,
moduleimpl_id=modimpl.id,
)
else:
href = ""
cells += f"""<td {attr}><a href="{href}" title="{modimpl.module.titre} ({
sco_users.user_info(modimpl.responsable_id)["nomcomplet"]})">{F[0][i]}</a></td>"""
else:
cells += f"<td {attr}>{F[0][i]}</td>"
if modejury:
cells += '<td class="recap_tit">Décision</td>'
return cells + "</tr>"
def _list_notes_evals(evals: list[Evaluation], etudid: int) -> list[str]:
"""Liste des notes des evaluations completes de ce module
(pour table xls avec evals)
"""
L = []
for e in evals:
notes_db = sco_evaluation_db.do_evaluation_get_all_notes(e.evaluation_id)
if etudid in notes_db:
val = notes_db[etudid]["value"]
else:
# Note manquante mais prise en compte immédiate: affiche ATT
val = scu.NOTES_ATTENTE
val_fmt = scu.fmt_note(val, keep_numeric=True)
L.append(val_fmt)
return L
def _list_notes_evals_titles(codemodule: str, evals: list[Evaluation]) -> list[str]:
"""Liste des titres des evals completes"""
L = []
eval_index = len(evals) - 1
for e in evals:
L.append(
codemodule
+ "-"
+ str(eval_index)
+ "-"
+ (e.jour.isoformat() if e.jour else "")
)
eval_index -= 1
return L
def _list_notes_evals_stats(evals: list[Evaluation], key: str) -> list[str]:
"""Liste des stats (moy, ou rien!) des evals completes"""
L = []
for e in evals:
if key == "moy":
# TODO #sco92
# val = e["etat"]["moy_num"]
# L.append(scu.fmt_note(val, keep_numeric=True))
L.append("")
elif key == "max":
L.append(e.note_max)
elif key == "min":
L.append(0.0)
elif key == "coef":
L.append(e.coefficient)
else:
L.append("") # on n'a pas sous la main min/max
return L
def _formsemestre_recapcomplet_xml(
formsemestre_id,
xml_nodate,
xml_with_decisions=False,
force_publishing=True,
):
"XML export: liste tous les bulletins XML."
formsemestre = FormSemestre.query.get_or_404(formsemestre_id)
nt: NotesTableCompat = res_sem.load_formsemestre_results(formsemestre)
T = nt.get_table_moyennes_triees()
if not T:
return "", "", "xml"
if xml_nodate:
docdate = ""
else:
docdate = datetime.datetime.now().isoformat()
doc = ElementTree.Element(
"recapsemestre", formsemestre_id=str(formsemestre_id), date=docdate
)
evals = sco_evaluations.do_evaluation_etat_in_sem(formsemestre_id)
doc.append(
ElementTree.Element(
"evals_info",
nb_evals_completes=str(evals["nb_evals_completes"]),
nb_evals_en_cours=str(evals["nb_evals_en_cours"]),
nb_evals_vides=str(evals["nb_evals_vides"]),
date_derniere_note=str(evals["last_modif"]),
)
)
for t in T:
etudid = t[-1]
sco_bulletins_xml.make_xml_formsemestre_bulletinetud(
formsemestre_id,
etudid,
doc=doc,
force_publishing=force_publishing,
xml_nodate=xml_nodate,
xml_with_decisions=xml_with_decisions,
)
return (
sco_xml.XML_HEADER + ElementTree.tostring(doc).decode(scu.SCO_ENCODING),
"",
"xml",
)
def _formsemestre_recapcomplet_json(
formsemestre_id,
xml_nodate=False,
xml_with_decisions=False,
force_publishing=True,
):
"""JSON export: liste tous les bulletins JSON
:param xml_nodate(bool): indique la date courante (attribut docdate)
:param force_publishing: donne les bulletins même si non "publiés sur portail"
:returns: dict, "", "json"
"""
formsemestre = FormSemestre.query.get_or_404(formsemestre_id)
is_apc = formsemestre.formation.is_apc()
if xml_nodate:
docdate = ""
else:
docdate = datetime.datetime.now().isoformat()
evals = sco_evaluations.do_evaluation_etat_in_sem(formsemestre_id)
J = {
"docdate": docdate,
"formsemestre_id": formsemestre_id,
"evals_info": {
"nb_evals_completes": evals["nb_evals_completes"],
"nb_evals_en_cours": evals["nb_evals_en_cours"],
"nb_evals_vides": evals["nb_evals_vides"],
"date_derniere_note": evals["last_modif"],
},
"bulletins": [],
}
bulletins = J["bulletins"]
formsemestre = FormSemestre.query.get_or_404(formsemestre_id)
nt: NotesTableCompat = res_sem.load_formsemestre_results(formsemestre)
T = nt.get_table_moyennes_triees()
for t in T:
etudid = t[-1]
if is_apc:
etud = Identite.query.get(etudid)
r = bulletin_but.BulletinBUT(formsemestre)
bul = r.bulletin_etud(etud, formsemestre)
else:
bul = sco_bulletins_json.formsemestre_bulletinetud_published_dict(
formsemestre_id,
etudid,
force_publishing=force_publishing,
xml_with_decisions=xml_with_decisions,
)
bulletins.append(bul)
return J, "", "json"
def formsemestres_bulletins(annee_scolaire):
"""Tous les bulletins des semestres publiés des semestres de l'année indiquée.
:param annee_scolaire(int): année de début de l'année scolaire
:returns: JSON
"""
jslist = []
sems = sco_formsemestre.list_formsemestre_by_etape(annee_scolaire=annee_scolaire)
log("formsemestres_bulletins(%s): %d sems" % (annee_scolaire, len(sems)))
for sem in sems:
J, _, _ = _formsemestre_recapcomplet_json(
sem["formsemestre_id"], force_publishing=False
)
jslist.append(J)
return scu.sendJSON(jslist)
def _gen_cell(key: str, row: dict, elt="td"):
"html table cell"
klass = row.get(f"_{key}_class")
attrs = f'class="{klass}"' if klass else ""
order = row.get(f"_{key}_order")
if order:
attrs += f' data-order="{order}"'
content = row.get(key, "")
target = row.get(f"_{key}_target")
if content or target: # avec lien
href = f'href="{target}"' if target else ""
content = f'<a {href} {row.get(f"_{key}_target_attrs", "")}>{content}</a>'
return f"<{elt} {attrs}>{content}</{elt}>"
def _gen_row(keys: list[str], row, elt="td"):
klass = row.get("_tr_class")
tr_class = f'class="{klass}"' if klass else ""
return f'<tr {tr_class}>{"".join([_gen_cell(key, row, elt) for key in keys])}</tr>'
def gen_formsemestre_recapcomplet_html(
formsemestre: FormSemestre, res: NotesTableCompat
):
"""Construit table recap pour le BUT
Return: data, filename
"""
rows, footer_rows, titles, column_ids = res.get_table_recap(convert_values=True)
if not rows:
return (
'<div class="table_recap"><div class="message">aucun étudiant !</div></div>',
"",
)
H = ['<div class="table_recap"><table class="table_recap">']
# header
H.append(
f"""
<thead>
{_gen_row(column_ids, titles, "th")}
</thead>
"""
)
# body
H.append("<tbody>")
for row in rows:
H.append(f"{_gen_row(column_ids, row)}\n")
H.append("</tbody>\n")
# footer
H.append("<tfoot>")
idx_last = len(footer_rows) - 1
for i, row in enumerate(footer_rows):
H.append(f'{_gen_row(column_ids, row, "th" if i == idx_last else "td")}\n')
H.append(
"""
</tfoot>
</table>
</div>
"""
)
return (
"".join(H),
f'recap-{formsemestre.titre_num().replace(" ", "_")}-{time.strftime("%d-%m-%Y")}',
) # suffix ?