TestsScoDoc7API/sco_pdf.py

356 lines
12 KiB
Python
Raw Normal View History

2020-09-26 16:19:37 +02:00
# -*- mode: python -*-
# -*- coding: utf-8 -*-
##############################################################################
#
# Gestion scolarite IUT
#
# Copyright (c) 1999 - 2020 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
#
##############################################################################
"""Generation de PDF: définitions diverses et gestion du verrou
reportlab n'est pas réentrante: il ne faut qu'une seule opération PDF au même moment.
Tout accès à ReportLab doit donc être précédé d'un PDFLOCK.acquire()
et terminé par un PDFLOCK.release()
"""
import time, cStringIO
from types import StringType
import unicodedata
import traceback
import reportlab
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, Frame, PageBreak
from reportlab.platypus import Table, TableStyle, Image, KeepInFrame
from reportlab.platypus.flowables import Flowable
from reportlab.platypus.doctemplate import PageTemplate, BaseDocTemplate
from reportlab.lib.styles import getSampleStyleSheet
from reportlab.rl_config import defaultPageSize
from reportlab.lib.units import inch, cm, mm
from reportlab.lib.colors import pink, black, red, blue, green, magenta, red
from reportlab.lib.colors import Color
from reportlab.lib.enums import TA_LEFT, TA_RIGHT, TA_CENTER, TA_JUSTIFY
from reportlab.lib import styles
from reportlab.lib.pagesizes import letter, A4, landscape
from sco_utils import *
from notes_log import log
from SuppressAccents import suppression_diacritics
from VERSION import SCOVERSION, SCONAME
PAGE_HEIGHT = defaultPageSize[1]
PAGE_WIDTH = defaultPageSize[0]
DEFAULT_PDF_FOOTER_TEMPLATE = CONFIG.DEFAULT_PDF_FOOTER_TEMPLATE
def SU(s):
"convert s from SCO default encoding to UTF8"
# Mis en service le 4/11/06, passage à ReportLab 2.0
if not s:
s = ""
# Remplace caractères composés
# eg 'e\xcc\x81' COMBINING ACUTE ACCENT par '\xc3\xa9' LATIN SMALL LETTER E WITH ACUTE
# car les "combining accents" ne sont pas traités par ReportLab mais peuvent
# nous être envoyés par certains navigateurs ou imports
2020-09-26 16:19:37 +02:00
u = unicodedata.normalize("NFC", unicode(s, SCO_ENCODING, "replace"))
return u.encode("utf8")
def _splitPara(txt):
"split a string, returns a list of <para > ... </para>"
L = []
closetag = "</para>"
l = len(closetag)
start = 0
e = -1
while 1:
b = txt.find("<para", start)
if b < 0:
if e < 0:
L.append(txt) # no para, just return text
break
e = txt.find(closetag, b)
if e < 0:
raise ValueError("unbalanced para tags")
L.append(txt[b : e + l])
start = e
# fix para: must be followed by a newline (? Reportlab bug turnaround ?)
L = [re.sub("<para(.*?)>([^\n\r])", "<para\\1>\n\\2", p) for p in L]
return L
def makeParas(txt, style, suppress_empty=False):
"""Returns a list of Paragraph instances from a text
with one or more <para> ... </para>
"""
try:
paras = _splitPara(txt)
if suppress_empty:
r = []
for para in paras:
m = re.match("\s*<\s*para.*>\s*(.*)\s*<\s*/\s*para\s*>\s*", para)
if not m:
r.append(para) # not a paragraph, keep it
else:
if m.group(1): # non empty paragraph
r.append(para)
paras = r
return [Paragraph(SU(s), style) for s in paras]
except Exception as e:
if type(e) is IOError:
detail = " " + e.message
else:
detail = ""
log(traceback.format_exc())
log("Invalid pdf para format: %s" % txt)
return [
Paragraph(
SU(
'<font color="red"><i>Erreur: format invalide{}</i></font>'.format(
detail
)
),
style,
)
]
def bold_paras(L, tag="b", close=None):
"""Put each (string) element of L between <b>
L is a dict or sequence. (if dict, elements with key begining by _ are unaffected)
"""
b = "<" + tag + ">"
if not close:
close = "</" + tag + ">"
if hasattr(L, "keys"):
# L is a dict
for k in L:
x = L[k]
if k[0] != "_":
L[k] = b + L[k] or "" + close
return L
else:
# L is a sequence
return [b + (x or "") + close for x in L]
class ScolarsPageTemplate(PageTemplate):
"""Our own page template."""
def __init__(
self,
document,
2020-10-21 22:56:25 +02:00
context=None,
2020-09-26 16:19:37 +02:00
pagesbookmarks={},
author=None,
title=None,
subject=None,
margins=(0, 0, 0, 0), # additional margins in mm (left,top,right, bottom)
server_name="",
footer_template=DEFAULT_PDF_FOOTER_TEMPLATE,
filigranne=None,
preferences=None, # dictionnary with preferences, required
):
"""Initialise our page template."""
self.preferences = preferences
self.pagesbookmarks = pagesbookmarks
self.pdfmeta_author = author
self.pdfmeta_title = title
self.pdfmeta_subject = subject
self.server_name = server_name
self.filigranne = filigranne
self.footer_template = footer_template
2020-10-21 22:56:25 +02:00
self.with_page_background = self.preferences["bul_pdf_with_background"]
self.background_image_filename = None
2020-09-26 16:19:37 +02:00
# Our doc is made of a single frame
left, top, right, bottom = [float(x) for x in margins]
content = Frame(
10.0 * mm + left * mm,
13.0 * mm + bottom * mm,
document.pagesize[0] - 20.0 * mm - left * mm - right * mm,
document.pagesize[1] - 18.0 * mm - top * mm - bottom * mm,
)
PageTemplate.__init__(self, "ScolarsPageTemplate", [content])
self.logo = None
2020-10-21 22:56:25 +02:00
# XXX COPIED from sco_pvpdf, to be refactored (no time now)
# Search background in dept specific dir, then in global config dir
for image_dir in (
SCODOC_LOGOS_DIR + "/logos_" + context.DeptId() + "/",
SCODOC_LOGOS_DIR + "/", # global logos
):
for suffix in LOGOS_IMAGES_ALLOWED_TYPES:
fn = image_dir + "/bul_pdf_background" + "." + suffix
if not self.background_image_filename and os.path.exists(fn):
self.background_image_filename = fn
# Also try to use PV background
fn = image_dir + "/letter_background" + "." + suffix
if not self.background_image_filename and os.path.exists(fn):
self.background_image_filename = fn
2020-09-26 16:19:37 +02:00
def beforeDrawPage(self, canvas, doc):
2020-10-21 22:56:25 +02:00
"""Draws (optional) background, logo and contribution message on each page.
2020-09-26 16:19:37 +02:00
day : Day of the month as a decimal number [01,31]
month : Month as a decimal number [01,12].
year : Year without century as a decimal number [00,99].
Year : Year with century as a decimal number.
hour : Hour (24-hour clock) as a decimal number [00,23].
minute: Minute as a decimal number [00,59].
server_url: URL du serveur ScoDoc
2020-09-26 16:19:37 +02:00
"""
canvas.saveState()
2020-10-21 22:56:25 +02:00
# ---- Background image
if self.background_image_filename and self.with_page_background:
canvas.drawImage(
self.background_image_filename, 0, 0, doc.pagesize[0], doc.pagesize[1]
)
# ---- Logo: a small image, positionned at top left of the page
2020-09-26 16:19:37 +02:00
if self.logo is not None:
# draws the logo if it exists
((width, height), image) = self.logo
canvas.drawImage(image, inch, doc.pagesize[1] - inch, width, height)
# ---- Filigranne (texte en diagonal en haut a gauche de chaque page)
if self.filigranne:
if type(self.filigranne) == StringType:
filigranne = self.filigranne # same for all pages
else:
filigranne = self.filigranne.get(doc.page, None)
if filigranne:
canvas.saveState()
canvas.translate(9 * cm, 27.6 * cm)
canvas.rotate(30)
canvas.scale(4.5, 4.5)
canvas.setFillColorRGB(1.0, 0.65, 0.65)
canvas.drawRightString(0, 0, SU(filigranne))
canvas.restoreState()
# ---- Add some meta data and bookmarks
if self.pdfmeta_author:
canvas.setAuthor(SU(self.pdfmeta_author))
if self.pdfmeta_title:
canvas.setTitle(SU(self.pdfmeta_title))
if self.pdfmeta_subject:
canvas.setSubject(SU(self.pdfmeta_subject))
bm = self.pagesbookmarks.get(doc.page, None)
if bm != None:
key = bm
txt = SU(bm)
canvas.bookmarkPage(key)
canvas.addOutlineEntry(txt, bm)
# ---- Footer
canvas.setFont(
self.preferences["SCOLAR_FONT"], self.preferences["SCOLAR_FONT_SIZE_FOOT"]
)
d = _makeTimeDict()
d["scodoc_name"] = VERSION.SCONAME
d["server_url"] = self.server_name
footer_str = SU(self.footer_template % d)
canvas.drawString(
self.preferences["pdf_footer_x"] * mm,
self.preferences["pdf_footer_y"] * mm,
footer_str,
)
canvas.restoreState()
def _makeTimeDict():
# ... suboptimal but we don't care
return {
"day": time.strftime("%d"),
"month": time.strftime("%m"),
"year": time.strftime("%y"),
"Year": time.strftime("%Y"),
"hour": time.strftime("%H"),
"minute": time.strftime("%M"),
}
def pdf_basic_page(
2020-10-21 22:56:25 +02:00
objects, title="", preferences=None, context=None
2020-09-26 16:19:37 +02:00
): # used by gen_table.make_page()
"""Simple convenience fonction: build a page from a list of platypus objects,
adding a title if specified.
"""
StyleSheet = styles.getSampleStyleSheet()
report = cStringIO.StringIO() # in-memory document, no disk file
document = BaseDocTemplate(report)
document.addPageTemplates(
ScolarsPageTemplate(
document,
2020-10-21 22:56:25 +02:00
context,
2020-09-26 16:19:37 +02:00
title=title,
author="%s %s (E. Viennet)" % (SCONAME, SCOVERSION),
footer_template="Edité par %(scodoc_name)s le %(day)s/%(month)s/%(year)s à %(hour)sh%(minute)s",
preferences=preferences,
)
)
if title:
head = Paragraph(SU(title), StyleSheet["Heading3"])
objects = [head] + objects
document.build(objects)
data = report.getvalue()
return data
# Gestion du lock pdf
import threading, time, Queue, thread
class PDFLock:
def __init__(self, timeout=15):
self.Q = Queue.Queue(1)
self.timeout = timeout
self.current_thread = None
self.nref = 0
def release(self):
"Release lock. Raise Empty if not acquired first"
if self.current_thread == thread.get_ident():
self.nref -= 1
if self.nref == 0:
log("PDFLock: release from %s" % self.current_thread)
self.current_thread = None
self.Q.get(False)
return
else:
self.Q.get(False)
def acquire(self):
"Acquire lock. Raise ScoGenError if can't lock after timeout."
if self.current_thread == thread.get_ident():
self.nref += 1
return # deja lock pour ce thread
try:
self.Q.put(1, True, self.timeout)
except Queue.Full:
raise ScoGenError(msg="Traitement PDF occupé: ré-essayez")
self.current_thread = thread.get_ident()
self.nref = 1
log("PDFLock: granted to %s" % self.current_thread)
PDFLOCK = PDFLock()