X-Git-Url: https://git.mdrn.pl/redakcja.git/blobdiff_plain/193b26891b312d8ab98457e1d29b54c0ffd94712..cbb14d2ed3eb0b72a6c3c68ab93847923e6142f4:/apps/wiki/views.py diff --git a/apps/wiki/views.py b/apps/wiki/views.py index 319cdbad..6f09ddd6 100644 --- a/apps/wiki/views.py +++ b/apps/wiki/views.py @@ -1,5 +1,4 @@ import os -import functools import logging logger = logging.getLogger("fnp.wiki") @@ -11,12 +10,15 @@ from django.core.urlresolvers import reverse from wiki.helpers import (JSONResponse, JSONFormInvalid, JSONServerError, ajax_require_permission, recursive_groupby) from django import http +from django.shortcuts import get_object_or_404, redirect -from wiki.models import getstorage, DocumentNotFound, normalize_name, split_name, join_name, Theme -from wiki.forms import DocumentTextSaveForm, DocumentTagForm, DocumentCreateForm, DocumentsUploadForm +from wiki.models import Book, Theme +from wiki.forms import DocumentTextSaveForm, DocumentTextRevertForm, DocumentTagForm, DocumentCreateForm, DocumentsUploadForm from datetime import datetime from django.utils.encoding import smart_unicode from django.utils.translation import ugettext_lazy as _ +from django.utils.decorators import decorator_from_middleware +from django.middleware.gzip import GZipMiddleware # @@ -24,64 +26,45 @@ from django.utils.translation import ugettext_lazy as _ # from django.views.decorators.cache import never_cache -import wlapi import nice_diff import operator MAX_LAST_DOCS = 10 -def normalized_name(view): - - @functools.wraps(view) - def decorated(request, name, *args): - normalized = normalize_name(name) - logger.debug('View check %r -> %r', name, normalized) - - if normalized != name: - return http.HttpResponseRedirect( - reverse('wiki_' + view.__name__, kwargs={'name': normalized})) - - return view(request, name, *args) - - return decorated - - @never_cache def document_list(request): return direct_to_template(request, 'wiki/document_list.html', extra_context={ - 'docs': getstorage().all(), - 'last_docs': sorted(request.session.get("wiki_last_docs", {}).items(), - key=operator.itemgetter(1), reverse=True), + 'books': Book.objects.all(), + 'last_books': sorted(request.session.get("wiki_last_books", {}).items(), + key=lambda x: x[1]['time'], reverse=True), }) @never_cache -@normalized_name -def editor(request, name, template_name='wiki/document_details.html'): - storage = getstorage() - +def editor(request, slug, template_name='wiki/document_details.html'): try: - document = storage.get(name) - except DocumentNotFound: - return http.HttpResponseRedirect(reverse("wiki_create_missing", args=[name])) + book = Book.objects.get(slug=slug) + except Book.DoesNotExist: + return http.HttpResponseRedirect(reverse("wiki_create_missing", args=[slug])) access_time = datetime.now() - last_documents = request.session.get("wiki_last_docs", {}) - last_documents[name] = access_time + last_books = request.session.get("wiki_last_books", {}) + last_books[slug] = { + 'time': access_time, + 'title': book.title, + } - if len(last_documents) > MAX_LAST_DOCS: - oldest_key = min(last_documents, key=last_documents.__getitem__) - del last_documents[oldest_key] - request.session['wiki_last_docs'] = last_documents + if len(last_books) > MAX_LAST_DOCS: + oldest_key = min(last_books, key=operator.itemgetter('time')) + del last_books[oldest_key] + request.session['wiki_last_books'] = last_books return direct_to_template(request, template_name, extra_context={ - 'document': document, - 'document_name': document.name, - 'document_info': document.info, - 'document_meta': document.meta, + 'book': book, 'forms': { "text_save": DocumentTextSaveForm(prefix="textsave"), + "text_revert": DocumentTextRevertForm(prefix="textrevert"), "add_tag": DocumentTagForm(prefix="addtag"), }, 'REDMINE_URL': settings.REDMINE_URL, @@ -89,95 +72,105 @@ def editor(request, name, template_name='wiki/document_details.html'): @require_GET -@normalized_name -def editor_readonly(request, name, template_name='wiki/document_details_readonly.html'): - name = normalize_name(name) - storage = getstorage() - +def editor_readonly(request, slug, template_name='wiki/document_details_readonly.html'): try: + book = Book.objects.get(slug=slug) revision = request.GET['revision'] - document = storage.get(name, revision) - except (KeyError, DocumentNotFound): + except KeyError: raise http.Http404 access_time = datetime.now() - last_documents = request.session.get("wiki_last_docs", {}) - last_documents[name] = access_time + last_books = request.session.get("wiki_last_books", {}) + last_books[slug] = { + 'time': access_time, + 'title': book.title, + } - if len(last_documents) > MAX_LAST_DOCS: - oldest_key = min(last_documents, key=last_documents.__getitem__) - del last_documents[oldest_key] - request.session['wiki_last_docs'] = last_documents + if len(last_books) > MAX_LAST_DOCS: + oldest_key = min(last_books, key=operator.itemgetter('time')) + del last_books[oldest_key] + request.session['wiki_last_books'] = last_books return direct_to_template(request, template_name, extra_context={ - 'document': document, - 'document_name': document.name, - 'document_info': dict(document.info(), readonly=True), - 'document_meta': document.meta, + 'book': book, + 'revision': revision, + 'readonly': True, 'REDMINE_URL': settings.REDMINE_URL, }) -@normalized_name -def create_missing(request, name): - storage = getstorage() +def create_missing(request, slug): + slug = slug.replace(' ', '-') if request.method == "POST": form = DocumentCreateForm(request.POST, request.FILES) if form.is_valid(): - doc = storage.create_document( - name=form.cleaned_data['id'], + + if request.user.is_authenticated(): + creator = request.user + else: + creator = None + book = Book.create(creator=creator, + slug=form.cleaned_data['slug'], + title=form.cleaned_data['title'], text=form.cleaned_data['text'], ) - return http.HttpResponseRedirect(reverse("wiki_editor", args=[doc.name])) + return http.HttpResponseRedirect(reverse("wiki_editor", args=[book.slug])) else: form = DocumentCreateForm(initial={ - "id": name.replace(" ", "_"), - "title": name.title(), + "slug": slug, + "title": slug.replace('-', ' ').title(), }) return direct_to_template(request, "wiki/document_create_missing.html", extra_context={ - "document_name": name, + "slug": slug, "form": form, }) def upload(request): - storage = getstorage() - if request.method == "POST": form = DocumentsUploadForm(request.POST, request.FILES) if form.is_valid(): + import slughifi + + if request.user.is_authenticated(): + creator = request.user + else: + creator = None + zip = form.cleaned_data['zip'] skipped_list = [] ok_list = [] error_list = [] - titles = {} - existing = storage.all() + slugs = {} + existing = [book.slug for book in Book.objects.all()] for filename in zip.namelist(): if filename[-1] == '/': continue - title = normalize_name(os.path.basename(filename)[:-4]) - if not (title and filename.endswith('.xml')): + title = os.path.basename(filename)[:-4] + slug = slughifi(title) + if not (slug and filename.endswith('.xml')): skipped_list.append(filename) - elif title in titles: - error_list.append((filename, title, _('Title already used for %s' % titles[title]))) - elif title in existing: - error_list.append((filename, title, _('Title already used in repository.'))) + elif slug in slugs: + error_list.append((filename, slug, _('Slug already used for %s' % slugs[slug]))) + elif slug in existing: + error_list.append((filename, slug, _('Slug already used in repository.'))) else: try: zip.read(filename).decode('utf-8') # test read - ok_list.append((filename, title)) + ok_list.append((filename, slug, title)) except UnicodeDecodeError: error_list.append((filename, title, _('File should be UTF-8 encoded.'))) - titles[title] = filename + slugs[slug] = filename if not error_list: - for filename, title in ok_list: - storage.create_document( - name=title, - text=zip.read(filename).decode('utf-8') + for filename, slug, title in ok_list: + Book.create(creator=creator, + slug=slug, + title=title, + text=zip.read(filename).decode('utf-8'), ) return direct_to_template(request, "wiki/document_upload.html", extra_context={ @@ -186,12 +179,6 @@ def upload(request): "skipped_list": skipped_list, "error_list": error_list, }) - #doc = storage.create_document( - # name=base, - # text=form.cleaned_data['text'], - - - return http.HttpResponse('\n'.join(yeslist) + '\n\n' + '\n'.join(nolist)) else: form = DocumentsUploadForm() @@ -201,74 +188,79 @@ def upload(request): @never_cache -@normalized_name -def text(request, name): - storage = getstorage() +@decorator_from_middleware(GZipMiddleware) +def text(request, slug): + doc = get_object_or_404(Book, slug=slug).doc if request.method == 'POST': form = DocumentTextSaveForm(request.POST, prefix="textsave") if form.is_valid(): - revision = form.cleaned_data['parent_revision'] - document = storage.get_or_404(name, revision) - document.text = form.cleaned_data['text'] - comment = form.cleaned_data['comment'] - if form.cleaned_data['stage_completed']: - comment += '\n#stage-finished: %s\n' % form.cleaned_data['stage_completed'] + # TODO: + # - stage completion should be stored (as a relation) + if request.user.is_authenticated(): - author_name = request.user - author_email = request.user.email + author = request.user else: - author_name = form.cleaned_data['author_name'] - author_email = form.cleaned_data['author_email'] - author = "%s <%s>" % (author_name, author_email) - storage.put(document, author=author, comment=comment, parent=revision) - document = storage.get(name) + author = None + text = form.cleaned_data['text'] + parent_revision = form.cleaned_data['parent_revision'] + parent = doc.at_revision(parent_revision) + doc.commit(author=author, + text=text, + parent=parent, + description=form.cleaned_data['comment'], + ) + revision = doc.revision() return JSONResponse({ - 'text': document.plain_text if revision != document.revision else None, - 'meta': document.meta(), - 'revision': document.revision, + 'text': doc.materialize() if parent_revision != revision else None, + 'meta': {}, + 'revision': revision, }) else: return JSONFormInvalid(form) else: revision = request.GET.get("revision", None) - + try: - try: - revision = revision and int(revision) - logger.info("Fetching %s", revision) - document = storage.get(name, revision) - except ValueError: - # treat as a tag - logger.info("Fetching tag %s", revision) - document = storage.get_by_tag(name, revision) - except DocumentNotFound: - raise http.Http404 + revision = int(revision) + except ValueError: + revision = None return JSONResponse({ - 'text': document.plain_text, - 'meta': document.meta(), - 'revision': document.revision, + 'text': doc.at_revision(revision).materialize(), + 'meta': {}, + 'revision': revision if revision else doc.revision(), }) @never_cache -@normalized_name @require_POST -def revert(request, name): - storage = getstorage() - revision = request.POST['target_revision'] +def revert(request, slug): + form = DocumentTextRevertForm(request.POST, prefix="textrevert") + if form.is_valid(): + doc = get_object_or_404(Book, slug=slug).doc + revision = form.cleaned_data['revision'] - try: - document = storage.revert(name, revision) + comment = form.cleaned_data['comment'] + comment += "\n#revert to %s" % revision + + if request.user.is_authenticated(): + author = request.user + else: + author = None + + before = doc.revision() + logger.info("Reverting %s to %s", slug, revision) + doc.at_revision(revision).revert(author=author, description=comment) return JSONResponse({ - 'text': document.plain_text if revision != document.revision else None, - 'meta': document.meta(), - 'revision': document.revision, + 'text': doc.materialize() if before != doc.revision() else None, + 'meta': {}, + 'revision': doc.revision(), }) - except DocumentNotFound: - raise http.Http404 + else: + return JSONFormInvalid(form) + @never_cache def gallery(request, directory): @@ -298,10 +290,7 @@ def gallery(request, directory): @never_cache -@normalized_name -def diff(request, name): - storage = getstorage() - +def diff(request, slug): revA = int(request.GET.get('from', 0)) revB = int(request.GET.get('to', 0)) @@ -311,22 +300,48 @@ def diff(request, name): if revB == 0: revB = None - docA = storage.get_or_404(name, int(revA)) - docB = storage.get_or_404(name, int(revB)) + doc = get_object_or_404(Book, slug=slug).doc + docA = doc.at_revision(revA).materialize() + docB = doc.at_revision(revB).materialize() - return http.HttpResponse(nice_diff.html_diff_table(docA.plain_text.splitlines(), - docB.plain_text.splitlines(), context=3)) + return http.HttpResponse(nice_diff.html_diff_table(docA.splitlines(), + docB.splitlines(), context=3)) @never_cache -@normalized_name -def history(request, name): - storage = getstorage() +def revision(request, slug): + book = get_object_or_404(Book, slug=slug) + return http.HttpResponse(str(book.doc.revision())) + +@never_cache +def history(request, slug): # TODO: pagination - changesets = list(storage.history(name)) + book = get_object_or_404(Book, slug=slug) + rev = book.doc.revision() + changes = [] + for change in book.doc.history().order_by('-created_at'): + if change.author: + author = "%s %s <%s>" % ( + change.author.first_name, + change.author.last_name, + change.author.email) + else: + author = None + changes.append({ + "version": rev, + "description": change.description, + "author": author, + "date": change.created_at, + "tag": [], + }) + rev -= 1 + return JSONResponse(changes) + - return JSONResponse(changesets) + +""" +import wlapi @require_POST @@ -360,7 +375,7 @@ def publish(request, name): return JSONResponse({"result": api.publish_book(document)}) except wlapi.APICallException, e: return JSONServerError({"message": str(e)}) - +""" def themes(request): prefix = request.GET.get('q', '')