fixes 807: convert to lowercase scriptlet
[redakcja.git] / apps / wiki / views.py
index 9a156a4..4348cb8 100644 (file)
+import os
+import functools
+import logging
+logger = logging.getLogger("fnp.wiki")
+
+from django.conf import settings
+
 from django.views.generic.simple import direct_to_template
-from django.http import HttpResponse
-from django.utils import simplejson as json
+from django.views.decorators.http import require_POST, require_GET
+from django.core.urlresolvers import reverse
+from wiki.helpers import (JSONResponse, JSONFormInvalid, JSONServerError,
+                ajax_require_permission, recursive_groupby)
+from django import http
 
-from wiki.models import storage, Document, DocumentNotFound
-from wiki.forms import DocumentForm
+from wiki.models import getstorage, DocumentNotFound, normalize_name, split_name, join_name, Theme
+from wiki.forms import DocumentTextSaveForm, 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
 
 
-def document_list(request, template_name='wiki/document_list.html'):
-    return direct_to_template(request, template_name, extra_context={
-        'document_list': storage.all(),
+#
+# Quick hack around caching problems, TODO: use ETags
+#
+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),
     })
 
 
-def document_detail(request, name, template_name='wiki/document_details.html'):
+@never_cache
+@normalized_name
+def editor(request, name, template_name='wiki/document_details.html'):
+    storage = getstorage()
+
     try:
         document = storage.get(name)
     except DocumentNotFound:
-        document = Document(storage, name=name, text='')
-    
+        return http.HttpResponseRedirect(reverse("wiki_create_missing", args=[name]))
+
+    access_time = datetime.now()
+    last_documents = request.session.get("wiki_last_docs", {})
+    last_documents[name] = access_time
+
+    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
+
+    return direct_to_template(request, template_name, extra_context={
+        'document': document,
+        'document_name': document.name,
+        'document_info': document.info,
+        'document_meta': document.meta,
+        'forms': {
+            "text_save": DocumentTextSaveForm(prefix="textsave"),
+            "add_tag": DocumentTagForm(prefix="addtag"),
+        },
+        'REDMINE_URL': settings.REDMINE_URL,
+    })
+
+
+@require_GET
+@normalized_name
+def editor_readonly(request, name, template_name='wiki/document_details_readonly.html'):
+    name = normalize_name(name)
+    storage = getstorage()
+
+    try:
+        revision = request.GET['revision']
+        document = storage.get(name, revision)
+    except (KeyError, DocumentNotFound):
+        raise http.Http404
+
+    access_time = datetime.now()
+    last_documents = request.session.get("wiki_last_docs", {})
+    last_documents[name] = access_time
+
+    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
+
+    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,
+        'REDMINE_URL': settings.REDMINE_URL,
+    })
+
+
+@normalized_name
+def create_missing(request, name):
+    storage = getstorage()
+
+    if request.method == "POST":
+        form = DocumentCreateForm(request.POST, request.FILES)
+        if form.is_valid():
+            doc = storage.create_document(
+                name=form.cleaned_data['id'],
+                text=form.cleaned_data['text'],
+            )
+
+            return http.HttpResponseRedirect(reverse("wiki_editor", args=[doc.name]))
+    else:
+        form = DocumentCreateForm(initial={
+                "id": name.replace(" ", "_"),
+                "title": name.title(),
+        })
+
+    return direct_to_template(request, "wiki/document_create_missing.html", extra_context={
+        "document_name": name,
+        "form": form,
+    })
+
+
+def upload(request):
+    storage = getstorage()
+
+    if request.method == "POST":
+        form = DocumentsUploadForm(request.POST, request.FILES)
+        if form.is_valid():
+            zip = form.cleaned_data['zip']
+            skipped_list = []
+            ok_list = []
+            error_list = []
+            titles = {}
+            existing = storage.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')):
+                    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.')))
+                else:
+                    try:
+                        zip.read(filename).decode('utf-8') # test read
+                        ok_list.append((filename, title))
+                    except UnicodeDecodeError:
+                        error_list.append((filename, title, _('File should be UTF-8 encoded.')))
+                    titles[title] = filename
+
+            if not error_list:
+                for filename, title in ok_list:
+                    storage.create_document(
+                        name=title,
+                        text=zip.read(filename).decode('utf-8')
+                    )
+
+            return direct_to_template(request, "wiki/document_upload.html", extra_context={
+                "form": form,
+                "ok_list": ok_list,
+                "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()
+
+    return direct_to_template(request, "wiki/document_upload.html", extra_context={
+        "form": form,
+    })
+
+
+@never_cache
+@normalized_name
+@decorator_from_middleware(GZipMiddleware)
+def text(request, name):
+    storage = getstorage()
 
     if request.method == 'POST':
-        form = DocumentForm(request.POST, instance=document)
+        form = DocumentTextSaveForm(request.POST, prefix="textsave")
         if form.is_valid():
-            document = form.save()
-            return HttpResponse(json.dumps({'text': document.plain_text(), 'meta': document.meta(), 'revision': document.revision()}))
+            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']         
+            if request.user.is_authenticated():
+                author_name = request.user
+                author_email = request.user.email
+            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)          
+            return JSONResponse({
+                'text': document.plain_text if revision != document.revision else None,
+                'meta': document.meta(),
+                'revision': document.revision,
+            })
         else:
-            return HttpResponse(json.dumps({'errors': form.errors}))
+            return JSONFormInvalid(form)
     else:
-        form = DocumentForm(instance=document)
-    
-    return direct_to_template(request, template_name, extra_context={
-        'document': document,
-        'form': form,
-    })
+        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
+
+        return JSONResponse({
+            'text': document.plain_text,
+            'meta': document.meta(),
+            'revision': document.revision,
+        })
+
+
+@never_cache
+@normalized_name
+@require_POST
+def revert(request, name):
+    storage = getstorage()
+    revision = request.POST['target_revision']
+
+    try:
+        document = storage.revert(name, revision)
+
+        return JSONResponse({
+            'text': document.plain_text if revision != document.revision else None,
+            'meta': document.meta(),
+            'revision': document.revision,
+        })
+    except DocumentNotFound:
+        raise http.Http404
+
+@never_cache
+def gallery(request, directory):
+    try:
+        base_url = ''.join((
+                        smart_unicode(settings.MEDIA_URL),
+                        smart_unicode(settings.FILEBROWSER_DIRECTORY),
+                        smart_unicode(directory)))
+
+        base_dir = os.path.join(
+                    smart_unicode(settings.MEDIA_ROOT),
+                    smart_unicode(settings.FILEBROWSER_DIRECTORY),
+                    smart_unicode(directory))
+
+        def map_to_url(filename):
+            return "%s/%s" % (base_url, smart_unicode(filename))
+
+        def is_image(filename):
+            return os.path.splitext(f)[1].lower() in (u'.jpg', u'.jpeg', u'.png')
+
+        images = [map_to_url(f) for f in map(smart_unicode, os.listdir(base_dir)) if is_image(f)]
+        images.sort()
+        return JSONResponse(images)
+    except (IndexError, OSError):
+        logger.exception("Unable to fetch gallery")
+        raise http.Http404
+
+
+@never_cache
+@normalized_name
+def diff(request, name):
+    storage = getstorage()
+
+    revA = int(request.GET.get('from', 0))
+    revB = int(request.GET.get('to', 0))
+
+    if revA > revB:
+        revA, revB = revB, revA
+
+    if revB == 0:
+        revB = None
+
+    docA = storage.get_or_404(name, int(revA))
+    docB = storage.get_or_404(name, int(revB))
+
+    return http.HttpResponse(nice_diff.html_diff_table(docA.plain_text.splitlines(),
+                                         docB.plain_text.splitlines(), context=3))
+
+
+@never_cache
+@normalized_name
+def history(request, name):
+    storage = getstorage()
+
+    # TODO: pagination
+    changesets = list(storage.history(name))
+
+    return JSONResponse(changesets)
+
+
+@require_POST
+@ajax_require_permission('wiki.can_change_tags')
+def add_tag(request, name):
+    name = normalize_name(name)
+    storage = getstorage()
+
+    form = DocumentTagForm(request.POST, prefix="addtag")
+    if form.is_valid():
+        doc = storage.get_or_404(form.cleaned_data['id'])
+        doc.add_tag(tag=form.cleaned_data['tag'],
+                    revision=form.cleaned_data['revision'],
+                    author=request.user.username)
+        return JSONResponse({"message": _("Tag added")})
+    else:
+        return JSONFormInvalid(form)
+
+
+@require_POST
+@ajax_require_permission('wiki.can_publish')
+def publish(request, name):
+    name = normalize_name(name)
+
+    storage = getstorage()
+    document = storage.get_by_tag(name, "ready_to_publish")
+
+    api = wlapi.WLAPI(**settings.WL_API_CONFIG)
+
+    try:
+        return JSONResponse({"result": api.publish_book(document)})
+    except wlapi.APICallException, e:
+        return JSONServerError({"message": str(e)})
+
+
+def themes(request):
+    prefix = request.GET.get('q', '')
+    return http.HttpResponse('\n'.join([str(t) for t in Theme.objects.filter(name__istartswith=prefix)]))