发布于 2016-11-06 10:43:32 | 209 次阅读 | 评论: 0 | 来源: 网络整理
Django 中自带了 sitemap框架,用来生成 xml 文件
sitemap 很重要,可以用来通知搜索引擎页面的地址,页面的重要性,帮助站点得到比较好的收录。
开启sitemap功能的步骤
settings.py 文件中 django.contrib.sitemaps 和 django.contrib.sites 要在 INSTALL_APPS 中
INSTALLED_APPS = (
'django.contrib.admin',
'django.contrib.auth',
'django.contrib.contenttypes',
'django.contrib.sessions',
'django.contrib.messages',
'django.contrib.staticfiles',
'django.contrib.sites',
'django.contrib.sitemaps',
'django.contrib.redirects',
#####
#othther apps
#####
)
Django 1.7 及以前版本:
TEMPLATE_LOADERS 中要加入 'django.template.loaders.app_directories.Loader',像这样:
TEMPLATE_LOADERS = (
'django.template.loaders.filesystem.Loader',
'django.template.loaders.app_directories.Loader',
)
Django 1.8 及以上版本新加入了 TEMPLATES 设置,其中 APP_DIRS 要为 True,比如:
# NOTICE: code for Django 1.8, not work on Django 1.7 and below
TEMPLATES = [
{
'BACKEND': 'django.template.backends.django.DjangoTemplates',
'DIRS': [
os.path.join(BASE_DIR,'templates').replace('\', '/'),
],
'APP_DIRS': True,
},
]
然后在 urls.py 中如下配置:
from django.conf.urls import url
from django.contrib.sitemaps import GenericSitemap
from django.contrib.sitemaps.views import sitemap
from blog.models import Entry
sitemaps = {
'blog': GenericSitemap({'queryset': Entry.objects.all(), 'date_field': 'pub_date'}, priority=0.6),
# 如果还要加其它的可以模仿上面的
}
urlpatterns = [
# some generic view using info_dict
# ...
# the sitemap
url(r'^sitemap.xml$', sitemap, {'sitemaps': sitemaps},
name='django.contrib.sitemaps.views.sitemap'),
]
但是这样生成的 sitemap,如果网站内容太多就很慢,很耗费资源,可以采用分页的功能:
from django.conf.urls import url
from django.contrib.sitemaps import GenericSitemap
from django.contrib.sitemaps.views import sitemap
from blog.models import Entry
from django.contrib.sitemaps import views as sitemaps_views
from django.views.decorators.cache import cache_page
sitemaps = {
'blog': GenericSitemap({'queryset': Entry.objects.all(), 'date_field': 'pub_date'}, priority=0.6),
# 如果还要加其它的可以模仿上面的
}
urlpatterns = [
url(r'^sitemap.xml$',
cache_page(86400)(sitemaps_views.index),
{'sitemaps': sitemaps, 'sitemap_url_name': 'sitemaps'}),
url(r'^sitemap-(?P<section>.+).xml$',
cache_page(86400)(sitemaps_views.sitemap),
{'sitemaps': sitemaps}, name='sitemaps'),
]
这样就可以看到类似如下的 sitemap,如果本地测试访问 http://localhost:8000/sitemap.xml
<?xml version="1.0" encoding="UTF-8"?>
<sitemapindex xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=2</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=3</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=4</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=5</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=6</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=7</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=8</loc></sitemap>
<sitemap><loc>http://www.phperz.com/sitemap-tutorials.xml?p=9</loc></sitemap>
</sitemapindex>
查看了下分页是实现了,但是全部显示成了 ?p=页面数,而且在百度站长平台上测试,发现这样的sitemap百度报错,于是看了下 Django的源代码:
在这里 https://github.com/django/django/blob/1.7.7/django/contrib/sitemaps/views.py
于是对源代码作了修改,变成了本站的sitemap的样子,比 ?p=2 这样更优雅
引入 下面这个 比如是 sitemap_views.py
import warnings
from functools import wraps
from django.contrib.sites.models import get_current_site
from django.core import urlresolvers
from django.core.paginator import EmptyPage, PageNotAnInteger
from django.http import Http404
from django.template.response import TemplateResponse
from django.utils import six
def x_robots_tag(func):
@wraps(func)
def inner(request, *args, **kwargs):
response = func(request, *args, **kwargs)
response['X-Robots-Tag'] = 'noindex, noodp, noarchive'
return response
return inner
@x_robots_tag
def index(request, sitemaps,
template_name='sitemap_index.xml', content_type='application/xml',
sitemap_url_name='django.contrib.sitemaps.views.sitemap',
mimetype=None):
if mimetype:
warnings.warn("The mimetype keyword argument is deprecated, use "
"content_type instead", DeprecationWarning, stacklevel=2)
content_type = mimetype
req_protocol = 'https' if request.is_secure() else 'http'
req_site = get_current_site(request)
sites = []
for section, site in sitemaps.items():
if callable(site):
site = site()
protocol = req_protocol if site.protocol is None else site.protocol
for page in range(1, site.paginator.num_pages + 1):
sitemap_url = urlresolvers.reverse(
sitemap_url_name, kwargs={'section': section, 'page': page})
absolute_url = '%s://%s%s' % (protocol, req_site.domain, sitemap_url)
sites.append(absolute_url)
return TemplateResponse(request, template_name, {'sitemaps': sites},
content_type=content_type)
@x_robots_tag
def sitemap(request, sitemaps, section=None, page=1,
template_name='sitemap.xml', content_type='application/xml',
mimetype=None):
if mimetype:
warnings.warn("The mimetype keyword argument is deprecated, use "
"content_type instead", DeprecationWarning, stacklevel=2)
content_type = mimetype
req_protocol = 'https' if request.is_secure() else 'http'
req_site = get_current_site(request)
if section is not None:
if section not in sitemaps:
raise Http404("No sitemap available for section: %r" % section)
maps = [sitemaps[section]]
else:
maps = list(six.itervalues(sitemaps))
urls = []
for site in maps:
try:
if callable(site):
site = site()
urls.extend(site.get_urls(page=page, site=req_site,
protocol=req_protocol))
except EmptyPage:
raise Http404("Page %s empty" % page)
except PageNotAnInteger:
raise Http404("No page '%s'" % page)
return TemplateResponse(request, template_name, {'urlset': urls},
content_type=content_type)
如果还是不懂,可以下载附件查看:zqxt_sitemap.zip
更多参考:
官方文档:https://docs.djangoproject.com/en/dev/ref/contrib/sitemaps/