You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1154 lines
27 KiB
1154 lines
27 KiB
<?php
|
|
// 公共函数文件
|
|
if (! function_exists('curl_request'))
|
|
{
|
|
function curl_request($api, $params = array(), $method = 'GET', $headers = array())
|
|
{
|
|
$curl = curl_init();
|
|
|
|
switch (strtoupper($method))
|
|
{
|
|
case 'GET' :
|
|
if (!empty($params))
|
|
{
|
|
$api .= (strpos($api, '?') ? '&' : '?') . http_build_query($params);
|
|
}
|
|
curl_setopt($curl, CURLOPT_HTTPGET, TRUE);
|
|
break;
|
|
case 'POST' :
|
|
curl_setopt($curl, CURLOPT_POST, TRUE);
|
|
curl_setopt($curl, CURLOPT_POSTFIELDS, $params);
|
|
|
|
break;
|
|
case 'PUT' :
|
|
curl_setopt($curl, CURLOPT_CUSTOMREQUEST, 'PUT');
|
|
curl_setopt($curl, CURLOPT_POSTFIELDS, $params);
|
|
break;
|
|
case 'DELETE' :
|
|
curl_setopt($curl, CURLOPT_CUSTOMREQUEST, 'DELETE');
|
|
curl_setopt($curl, CURLOPT_POSTFIELDS, $params);
|
|
break;
|
|
}
|
|
|
|
curl_setopt($curl, CURLOPT_URL, $api);
|
|
curl_setopt($curl, CURLOPT_RETURNTRANSFER, TRUE);
|
|
curl_setopt($curl, CURLOPT_SSL_VERIFYPEER, FALSE);
|
|
curl_setopt($curl, CURLOPT_HEADER, 0);
|
|
|
|
curl_setopt($curl, CURLOPT_HTTPHEADER, $headers);
|
|
|
|
$response = curl_exec($curl);
|
|
|
|
if ($response === FALSE)
|
|
{
|
|
$error = curl_error($curl);
|
|
curl_close($curl);
|
|
return FALSE;
|
|
}
|
|
else
|
|
{
|
|
// 解决windows 服务器 BOM 问题
|
|
$response = trim($response,chr(239).chr(187).chr(191));
|
|
$response = json_decode($response, true);
|
|
}
|
|
|
|
curl_close($curl);
|
|
|
|
return $response;
|
|
}
|
|
}
|
|
|
|
//获取数据
|
|
function dataList($modelname, $where = [], $size = 15, $page = 1)
|
|
{
|
|
$model = \DB::table($modelname);
|
|
|
|
$page = 1;$skip = 0;
|
|
if(isset($where['limit'])){$limit=explode(',',$where['limit']); $skip = $limit[0]; $size = $limit[1];}else{if(isset($where['row'])){$size = $where['row'];}} // 参数格式:$where['limit'] = '2,10';$where['row'] = 10;
|
|
|
|
//原生sql
|
|
if(isset($where['sql']))
|
|
{
|
|
$model = $model->whereRaw($where['sql']);
|
|
}
|
|
|
|
//排序
|
|
if(isset($where['orderby']))
|
|
{
|
|
$orderby = $where['orderby'];
|
|
|
|
if($orderby == 'rand()')
|
|
{
|
|
$model = $model->orderBy(\DB::raw('rand()'));
|
|
}
|
|
else
|
|
{
|
|
if(count($orderby) == count($orderby, 1))
|
|
{
|
|
$model = $model->orderBy($orderby[0], $orderby[1]);
|
|
}
|
|
else
|
|
{
|
|
foreach($orderby as $row)
|
|
{
|
|
$model = $model->orderBy($row[0], $row[1]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
$model = $model->orderBy('id', 'desc');
|
|
}
|
|
|
|
//要返回的字段
|
|
if(isset($where['field'])){$model = $model->select(\DB::raw($where['field']));}
|
|
|
|
//查询条件
|
|
$where = function ($query) use ($where) {
|
|
if(isset($where['expression']))
|
|
{
|
|
foreach($where['expression'] as $row)
|
|
{
|
|
$query->where($row[0], $row[1], $row[2]);
|
|
}
|
|
}
|
|
};
|
|
|
|
if(!empty($where)){$model = $model->where($where);}
|
|
|
|
if($skip==0){$skip = ($page-1)*$size;}
|
|
|
|
return object_to_array($model->skip($skip)->take($size)->get());
|
|
}
|
|
|
|
//pc前台栏目、标签、内容页面地址生成
|
|
function get_front_url($param='')
|
|
{
|
|
$url = '';
|
|
|
|
if($param['type'] == 'list')
|
|
{
|
|
//列表页
|
|
$url .= '/cat'.$param['catid'];
|
|
}
|
|
else if($param['type'] == 'content')
|
|
{
|
|
//内容页
|
|
$url .= '/p/'.$param['id'];
|
|
}
|
|
else if($param['type'] == 'tags')
|
|
{
|
|
//tags页面
|
|
$url .= '/tag'.$param['tagid'];
|
|
}
|
|
else if($param['type'] == 'page')
|
|
{
|
|
//单页面
|
|
$url .= '/page/'.$param['pagename'];
|
|
}
|
|
else if($param['type'] == 'search')
|
|
{
|
|
//搜索关键词页面
|
|
$url .= '/s'.$param['searchid'];
|
|
}
|
|
else if($param['type'] == 'goodslist')
|
|
{
|
|
//商品列表页
|
|
$url .= '/product'.$param['catid'];
|
|
}
|
|
else if($param['type'] == 'goodsdetail')
|
|
{
|
|
//商品内容页
|
|
$url .= '/goods/'.$param['id'];
|
|
}
|
|
|
|
return $url;
|
|
}
|
|
|
|
//wap前台栏目、标签、内容页面地址生成
|
|
function get_wap_front_url(array $param)
|
|
{
|
|
$url = '';
|
|
|
|
if($param['type'] == 'list')
|
|
{
|
|
//列表页
|
|
$url .= '/cat'.$param['catid'];
|
|
}
|
|
else if($param['type'] == 'content')
|
|
{
|
|
//内容页
|
|
$url .= '/p/'.$param['id'];
|
|
}
|
|
else if($param['type'] == 'tags')
|
|
{
|
|
//tags页面
|
|
$url .= '/tag'.$param['tagid'];
|
|
}
|
|
else if($param['type'] == 'page')
|
|
{
|
|
//单页面
|
|
$url .= '/page/'.$param['pagename'];
|
|
}
|
|
else if($param['type'] == 'search')
|
|
{
|
|
//tags页面
|
|
$url .= '/s'.$param['searchid'];
|
|
}
|
|
else if($param['type'] == 'goodslist')
|
|
{
|
|
//商品列表页
|
|
$url .= '/product'.$param['catid'];
|
|
}
|
|
else if($param['type'] == 'goodsdetail')
|
|
{
|
|
//商品内容页
|
|
$url .= '/goods/'.$param['id'];
|
|
}
|
|
|
|
return $url;
|
|
}
|
|
|
|
/**
|
|
* 获取文章列表
|
|
* @param int $tuijian=0 推荐等级
|
|
* @param int $typeid=0 分类
|
|
* @param int $image=1 是否存在图片
|
|
* @param int $row=10 需要返回的数量
|
|
* @param string $orderby='id desc' 排序,默认id降序,随机rand()
|
|
* @param string $limit='0,10' 如果存在$row,$limit就无效
|
|
* @return string
|
|
*/
|
|
function arclist(array $param)
|
|
{
|
|
$modelname = 'article';
|
|
if(isset($param['table'])){$modelname = $param['table'];}
|
|
|
|
$model = \DB::table($modelname);
|
|
|
|
$size = sysconfig('CMS_PAGESIZE');$page = 1;$skip = 0;
|
|
if(isset($param['limit'])){$limit=explode(',',$param['limit']); $skip = $limit[0]; $size = $limit[1];}else{if(isset($param['row'])){$size = $param['row'];}} // 参数格式:$param['limit'] = '2,10';$param['row'] = 10;
|
|
|
|
//查询条件
|
|
$where = function ($query) use ($param) {
|
|
if(isset($param['tuijian']))
|
|
{
|
|
if(is_array($param['tuijian']))
|
|
{
|
|
$query->where('tuijian', $param['tuijian'][0], $param['tuijian'][1]);
|
|
}
|
|
else
|
|
{
|
|
$query->where('tuijian', $param['tuijian']);
|
|
}
|
|
}
|
|
|
|
if(isset($param['expression']))
|
|
{
|
|
foreach($param['expression'] as $row)
|
|
{
|
|
$query->where($row[0], $row[1], $row[2]);
|
|
}
|
|
}
|
|
|
|
if(isset($param['typeid']))
|
|
{
|
|
$query->where('typeid', $param["typeid"]);
|
|
}
|
|
|
|
if(isset($param['image']))
|
|
{
|
|
$query->where('litpic', '<>', '');
|
|
}
|
|
};
|
|
|
|
if(!empty($where)){$model = $model->where($where);}
|
|
|
|
//原生sql
|
|
if(isset($param['sql']))
|
|
{
|
|
$model = $model->whereRaw($param['sql']);
|
|
}
|
|
|
|
//排序
|
|
if(isset($param['orderby']))
|
|
{
|
|
$orderby = $param['orderby'];
|
|
|
|
if($orderby == 'rand()')
|
|
{
|
|
$model = $model->orderBy(\DB::raw('rand()'));
|
|
}
|
|
else
|
|
{
|
|
if(count($orderby) == count($orderby, 1))
|
|
{
|
|
$model = $model->orderBy($orderby[0], $orderby[1]);
|
|
}
|
|
else
|
|
{
|
|
foreach($orderby as $row)
|
|
{
|
|
$model = $model->orderBy($row[0], $row[1]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
$model = $model->orderBy('id', 'desc');
|
|
}
|
|
|
|
//要返回的字段
|
|
if(isset($param['field'])){$model = $model->select(\DB::raw($param['field']));}
|
|
|
|
if($skip==0){$skip = ($page-1)*$size;}
|
|
|
|
return object_to_array($model->skip($skip)->take($size)->get());
|
|
}
|
|
|
|
/**
|
|
* 获取tag标签列表
|
|
* @param int $row=10 需要返回的数量,如果存在$limit,$row就无效
|
|
* @param string $orderby='id desc' 排序,默认id降序,随机rand()
|
|
* @param string $limit='0,10'
|
|
* @return string
|
|
*/
|
|
function tagslist($param="")
|
|
{
|
|
$tagindex = \DB::table("tagindex");
|
|
$orderby=$limit="";
|
|
if(isset($param['row'])){$tagindex = $tagindex->take($param['row']);}
|
|
if(isset($param['orderby'])){if($param['orderby']=='rand()'){$tagindex = $tagindex->orderBy(\DB::Raw('rand()'));}else{$tagindex = $tagindex->orderBy($param['orderby'][0],$param['orderby'][1]);}}else{$tagindex = $tagindex->orderBy('id','desc');}
|
|
|
|
return object_to_array($tagindex->get());
|
|
}
|
|
|
|
/**
|
|
* 获取友情链接
|
|
* @param string $orderby='id desc' 排序,默认id降序,随机rand()
|
|
* @param int||string $limit='0,10'
|
|
* @return string
|
|
*/
|
|
function flinklist($param="")
|
|
{
|
|
return \DB::table("friendlink")->orderBy('rank','desc')->take($param['row'])->get();
|
|
}
|
|
|
|
/**
|
|
* 获取文章上一篇,下一篇id
|
|
* @param $param['aid'] 当前文章id
|
|
* @param $param['typeid'] 当前文章typeid
|
|
* @param string $type 获取类型
|
|
* pre:上一篇 next:下一篇
|
|
* @return array
|
|
*/
|
|
function get_article_prenext(array $param)
|
|
{
|
|
$typeid = $res = '';
|
|
|
|
if(!empty($param["typeid"]))
|
|
{
|
|
$typeid = $param["typeid"];
|
|
}
|
|
else
|
|
{
|
|
$Article = DB::table("article")->select('typeid')->where('id', $param["aid"])->first();
|
|
$typeid = $Article["typeid"];
|
|
}
|
|
|
|
$res = DB::table("article")->select('id','typeid','title')->where('typeid', $typeid);
|
|
if($param["type"]=='pre')
|
|
{
|
|
$res = $res->where('id', '<', $param["aid"])->orderBy('id', 'desc');
|
|
}
|
|
elseif($param["type"]=='next')
|
|
{
|
|
$res = $res->where('id', '>', $param["aid"])->orderBy('id', 'asc');
|
|
}
|
|
|
|
return object_to_array($res->first(), 1);
|
|
}
|
|
|
|
/**
|
|
* 获取列表分页
|
|
* @param $param['pagenow'] 当前第几页
|
|
* @param $param['counts'] 总条数
|
|
* @param $param['pagesize'] 每页显示数量
|
|
* @param $param['catid'] 栏目id
|
|
* @param $param['offset'] 偏移量
|
|
* @return array
|
|
*/
|
|
function get_listnav(array $param)
|
|
{
|
|
$catid = $param["catid"];
|
|
$pagenow = $param["pagenow"];
|
|
$prepage = $nextpage = '';
|
|
$prepagenum = $pagenow-1;
|
|
$nextpagenum = $pagenow+1;
|
|
|
|
$counts=$param["counts"];
|
|
$totalpage=get_totalpage(array("counts"=>$counts,"pagesize"=>$param["pagesize"]));
|
|
|
|
if($totalpage<=1 && $counts>0)
|
|
{
|
|
return "<li><span class=\"pageinfo\">共1页/".$counts."条记录</span></li>";
|
|
}
|
|
if($counts == 0)
|
|
{
|
|
return "<li><span class=\"pageinfo\">共0页/".$counts."条记录</span></li>";
|
|
}
|
|
$maininfo = "<li><span class=\"pageinfo\">共".$totalpage."页".$counts."条</span></li>";
|
|
|
|
if(!empty($param["urltype"]))
|
|
{
|
|
$urltype = $param["urltype"];
|
|
}
|
|
else
|
|
{
|
|
$urltype = 'cat';
|
|
}
|
|
|
|
//获得上一页和下一页的链接
|
|
if($pagenow != 1)
|
|
{
|
|
if($pagenow == 2)
|
|
{
|
|
$prepage.="<li><a href='/".$urltype.$catid."'>上一页</a></li>";
|
|
}
|
|
else
|
|
{
|
|
$prepage.="<li><a href='/".$urltype.$catid."/$prepagenum'>上一页</a></li>";
|
|
}
|
|
|
|
$indexpage="<li><a href='/".$urltype.$catid."'>首页</a></li>";
|
|
}
|
|
else
|
|
{
|
|
$indexpage="<li><a>首页</a></li>";
|
|
}
|
|
if($pagenow!=$totalpage && $totalpage>1)
|
|
{
|
|
$nextpage.="<li><a href='/".$urltype.$catid."/$nextpagenum'>下一页</a></li>";
|
|
$endpage="<li><a href='/".$urltype.$catid."/$totalpage'>末页</a></li>";
|
|
}
|
|
else
|
|
{
|
|
$endpage="<li><a>末页</a></li>";
|
|
}
|
|
|
|
//获得数字链接
|
|
$listdd="";
|
|
if(!empty($param["offset"])){$offset=$param["offset"];}else{$offset=2;}
|
|
|
|
$minnum=$pagenow-$offset;
|
|
$maxnum=$pagenow+$offset;
|
|
|
|
if($minnum<1){$minnum=1;}
|
|
if($maxnum>$totalpage){$maxnum=$totalpage;}
|
|
|
|
for($minnum;$minnum<=$maxnum;$minnum++)
|
|
{
|
|
if($minnum==$pagenow)
|
|
{
|
|
$listdd.= "<li class=\"thisclass\"><a>$minnum</a></li>";
|
|
}
|
|
else
|
|
{
|
|
if($minnum==1)
|
|
{
|
|
$listdd.="<li><a href='/".$urltype.$catid."'>$minnum</a></li>";
|
|
}
|
|
else
|
|
{
|
|
$listdd.="<li><a href='/".$urltype.$catid."/$minnum'>$minnum</a></li>";
|
|
}
|
|
}
|
|
}
|
|
|
|
$plist = '';
|
|
$plist .= $indexpage; //首页链接
|
|
$plist .= $prepage; //上一页链接
|
|
$plist .= $listdd; //数字链接
|
|
$plist .= $nextpage; //下一页链接
|
|
$plist .= $endpage; //末页链接
|
|
$plist .= $maininfo;
|
|
|
|
return $plist;
|
|
}
|
|
|
|
/**
|
|
* 获取列表上一页、下一页
|
|
* @param $param['pagenow'] 当前第几页
|
|
* @param $param['counts'] 总条数
|
|
* @param $param['pagesize'] 每页显示数量
|
|
* @param $param['catid'] 栏目id
|
|
* @return array
|
|
*/
|
|
function get_prenext(array $param)
|
|
{
|
|
$counts=$param['counts'];
|
|
$pagenow=$param["pagenow"];
|
|
$prepage = $nextpage = '';
|
|
$prepagenum = $pagenow-1;
|
|
$nextpagenum = $pagenow+1;
|
|
$cat=$param['catid'];
|
|
|
|
if(!empty($param["urltype"]))
|
|
{
|
|
$urltype = $param["urltype"];
|
|
}
|
|
else
|
|
{
|
|
$urltype = 'cat';
|
|
}
|
|
|
|
$totalpage=get_totalpage(array("counts"=>$counts,"pagesize"=>$param["pagesize"]));
|
|
|
|
//获取上一页
|
|
if($pagenow == 1)
|
|
{
|
|
|
|
}
|
|
elseif($pagenow==2)
|
|
{
|
|
$prepage='<a class="prep" href="/'.$urltype.$cat.'">上一页</a> ';
|
|
}
|
|
else
|
|
{
|
|
$prepage='<a class="prep" href="/'.$urltype.$cat.'/'.$prepagenum.'">上一页</a> ';
|
|
}
|
|
|
|
//获取下一页
|
|
if($pagenow<$totalpage && $totalpage>1)
|
|
{
|
|
$nextpage='<a class="nextp" href="/'.$urltype.$cat.'/'.$nextpagenum.'">下一页</a>';
|
|
}
|
|
|
|
$plist = '';
|
|
$plist .= $indexpage; //首页链接
|
|
$plist .= $prepage; //上一页链接
|
|
$plist .= $nextpage; //下一页链接
|
|
|
|
return $plist;
|
|
}
|
|
|
|
/**
|
|
* 获取分页列表
|
|
* @access public
|
|
* @param string $list_len 列表宽度
|
|
* @param string $list_len 列表样式
|
|
* @return string
|
|
*/
|
|
function pagenav(array $param)
|
|
{
|
|
$prepage = $nextpage = '';
|
|
$prepagenum = $param["pagenow"]-1;
|
|
$nextpagenum = $param["pagenow"]+1;
|
|
|
|
if(!empty($param['tuijian'])){$map['tuijian']=$param['tuijian'];}
|
|
if(!empty($param['typeid'])){$map['typeid']=$param['typeid'];}
|
|
if(!empty($param['image'])){$map['litpic']=array('NEQ','');}
|
|
if(!empty($param['row'])){$limit="0,".$param['row'];}else{if(!empty($param['limit'])){$limit=$param['limit'];}else{$limit='0,8';}}
|
|
if(!empty($param['orderby'])){$orderby=$param['orderby'];}else{$orderby='id desc';}
|
|
|
|
return db("article")->field('body',true)->where($map)->order($orderby)->limit($limit)->select();
|
|
}
|
|
|
|
//根据总数与每页条数,获取总页数
|
|
function get_totalpage(array $param)
|
|
{
|
|
if(!empty($param['pagesize'] || $param['pagesize']==0)){$pagesize=$param["pagesize"];}else{$pagesize=CMS_PAGESIZE;}
|
|
$counts=$param["counts"];
|
|
|
|
//取总数据量除以每页数的余数
|
|
if($counts % $pagesize)
|
|
{
|
|
$totalpage = intval($counts/$pagesize) + 1; //如果有余数,则页数等于总数据量除以每页数的结果取整再加一,如果没有余数,则页数等于总数据量除以每页数的结果
|
|
}
|
|
else
|
|
{
|
|
$totalpage = $counts/$pagesize;
|
|
}
|
|
|
|
return $totalpage;
|
|
}
|
|
|
|
/**
|
|
* 获得当前的页面文件的url
|
|
* @access public
|
|
* @return string
|
|
*/
|
|
function GetCurUrl()
|
|
{
|
|
if(!empty($_SERVER['REQUEST_URI']))
|
|
{
|
|
$nowurl = $_SERVER['REQUEST_URI'];
|
|
$nowurls = explode('?', $nowurl);
|
|
$nowurl = $nowurls[0];
|
|
}
|
|
else
|
|
{
|
|
$nowurl = $_SERVER['PHP_SELF'];
|
|
}
|
|
return $nowurl;
|
|
}
|
|
|
|
/**
|
|
* 获取单页列表
|
|
* @param int $row=8 需要返回的数量
|
|
* @param string $orderby='id desc' 排序,默认id降序,随机rand()
|
|
* @param string $limit='0,8' 如果存在$row,$limit就无效
|
|
* @return string
|
|
*/
|
|
function pagelist($param="")
|
|
{
|
|
if(!empty($param['row'])){$limit="0,".$param['row'];}else{if(!empty($param['limit'])){$limit=$param['limit'];}else{$limit='0,8';}}
|
|
if(!empty($param['orderby'])){$orderby=$param['orderby'];}else{$orderby='id desc';}
|
|
|
|
return db("page")->field('body',true)->order($orderby)->limit($limit)->select();
|
|
}
|
|
|
|
/**
|
|
* 截取中文字符串
|
|
* @param string $string 中文字符串
|
|
* @param int $sublen 截取长度
|
|
* @param int $start 开始长度 默认0
|
|
* @param string $code 编码方式 默认UTF-8
|
|
* @param string $omitted 末尾省略符 默认...
|
|
* @return string
|
|
*/
|
|
function cut_str($string, $sublen=250, $omitted = '', $start=0, $code='UTF-8')
|
|
{
|
|
$string = strip_tags($string);
|
|
$string = str_replace(" ","",$string);
|
|
$string = mb_strcut($string,$start,$sublen,$code);
|
|
$string.= $omitted;
|
|
return $string;
|
|
}
|
|
|
|
//PhpAnalysis获取中文分词
|
|
function get_keywords($keyword)
|
|
{
|
|
require_once(resource_path('org/phpAnalysis/phpAnalysis.php'));
|
|
//import("Vendor.phpAnalysis.phpAnalysis");
|
|
//初始化类
|
|
PhpAnalysis::$loadInit = false;
|
|
$pa = new PhpAnalysis('utf-8', 'utf-8', false);
|
|
//载入词典
|
|
$pa->LoadDict();
|
|
//执行分词
|
|
$pa->SetSource($keyword);
|
|
$pa->StartAnalysis( false );
|
|
$keywords = $pa->GetFinallyResult(',');
|
|
|
|
return ltrim($keywords, ",");
|
|
}
|
|
|
|
//获取二维码
|
|
function get_erweima($url="")
|
|
{
|
|
Vendor('phpqrcode.qrlib');
|
|
|
|
$url = str_replace("%26","&",$url);
|
|
$url = str_replace("%3F","?",$url);
|
|
$url = str_replace("%3D","=",$url);
|
|
|
|
return QRcode::png($url, false, "H", 6);
|
|
}
|
|
|
|
//根据栏目id获取栏目信息
|
|
function typeinfo($typeid)
|
|
{
|
|
return db("arctype")->where("id=$typeid")->find();
|
|
}
|
|
|
|
//根据栏目id获取该栏目下文章/商品的数量
|
|
function catarcnum($typeid, $modelname='article')
|
|
{
|
|
$map['typeid']=$typeid;
|
|
return \DB::table($modelname)->where($map)->count('id');
|
|
}
|
|
|
|
//根据Tag id获取该Tag标签下文章的数量
|
|
function tagarcnum($tagid)
|
|
{
|
|
$taglist = \DB::table("taglist");
|
|
if(!empty($tagid)){$map['tid']=$tagid; $taglist = $taglist->where($map);}
|
|
return $taglist->count();
|
|
}
|
|
|
|
//判断是否是图片格式,是返回true
|
|
function imgmatch($url)
|
|
{
|
|
$info = pathinfo($url);
|
|
if (isset($info['extension']))
|
|
{
|
|
if (($info['extension'] == 'jpg') || ($info['extension'] == 'jpeg') || ($info['extension'] == 'gif') || ($info['extension'] == 'png'))
|
|
{
|
|
return true;
|
|
}
|
|
else
|
|
{
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
|
|
//将栏目列表生成数组
|
|
function get_category($modelname, $parent_id=0, $pad=0)
|
|
{
|
|
$arr = array();
|
|
|
|
$temp = \DB::table($modelname)->where('pid', $parent_id);
|
|
if(get_table_columns($modelname, 'listorder'))
|
|
{
|
|
$temp = $temp->orderBy('listorder', 'asc');
|
|
}
|
|
else
|
|
{
|
|
$temp = $temp->orderBy('id', 'asc');
|
|
}
|
|
|
|
$temp = $temp->get();
|
|
|
|
$cats = object_to_array($temp);
|
|
|
|
if($cats)
|
|
{
|
|
foreach($cats as $row)//循环数组
|
|
{
|
|
$row['deep'] = $pad;
|
|
|
|
if($child = get_category($modelname, $row["id"], $pad+1))//如果子级不为空
|
|
{
|
|
$row['child'] = $child;
|
|
}
|
|
|
|
$arr[] = $row;
|
|
}
|
|
|
|
return $arr;
|
|
}
|
|
}
|
|
|
|
function category_tree($list,$pid=0)
|
|
{
|
|
global $temp;
|
|
|
|
if(!empty($list))
|
|
{
|
|
foreach($list as $v)
|
|
{
|
|
$temp[] = array("id"=>$v['id'],"deep"=>$v['deep'],"name"=>$v['name'],"pid"=>$v['pid']);
|
|
//echo $v['id'];
|
|
if(array_key_exists("child",$v))
|
|
{
|
|
category_tree($v['child'],$v['pid']);
|
|
}
|
|
}
|
|
}
|
|
|
|
return $temp;
|
|
}
|
|
|
|
//递归获取面包屑导航
|
|
function get_cat_path($cat,$table='arctype',$type='list')
|
|
{
|
|
global $temp;
|
|
|
|
$row = \DB::table($table)->select('name','pid','id')->where('id',$cat)->first();
|
|
|
|
$temp = '<a href="'.get_front_url(array("catid"=>$row->id,"type"=>$type)).'">'.$row->name."</a> > ".$temp;
|
|
|
|
if($row->pid<>0)
|
|
{
|
|
get_cat_path($row->pid, $table, $type);
|
|
}
|
|
|
|
return $temp;
|
|
}
|
|
|
|
//根据文章id获得tag,$id表示文章id,$tagid表示要排除的标签id
|
|
function taglist($id,$tagid=0)
|
|
{
|
|
$tags="";
|
|
if($tagid!=0)
|
|
{
|
|
$Taglist = \DB::table("taglist")->where('aid',$id)->where('tid', '<>', $tagid)->get();
|
|
}
|
|
else
|
|
{
|
|
$Taglist = \DB::table("taglist")->where('aid',$id)->get();
|
|
}
|
|
|
|
foreach($Taglist as $row)
|
|
{
|
|
if($tags==""){$tags='id='.$row->tid;}else{$tags=$tags.' or id='.$row->tid;}
|
|
}
|
|
|
|
if($tags!=""){return object_to_array(\DB::table("tagindex")->whereRaw(\DB::raw($tags))->get());}
|
|
}
|
|
|
|
//读取动态配置
|
|
function sysconfig($varname='')
|
|
{
|
|
$sysconfig = cache('sysconfig');
|
|
$res = '';
|
|
|
|
if(empty($sysconfig))
|
|
{
|
|
cache()->forget('sysconfig');
|
|
|
|
$sysconfig = \App\Http\Model\Sysconfig::orderBy('id')->select('varname', 'value')->get()->toArray();
|
|
|
|
cache(['sysconfig' => $sysconfig], \Carbon\Carbon::now()->addMinutes(86400));
|
|
}
|
|
|
|
if($varname != '')
|
|
{
|
|
foreach($sysconfig as $row)
|
|
{
|
|
if($varname == $row['varname'])
|
|
{
|
|
$res = $row['value'];
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
$res = $sysconfig;
|
|
}
|
|
|
|
return $res;
|
|
}
|
|
|
|
//获取https的get请求结果
|
|
function curl_post($c_url,$data='')
|
|
{
|
|
$curl = curl_init(); // 启动一个CURL会话
|
|
curl_setopt($curl, CURLOPT_URL, $c_url); // 要访问的地址
|
|
curl_setopt($curl, CURLOPT_SSL_VERIFYPEER, 0); // 对认证证书来源的检查
|
|
curl_setopt($curl, CURLOPT_SSL_VERIFYHOST, 1); // 从证书中检查SSL加密算法是否存在
|
|
curl_setopt($curl, CURLOPT_USERAGENT, $_SERVER['HTTP_USER_AGENT']); // 模拟用户使用的浏览器
|
|
curl_setopt($curl, CURLOPT_FOLLOWLOCATION, 1); // 使用自动跳转
|
|
curl_setopt($curl, CURLOPT_AUTOREFERER, 1); // 自动设置Referer
|
|
|
|
if($data)
|
|
{
|
|
curl_setopt($curl, CURLOPT_POST, 1); // 发送一个常规的Post请求
|
|
curl_setopt($curl, CURLOPT_POSTFIELDS, $data); // Post提交的数据包
|
|
}
|
|
|
|
curl_setopt($curl, CURLOPT_TIMEOUT, 30); // 设置超时限制防止死循环
|
|
curl_setopt($curl, CURLOPT_HEADER, 0); // 显示返回的Header区域内容
|
|
curl_setopt($curl, CURLOPT_RETURNTRANSFER, 1); // 获取的信息以文件流的形式返回
|
|
|
|
$tmpInfo = curl_exec($curl); // 执行操作
|
|
|
|
if (curl_errno($curl))
|
|
{
|
|
echo 'Errno'.curl_error($curl);//捕抓异常
|
|
}
|
|
|
|
curl_close($curl); // 关闭CURL会话
|
|
|
|
return $tmpInfo; // 返回数据
|
|
}
|
|
|
|
//通过file_get_content获取远程数据
|
|
function http_request_post($url,$data,$type='POST')
|
|
{
|
|
$content = http_build_query($data);
|
|
$content_length = strlen($content);
|
|
$options = array(
|
|
'http' => array(
|
|
'method' => $type,
|
|
'header' => "Content-type: application/x-www-form-urlencoded\r\n" . "Content-length: $content_length\r\n",
|
|
'content' => $content
|
|
)
|
|
);
|
|
|
|
$result = file_get_contents($url,false,stream_context_create($options));
|
|
|
|
return $result;
|
|
}
|
|
|
|
function json_to_array($json)
|
|
{
|
|
return json_decode($json,true);
|
|
}
|
|
|
|
function imageResize($url, $width, $height)
|
|
{
|
|
header('Content-type: image/jpeg');
|
|
|
|
list($width_orig, $height_orig) = getimagesize($url);
|
|
$ratio_orig = $width_orig/$height_orig;
|
|
|
|
if($width/$height > $ratio_orig)
|
|
{
|
|
$width = $height*$ratio_orig;
|
|
}
|
|
else
|
|
{
|
|
$height = $width/$ratio_orig;
|
|
}
|
|
|
|
// This resamples the image
|
|
$image_p = imagecreatetruecolor($width, $height);
|
|
$image = imagecreatefromjpeg($url);
|
|
imagecopyresampled($image_p, $image, 0, 0, 0, 0, $width, $height, $width_orig, $height_orig);
|
|
// Output the image
|
|
imagejpeg($image_p, null, 100);
|
|
}
|
|
|
|
/**
|
|
* 为文章内容添加内敛, 排除alt title <a></a>直接的字符替换
|
|
*
|
|
* @param string $body
|
|
* @return string
|
|
*/
|
|
function ReplaceKeyword($body)
|
|
{
|
|
$karr = $kaarr = array();
|
|
|
|
//暂时屏蔽超链接
|
|
$body = preg_replace("#(<a(.*))(>)(.*)(<)(\/a>)#isU", '\\1-]-\\4-[-\\6', $body);
|
|
|
|
if(cache("keywordlist")){$posts=cache("keywordlist");}else{$posts = object_to_array(DB::table("keyword")->get());cache(["keywordlist"=>$posts], \Carbon\Carbon::now()->addMinutes(2592000));}
|
|
|
|
foreach($posts as $row)
|
|
{
|
|
$keyword = trim($row['keyword']);
|
|
$key_url=trim($row['rpurl']);
|
|
$karr[] = $keyword;
|
|
$kaarr[] = "<a href='$key_url' target='_blank'><u>$keyword</u></a>";
|
|
}
|
|
|
|
asort($karr);
|
|
|
|
$body = str_replace('\"', '"', $body);
|
|
|
|
foreach ($karr as $key => $word)
|
|
{
|
|
$body = preg_replace("#".preg_quote($word)."#isU", $kaarr[$key], $body, 1);
|
|
}
|
|
|
|
//恢复超链接
|
|
return preg_replace("#(<a(.*))-\]-(.*)-\[-(\/a>)#isU", '\\1>\\3<\\4', $body);
|
|
}
|
|
|
|
/**
|
|
* 删除非站内链接
|
|
*
|
|
* @access public
|
|
* @param string $body 内容
|
|
* @param array $allow_urls 允许的超链接
|
|
* @return string
|
|
*/
|
|
function replacelinks($body, $allow_urls=array())
|
|
{
|
|
$host_rule = join('|', $allow_urls);
|
|
$host_rule = preg_replace("#[\n\r]#", '', $host_rule);
|
|
$host_rule = str_replace('.', "\\.", $host_rule);
|
|
$host_rule = str_replace('/', "\\/", $host_rule);
|
|
$arr = '';
|
|
|
|
preg_match_all("#<a([^>]*)>(.*)<\/a>#iU", $body, $arr);
|
|
|
|
if( is_array($arr[0]) )
|
|
{
|
|
$rparr = array();
|
|
$tgarr = array();
|
|
|
|
foreach($arr[0] as $i=>$v)
|
|
{
|
|
if( $host_rule != '' && preg_match('#'.$host_rule.'#i', $arr[1][$i]) )
|
|
{
|
|
continue;
|
|
}
|
|
else
|
|
{
|
|
$rparr[] = $v;
|
|
$tgarr[] = $arr[2][$i];
|
|
}
|
|
}
|
|
|
|
if( !empty($rparr) )
|
|
{
|
|
$body = str_replace($rparr, $tgarr, $body);
|
|
}
|
|
}
|
|
$arr = $rparr = $tgarr = '';
|
|
return $body;
|
|
}
|
|
|
|
/**
|
|
* 获取文本中首张图片地址
|
|
* @param [type] $content
|
|
* @return [type]
|
|
*/
|
|
function getfirstpic($content)
|
|
{
|
|
if(preg_match_all("/(src)=([\"|']?)([^ \"'>]+\.(gif|jpg|jpeg|bmp|png))\\2/i", $content, $matches))
|
|
{
|
|
$file=$_SERVER['DOCUMENT_ROOT'].$matches[3][0];
|
|
|
|
if(file_exists($file))
|
|
{
|
|
return $matches[3][0];
|
|
}
|
|
}
|
|
else
|
|
{
|
|
return false;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* 更新配置文件 / 更新系统缓存
|
|
*/
|
|
function updateconfig()
|
|
{
|
|
$str_tmp="<?php\r\n"; //得到php的起始符。$str_tmp将累加
|
|
$str_end="?>"; //php结束符
|
|
$str_tmp.="//全站配置文件\r\n";
|
|
|
|
$param = db("sysconfig")->select();
|
|
foreach($param as $row)
|
|
{
|
|
$str_tmp .= 'define("'.$row['varname'].'","'.$row['value'].'"); // '.$row['info']."\r\n";
|
|
}
|
|
|
|
$str_tmp .= $str_end; //加入结束符
|
|
//保存文件
|
|
$sf = APP_PATH."common.inc.php"; //文件名
|
|
$fp = fopen($sf,"w"); //写方式打开文件
|
|
fwrite($fp,$str_tmp); //存入内容
|
|
fclose($fp); //关闭文件
|
|
}
|
|
|
|
//清空文件夹
|
|
function dir_delete($dir)
|
|
{
|
|
//$dir = dir_path($dir);
|
|
if (!is_dir($dir)) return FALSE;
|
|
$handle = opendir($dir); //打开目录
|
|
|
|
while(($file = readdir($handle)) !== false)
|
|
{
|
|
if($file == '.' || $file == '..')continue;
|
|
$d = $dir.DIRECTORY_SEPARATOR.$file;
|
|
is_dir($d) ? dir_delete($d) : @unlink($d);
|
|
}
|
|
|
|
closedir($handle);
|
|
return @rmdir($dir);
|
|
}
|
|
|
|
//对象转数组
|
|
function object_to_array($object, $get=0)
|
|
{
|
|
$res = '';
|
|
if(!empty($object))
|
|
{
|
|
if($get==0)
|
|
{
|
|
foreach($object as $key=>$value)
|
|
{
|
|
$res[$key] = (array)$value;
|
|
}
|
|
}
|
|
elseif($get==1)
|
|
{
|
|
$res = (array)$object;
|
|
}
|
|
}
|
|
|
|
return $res;
|
|
}
|
|
|
|
/**
|
|
* 操作错误跳转的快捷方法
|
|
* @access protected
|
|
* @param string $msg 错误信息
|
|
* @param string $url 页面跳转地址
|
|
* @param mixed $time 当数字时指定跳转时间
|
|
* @return void
|
|
*/
|
|
function error_jump($msg='', $url='', $time=3)
|
|
{
|
|
if ($url=='' && isset($_SERVER["HTTP_REFERER"]))
|
|
{
|
|
$url = $_SERVER["HTTP_REFERER"];
|
|
}
|
|
|
|
if(!headers_sent())
|
|
{
|
|
header("Location:".route('admin_jump')."?error=$msg&url=$url&time=$time");
|
|
exit();
|
|
}
|
|
else
|
|
{
|
|
$str = "<meta http-equiv='Refresh' content='URL=".route('admin_jump')."?error=$msg&url=$url&time=$time"."'>";
|
|
exit($str);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* 操作成功跳转的快捷方法
|
|
* @access protected
|
|
* @param string $msg 提示信息
|
|
* @param string $url 页面跳转地址
|
|
* @param mixed $time 当数字时指定跳转时间
|
|
* @return void
|
|
*/
|
|
function success_jump($msg='', $url='', $time=1)
|
|
{
|
|
if ($url=='' && isset($_SERVER["HTTP_REFERER"]))
|
|
{
|
|
$url = $_SERVER["HTTP_REFERER"];
|
|
}
|
|
|
|
if(!headers_sent())
|
|
{
|
|
header("Location:".route('admin_jump')."?message=$msg&url=$url&time=$time");
|
|
exit();
|
|
}
|
|
else
|
|
{
|
|
$str = "<meta http-equiv='Refresh' content='URL=".route('admin_jump')."?message=$msg&url=$url&time=$time"."'>";
|
|
exit($str);
|
|
}
|
|
}
|
|
|
|
//获取表所有字段
|
|
function get_table_columns($table, $field='')
|
|
{
|
|
$res = \Illuminate\Support\Facades\Schema::getColumnListing($table);
|
|
|
|
if($field != '')
|
|
{
|
|
//判断字段是否在表里面
|
|
if(in_array($field, $res))
|
|
{
|
|
return true;
|
|
}
|
|
else
|
|
{
|
|
return false;
|
|
}
|
|
}
|
|
|
|
return $res;
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|