123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931 |
- import os
- from typing import Optional
- import mysql.connector
- from dataset.util import ResultIter
- from fastapi import FastAPI
- from fastapi.middleware.cors import CORSMiddleware
- from pytrends.request import TrendReq
- from datetime import tzinfo
- import datetime
- from io import BytesIO
- from fastapi.responses import StreamingResponse
- import xlsxwriter
- import pandas as pd
- import dataset
- import json
- from pytube import extract
- app = FastAPI()
- origins = [
- "*"
- ]
- app.add_middleware(
- CORSMiddleware,
- allow_origins=origins,
- allow_credentials=True,
- allow_methods=["*"],
- allow_headers=["*"],
- )
- hhhMBPath = '../hhh-home-mb'
- hhhPCPath = '../hhh-home-pc'
- connstr = 'mysql://hhh7796hhh:lYmWsu^ujcA1@hhh-v57.cmab1ctkglka.ap-northeast-2.rds.amazonaws.com:3306/xoops?charset=utf8mb4'
- def ExecuteQuery(isql):
- with mysql.connector.connect(
- host='hhh-v57.cmab1ctkglka.ap-northeast-2.rds.amazonaws.com',
- database='xoops',
- user='hhh7796hhh',
- password='lYmWsu^ujcA1',
- use_unicode=True,
- charset='utf8',
- collation='utf8_unicode_ci'
- ) as connection :
- with connection.cursor(dictionary=True) as cursor:
- # connection.set_charset_collation('utf8','utf8_general_ci')
- #cursor = connection.cursor(dictionary=True)
- cursor.execute(isql)
- return cursor.fetchall()
- def ExecuteCmd(isql):
- with mysql.connector.connect(
- host='hhh-v57.cmab1ctkglka.ap-northeast-2.rds.amazonaws.com',
- database='xoops',
- user='hhh7796hhh',
- password='lYmWsu^ujcA1',
- use_unicode=True,
- charset='utf8',
- collation='utf8_unicode_ci'
- ) as connection:
- cursor = connection.cursor(dictionary=True)
- cursor.execute(isql)
- print(cursor.rowcount)
- connection.commit()
- return None
- @app.get("/ExportExecuteDetail")
- async def ExportExecuteDetail():
- with dataset.connect(connstr) as db:
- output = BytesIO()
- records = db.query(""" select f.exf_id, num 合約, company 合約公司,lv1 大項目,lv2 執行項, contract_time 合約到期日,price 金額,sales_man 業務,quota 額度,creator 建立者,is_close 狀態,sdate 上架日期,edate 下架日期,f.note 備註,designer 設計師,mobile 手機,telete 電話,contract_person 聯絡人,detail_status 合約名稱,d.create_time 建立時間,d.update_time 更新時間,last_update 最後更新 from execute_form f
- left join execute_detail d on f.exf_id=d.exf_id
- where f.is_delete='N' order BY f.exf_id DESC, exd_id
- """)
- df = pd.DataFrame(list(records))
- writer = pd.ExcelWriter(output)
- df.to_excel(writer, sheet_name='bar')
- writer.save()
- """ workbook = xlsxwriter.Workbook(output)
- worksheet = workbook.add_worksheet()
- for cols in records:
- worksheet.write(0, 0, 'ISBN')
- worksheet.write(0, 1, 'Name')
- worksheet.write(0, 2, 'Takedown date')
- worksheet.write(0, 3, 'Last updated')
-
- workbook.close() """
- output.seek(0)
- headers = {
- 'Content-Disposition': 'attachment; filename="execute_detail_all.xlsx"'
- }
- return StreamingResponse(output, headers=headers, media_type='application/octet-stream')
- @app.get("/")
- def read_root():
- return {"Hello": "World"}
- @app.get("/movexoopstostage")
- def movexoopstostage(designerid: str = "0", caseid: str = "0"):
- with dataset.connect(connstr) as db:
- db.query(
- "replace INTO stage._hdesigner SELECT * FROM xoops._hdesigner WHERE hdesigner_id IN ('" + designerid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hcase SELECT * FROM xoops._hcase WHERE hcase_id IN ('"+caseid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hcase_img SELECT * FROM xoops._hcase_img WHERE hcase_id IN ('"+caseid.replace(',', "','")+"');")
- return {"success"}
- @app.get("/movepxoopstostage")
- def movepxoopstostage(brandid: str = "0", productid: str = "0"):
- with dataset.connect(connstr) as db:
- db.query(
- "replace INTO stage._hbrand SELECT * FROM xoops._hbrand WHERE hbrand_id IN ('" + brandid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hbrand_page SELECT * FROM xoops._hbrand_page WHERE hbrand_id IN ('" + brandid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hproduct SELECT * FROM xoops._hproduct WHERE id IN ('"+productid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hproduct_img SELECT * FROM xoops._hproduct_img WHERE hproduct_id IN ('"+productid.replace(',', "','")+"');")
- return {"success"}
- @app.get("/movecxoopstostage")
- def movecxoopstostage(columnid: str = "0"):
- with dataset.connect(connstr) as db:
- db.query(
- "replace INTO stage._hcolumn SELECT * FROM xoops._hcolumn WHERE hcolumn_id IN ('" + columnid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hcolumn_img SELECT * FROM xoops._hcolumn_img WHERE hcolumn_id IN ('" + columnid.replace(',', "','")+"');")
- db.query(
- "replace INTO stage._hcolumn_page SELECT * FROM xoops._hcolumn_page WHERE hcolumn_id IN ('" + columnid.replace(',', "','")+"');")
- return {"success"}
- @app.get("/genjson")
- def genjson(filename: str = "realtime.json"):
-
- jData = json.load(open(hhhMBPath+'/json/data.json', encoding='utf8'))
- records = ExecuteQuery("SELECT * FROM _had where (now() between start_time and end_time or ( start_time is null and end_time is null) or ( start_time = '0000-00-00 00:00:00' and end_time = '0000-00-00 00:00:00')) and onoff='1' and adtype like '首八大%' ")
- for x in jData:
- # 頂部輪播區-新刊頭
- if x['id'] == 0:
- records = ExecuteQuery("""SELECT adlogo lo,adlogo_mobile mlo, adhref lk, adlogo_mobile_webp lomwebp, adlogo_webp dwebp FROM _had
- WHERE adtype LIKE '新刊頭%'
- AND onoff='1'
- AND(NOW() BETWEEN start_time AND end_time OR(start_time='0000-00-00 00:00:00' and end_time='0000-00-00 00:00:00') or (start_time is null and end_time is NULL))
- ORDER BY cast(SUBSTR(adtype,4) AS DECIMAL)""")
- x["data"] = []
- for c in records:
- a = {'imgUrl': c['mlo'], 'link': str(
- c['lk']), 'DimgUrl': c['lo'], 'webp': str(c['lomwebp']), 'Dwebp': str(c['dwebp'])}
- x["data"].append(a)
- # print(x["data"])
- # 主要輪播區-首八大
- if x['id'] == 1:
- records = ExecuteQuery("""SELECT adlogo lo,adlogo_mobile mlo, adhref lk, adlogo_mobile_webp lomwebp, adlogo_webp dwebp FROM _had
- WHERE adtype LIKE '首八大%'
- AND onoff='1'
- AND(NOW() BETWEEN start_time AND end_time OR(start_time='0000-00-00 00:00:00' and end_time='0000-00-00 00:00:00') or (start_time is null and end_time is NULL))
- ORDER BY cast(SUBSTR(adtype,4) AS DECIMAL)""")
- x["data"] = []
- for c in records:
- a = {'imgUrl': c['mlo'], 'link': str(
- c['lk']), 'DimgUrl': c['lo'], 'webp': str(c['lomwebp']), 'Dwebp': str(c['dwebp'])}
- x["data"].append(a)
- # print(x["data"])
- #tab區塊-最夯設計, 影音實錄, 專欄文章
- if x['id'] == 2:
- x["data"] = []
- records = ExecuteQuery("""SELECT caption TT ,cover IMG, CONCAT('https://hhh.com.tw/cases/detail/',hcase_id,'/') LK, short_desc txt
- from _hcase
- left join _hdesigner ON _hcase.hdesigner_id=_hdesigner.hdesigner_id
- WHERE
- _hcase.onoff='1' AND _hdesigner.onoff='1'
- AND(NOW() > sdate)
- ORDER BY hcase_id DESC
- LIMIT 3""")
- a = {'tab': '最夯設計', 'data': []}
- for c in records:
- ad = {'imgUrl': c['IMG'], 'link': c['LK'],
- 'title': c['TT'], 'description': c['txt']}
- a['data'].append(ad)
- x["data"].append(a)
- records = ExecuteQuery("""SELECT title TT,iframe IMG , CONCAT('https://hhh.com.tw/video-post.php?id=',hvideo_id) LK , name
- from _hvideo
- ORDER BY hvideo_id DESC
- LIMIT 4""")
- a = {'tab': '影音實錄', 'data': []}
- cnt = 0
- for c in records:
- if cnt == 0:
- cnt += 1
- continue
- tid = extract.video_id(c['IMG'])
- timg = "https://img.youtube.com/vi/" + tid+"/hqdefault.jpg"
- ad = {'imgUrl': timg, 'link': c['LK'],
- 'title': c['name'], 'description': c['TT']}
- a['data'].append(ad)
- x["data"].append(a)
- records = ExecuteQuery("""SELECT ctitle TT,clogo IMG, CONCAT('https://hhh.com.tw/columns/detail/',hcolumn_id,'/') LK, cdesc
- from _hcolumn
- WHERE onoff='1'
- AND NOW() > sdate
- ORDER BY hcolumn_id DESC
- LIMIT 3""")
- a = {'tab': '專欄文章', 'data': []}
- for c in records:
- ad = {'imgUrl': c['IMG'], 'link': c['LK'],
- 'title': c['TT'], 'description': c['cdesc']}
- a['data'].append(ad)
- x["data"].append(a)
- # print(x["data"])
- # 主題企劃區
- if x['id'] == 3:
- records = ExecuteQuery("""SELECT logo lo, CONCAT('https://hhh.com.tw/topic/detail/',htopic_id,'/') lk, `desc`, title FROM _htopic
- WHERE onoff = '1'
- ORDER BY htopic_id DESC limit 3""")
- x["data"] = []
- for c in records:
- a = {'imgUrl': c['lo'], 'link': str(
- c['lk']), 'video': 'false', 'description': c['desc'], 'title': c['title']}
- x["data"].append(a)
- # print(x["data"])
- # 編輯精選
- if x['id'] == 4:
- records = ExecuteQuery("""SELECT hcolumn_id, ctitle, clogo,cdesc
- FROM homepage_set
- LEFT JOIN _hcolumn ON mapping_id = hcolumn_id
- WHERE outer_set=8
- AND homepage_set.onoff='Y'
- AND(NOW() BETWEEN homepage_set.start_time AND homepage_set.end_time OR(homepage_set.start_time='0000-00-00 00:00:00' and homepage_set.end_time='0000-00-00 00:00:00') or (homepage_set.start_time is null and homepage_set.end_time is NULL))
- ORDER BY inner_sort""")
- x["data"] = []
- for c in records:
- a = {'imgUrl': c['clogo'], 'link': "https://hhh.com.tw/columns/detail/" + str(
- c['hcolumn_id']) + "/", 'title': c['ctitle'], 'video': 'false', 'description': c['cdesc']}
- x["data"].append(a)
- # print(x["data"])
- # 首列表廣告
- if x['id'] == 5:
- records = ExecuteQuery("""SELECT adlogo lo,adlogo_mobile mlo, adhref lk, adlogo_mobile_webp lomwebp, adlogo_webp dwebp FROM _had
- WHERE adtype LIKE '首列表廣告%'
- AND onoff='1'
- AND(NOW() BETWEEN start_time AND end_time OR(start_time='0000-00-00 00:00:00' and end_time='0000-00-00 00:00:00') or (start_time is null and end_time is NULL))
- ORDER BY adtype""")
- x["data"] = []
- for c in records:
- a = {'imgUrl': c['mlo'], 'link': str(
- c['lk']), 'DimgUrl': c['lo'], 'webp': str(c['lomwebp']), 'Dwebp': str(c['dwebp'])}
- x["data"].append(a)
- # print(x["data"])
- # 來選好物區
- if x['id'] == 6:
- records = ExecuteQuery(
- "SELECT max_row from outer_site_set WHERE title='來選好貨'")
- maxrow = 1
- for c in records:
- maxrow = c['max_row']
- records = ExecuteQuery("""(SELECT theme_type, mapping_id, IFNULL(ifnull(ifnull(_hcase.caption,_hcolumn.ctitle),_hproduct.name),_hvideo.title) COLLATE utf8_general_ci caption , IFNULL(ifnull(_hcase.cover,_hcolumn.clogo),_hproduct.cover) COLLATE utf8_general_ci J, iframe , IFNULL(ifnull(ifnull(_hcase.short_desc,_hcolumn.cdesc),_hproduct.descr),_hvideo.`desc`) COLLATE utf8_general_ci short_desc
- , (case when theme_type='case' then CONCAT('https://hhh.com.tw/cases/detail/d/',mapping_id) when theme_type='column' then CONCAT('https://hhh.com.tw/columns/detail/',mapping_id) when theme_type='product' then CONCAT('https://hhh.com.tw/product-post.php?id=',mapping_id) when theme_type='video' then CONCAT('https://hhh.com.tw/video-post.php?id=',mapping_id) ELSE '' END) url
- -- SELECT *
- FROM homepage_set
- left join _hcase ON _hcase.hcase_id=homepage_set.mapping_id AND theme_type='case'-- AND _hcase.onoff = '1'
- LEFT JOIN _hproduct ON mapping_id = _hproduct.id AND theme_type='product'-- AND _hproduct.onoff = '1'
- LEFT JOIN _hcolumn ON mapping_id = _hcolumn.hcolumn_id AND theme_type='column'-- AND _hcolumn.onoff = '1'
- LEFT JOIN _hvideo ON mapping_id = _hvideo.hvideo_id AND theme_type='video'
- WHERE homepage_set.onoff='Y'
- AND outer_set = (SELECT oss_id from outer_site_set WHERE title='來選好貨')
- AND(NOW() BETWEEN homepage_set.start_time AND homepage_set.end_time OR(homepage_set.start_time='0000-00-00 00:00:00' and homepage_set.end_time='0000-00-00 00:00:00') or (homepage_set.start_time is null and homepage_set.end_time is NULL))
- ORDER BY outer_set, inner_sort)
- UNION
- (SELECT 'product', id, `name`, cover, NULL ,descr ,CONCAT('https://hhh.com.tw/product-post.php?id=',id) FROM _hproduct WHERE onoff='1' ORDER BY id DESC LIMIT """ + str(maxrow) + """)
- LIMIT """ + str(maxrow))
- x["data"] = []
- for c in records:
- #print(c)
- if c['iframe'] is None:
- if isinstance(c['J'], bytearray) or isinstance(c['J'], bytes):
- c['J'] = c['J'].decode('utf8')
- if isinstance(c['caption'], bytearray) or isinstance(c['caption'], bytes):
- c['caption'] = c['caption'].decode('utf8')
- if isinstance(c['short_desc'], bytearray) or isinstance(c['short_desc'], bytes):
- c['short_desc'] = c['short_desc'].decode('utf8')
- a = {'imgUrl': c['J'], 'link': c['url'], 'title': c['caption'],
- 'description': c['short_desc'], 'video': 'false'}
- else:
- tid = extract.video_id(str(c['iframe']))
- timg = "https://img.youtube.com/vi/" + tid+"/hqdefault.jpg"
- ccaption = ""
- cdescription = ""
- if isinstance(c['caption'], bytearray):
- ccaption = str(c['caption'].decode('utf8'))
- else:
- ccaption = str(c['caption'])
- if c['short_desc'] is not None:
- if isinstance(c['short_desc'], bytes):
- cdescription = str(c['short_desc'].decode('utf8'))
- else:
- cdescription = str(c['short_desc'])
- a = {'imgUrl': timg, 'link': c['url'], 'title': ccaption,
- 'description': cdescription, 'video': tid}
- x["data"].append(a)
- # print(x["data"])
- # 本週推薦
- if x['id'] == 7:
- records = ExecuteQuery(
- "SELECT max_row from outer_site_set WHERE title='本週推薦'")
- maxrow = 1
- for c in records:
- maxrow = c['max_row']
- records = ExecuteQuery("""SELECT theme_type, mapping_id, IFNULL(ifnull(ifnull(_hcase.caption,_hcolumn.ctitle),_hproduct.name),_hvideo.title) caption , IFNULL(ifnull(_hcase.cover,_hcolumn.clogo),_hproduct.cover) J, iframe , IFNULL(ifnull(ifnull(_hcase.short_desc,_hcolumn.cdesc),_hproduct.descr),_hvideo.`desc`) short_desc
- , (case when theme_type='case' then CONCAT('https://hhh.com.tw/cases/detail/d/',mapping_id) when theme_type='column' then CONCAT('https://hhh.com.tw/columns/detail/',mapping_id) when theme_type='product' then CONCAT('https://hhh.com.tw/product-post.php?id=',mapping_id) when theme_type='video' then CONCAT('https://hhh.com.tw/video-post.php?id=',mapping_id) ELSE '' END) url
- -- SELECT *
- FROM homepage_set
- left join _hcase ON _hcase.hcase_id=homepage_set.mapping_id AND theme_type='case'-- AND _hcase.onoff = '1'
- LEFT JOIN _hproduct ON mapping_id = _hproduct.id AND theme_type='product'-- AND _hproduct.onoff = '1'
- LEFT JOIN _hcolumn ON mapping_id = _hcolumn.hcolumn_id AND theme_type='column'-- AND _hcolumn.onoff = '1'
- LEFT JOIN _hvideo ON mapping_id = _hvideo.hvideo_id AND theme_type='video'
- WHERE homepage_set.onoff='Y'
- AND outer_set = (SELECT oss_id from outer_site_set WHERE title='本週推薦')
- AND(NOW() BETWEEN homepage_set.start_time AND homepage_set.end_time OR(homepage_set.start_time='0000-00-00 00:00:00' and homepage_set.end_time='0000-00-00 00:00:00') or (homepage_set.start_time is null and homepage_set.end_time is NULL))
- ORDER BY outer_set, inner_sort
- LIMIT """ + str(maxrow))
- x["data"] = []
- for c in records:
- if c['iframe'] is None:
- if isinstance(c['J'], bytearray) or isinstance(c['J'], bytes):
- c['J'] = c['J'].decode('utf8')
- if isinstance(c['caption'], bytearray) or isinstance(c['caption'], bytes):
- c['caption'] = c['caption'].decode('utf8')
- if isinstance(c['short_desc'], bytearray) or isinstance(c['short_desc'], bytes):
- c['short_desc'] = c['short_desc'].decode('utf8')
- a = {'imgUrl': c['J'], 'link': c['url'], 'title': c['caption'],
- 'description': c['short_desc'], 'video': 'false'}
- else:
- tid = extract.video_id(str(c['iframe']))
- timg = "https://img.youtube.com/vi/" + tid+"/hqdefault.jpg"
- ccaption = ""
- cdescription = ""
- if isinstance(c['caption'], bytearray):
- ccaption = str(c['caption'].decode('utf8'))
- else:
- ccaption = str(c['caption'])
- if c['short_desc'] is not None:
- if isinstance(c['short_desc'], bytes):
- cdescription = str(c['short_desc'].decode('utf8'))
- else:
- cdescription = str(c['short_desc'])
- a = {'imgUrl': timg, 'link': c['url'], 'title': ccaption,
- 'description': cdescription, 'video': tid}
- x["data"].append(a)
- # print(x["data"])
- # 粉絲推薦
- if x['id'] == 8:
- records = ExecuteQuery(
- "SELECT max_row from outer_site_set WHERE title='粉絲推薦'")
- maxrow = 1
- for c in records:
- maxrow = c['max_row']
- records = ExecuteQuery("""SELECT theme_type, mapping_id, IFNULL(ifnull(ifnull(_hcase.caption,_hcolumn.ctitle),_hproduct.name),_hvideo.title) caption , IFNULL(ifnull(_hcase.cover,_hcolumn.clogo),_hproduct.cover) J, iframe , IFNULL(ifnull(ifnull(_hcase.short_desc,_hcolumn.cdesc),_hproduct.descr),_hvideo.`desc`) short_desc
- , (case when theme_type='case' then CONCAT('https://hhh.com.tw/cases/detail/d/',mapping_id) when theme_type='column' then CONCAT('https://hhh.com.tw/columns/detail/',mapping_id) when theme_type='product' then CONCAT('https://hhh.com.tw/product-post.php?id=',mapping_id) when theme_type='video' then CONCAT('https://hhh.com.tw/video-post.php?id=',mapping_id) ELSE '' END) url
- -- SELECT *
- FROM homepage_set
- left join _hcase ON _hcase.hcase_id=homepage_set.mapping_id AND theme_type='case'-- AND _hcase.onoff = '1'
- LEFT JOIN _hproduct ON mapping_id = _hproduct.id AND theme_type='product'-- AND _hproduct.onoff = '1'
- LEFT JOIN _hcolumn ON mapping_id = _hcolumn.hcolumn_id AND theme_type='column'-- AND _hcolumn.onoff = '1'
- LEFT JOIN _hvideo ON mapping_id = _hvideo.hvideo_id AND theme_type='video'
- WHERE homepage_set.onoff='Y'
- AND outer_set = (SELECT oss_id from outer_site_set WHERE title='粉絲推薦')
- AND(NOW() BETWEEN homepage_set.start_time AND homepage_set.end_time OR(homepage_set.start_time='0000-00-00 00:00:00' and homepage_set.end_time='0000-00-00 00:00:00') or (homepage_set.start_time is null and homepage_set.end_time is NULL))
- ORDER BY outer_set, inner_sort
- LIMIT """ + str(maxrow))
- x["data"] = []
- for c in records:
- if c['iframe'] is None:
- if isinstance(c['J'], bytearray) or isinstance(c['J'], bytes):
- c['J'] = c['J'].decode('utf8')
- if isinstance(c['caption'], bytearray) or isinstance(c['caption'], bytes):
- c['caption'] = c['caption'].decode('utf8')
- if isinstance(c['short_desc'], bytearray) or isinstance(c['short_desc'], bytes):
- c['short_desc'] = c['short_desc'].decode('utf8')
- a = {'imgUrl': c['J'], 'link': c['url'], 'title': c['caption'],
- 'description': c['short_desc'], 'video': 'false'}
- else:
- tid = extract.video_id(str(c['iframe']))
- timg = "https://img.youtube.com/vi/" + tid+"/hqdefault.jpg"
- ccaption = ""
- cdescription = ""
- if isinstance(c['caption'], bytearray):
- ccaption = str(c['caption'].decode('utf8'))
- else:
- ccaption = str(c['caption'])
- if c['short_desc'] is not None:
- if isinstance(c['short_desc'], bytes):
- cdescription = str(c['short_desc'].decode('utf8'))
- else:
- cdescription = str(c['short_desc'])
- a = {'imgUrl': timg, 'link': c['url'], 'title': ccaption,
- 'description': cdescription, 'video': tid}
- x["data"].append(a)
- # print(x["data"])
- if x['id'] == 9:
- records = ExecuteQuery(
- "SELECT id, (case when youtube_title = '' OR youtube_title IS NULL then (SELECT title FROM _hvideo ORDER BY hvideo_id DESC LIMIT 1) ELSE youtube_title END) T, (case when youtube_id = '' OR youtube_id IS NULL then (SELECT iframe FROM _hvideo ORDER BY hvideo_id DESC LIMIT 1) ELSE youtube_id end) Y FROM site_setup")
- for c in records:
- x['title'] = ""
- if isinstance(c['T'], bytearray):
- x['title'] = str(c['T'].decode('utf8'))
- else:
- x['title'] = str(c['T'])
- x['yt'] = extract.video_id(str(c['Y']))
- # print(id)
- if x['id'] == 10:
- records = ExecuteQuery(
- "SELECT all_search_tag ast FROM site_setup")
- x["data"] = []
- for c in records:
- x["data"] = c['ast'].split(',')
- # print(id)
- # print(jData)
- """ if not os.path.exists(hhhMBPath):
- os.mkdir(hhhMBPath)
- with open(hhhMBPath+'/json/' + filename, 'w', encoding='utf-8') as f:
- json.dump(jData, f, ensure_ascii=False, indent=4)
- if not os.path.exists(hhhPCPath):
- os.mkdir(hhhPCPath)
- with open(hhhPCPath+'/json/' + filename, 'w', encoding='utf-8') as f:
- json.dump(jData, f, ensure_ascii=False, indent=4) """
-
- return jData
- @app.get("/gencase")
- def gencase(id: str = "14151", sort: str = "new", page: str="1"):
- with dataset.connect(connstr) as db:
- jData = json.load(open(hhhMBPath+'/json/cases.json', encoding='utf8'))
- records = db.query("""SELECT *, c.style cstyle, c.style2 cstyle2 FROM _hcase c
- LEFT JOIN _hdesigner d ON c.hdesigner_id = d.hdesigner_id
- WHERE c.hcase_id = '""" + id + """' AND c.onoff='1' AND d.onoff='1' and c.sdate < now() """)
- # print(jData)
- for x in jData:
- tmpCaseDetail = []
- icount = 0
- for c in records:
- icount += 1
- #tmpCaseDetail.append({"CaseDetailImg": c["cimg"]})
- #x["CaseDetail"] = tmpCaseDetail
- if c != None:
- x["designerid"] = str(c["hdesigner_id"])
- x["CaseId"] = str(c["hcase_id"])
- x["Casetitle"] = c["caption"]
- x["CaseTeamName"] = c["name"]
- x["CaseCompany"] = c["title"]
- x["CaseCompanyAddress"] = c["address"]
- x["CaseCompanyTel"] = c["phone"]
- x["CaseCompanyEmail"] = c["mail"]
- x["CaseCompanyWeb"] = c["website"]
- x["CaseDate"] = str(c["sdate"])
- x["CaseViews"] = c["viewed"]
- x["CaseCoverImg"] = c["cover"]
- x["CaseImgAmount"] = icount
- x["CaseStyle"] = c["cstyle"]
- x["CaseHouse"] = c["layout"]
- x["CaseSize"] = c["area"]
- x["CaseProject"] = ""
- x["CaseDataMember"] = c["member"]
- x["CaseDataSize"] = c["area"]
- x["CaseDataStyle"] = c["cstyle"] + c["cstyle2"]
- x["CaseDataType"] = c["type"]
- x["CaseDataSituation"] = c["condition"]
- x["CaseDataImgProvide"] = c["provider"]
- x["CaseDataSpace"] = c["layout"]
- x["CaseDataMaterial"] = c["materials"]
- x["ContactFreeTel"] = c["service_phone"]
- x["ContactDesignerImg"] = c["img_path"]
- x["CasePageLink"] = ""
- x["CasePageprev"] = ""
- #x["CaseTag"]= []
-
- #相同設計師的個案
- sql = """SELECT * FROM _hcase c
- WHERE hdesigner_id = '""" + x["designerid"] + """' and hcase_id <> '""" + x["CaseId"] + """' and sdate < now() AND c.onoff='1'
- ORDER BY """ + ("sdate" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """
- cases = db.query(sql)
-
- tmpOtherCases = []
- for other in cases:
- tmpOtherCase = {}
- tmpOtherCase["designerid"] = str(other["hdesigner_id"])
- tmpOtherCase["casesid"] = str(other["hcase_id"])
- tmpOtherCase["Views"] = other["viewed"]
- tmpOtherCase["ProfileImg"] = other["cover"]
- tmpTags = []
- for tag in other["tag"].split(','):
- tmpTags.append({"Tag": tag , "TagLink": "" })
- tmpOtherCase["ProfileTag"] = tmpTags
- tmpOtherCases.append(tmpOtherCase)
-
- x["DesignerProfile"] = tmpOtherCases
- #相同風格的個案
- sql = """SELECT * FROM _hcase c
- WHERE style = '""" + x["CaseStyle"] + """' and hcase_id <> '""" + x["CaseId"] + """' and sdate < now() AND c.onoff='1'
- ORDER BY """ + ("sdate" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """
- cases = db.query(sql)
-
- tmpOtherCases = []
- for other in cases:
- tmpOtherCase = {}
- tmpOtherCase["designerid"] = str(other["hdesigner_id"])
- tmpOtherCase["casesid"] = str(other["hcase_id"])
- tmpOtherCase["Views"] = other["viewed"]
- tmpOtherCase["ProfileImg"] = other["cover"]
- tmpTags = []
- for tag in other["tag"].split(','):
- tmpTags.append({"Tag": tag , "TagLink": "" })
- tmpOtherCase["ProfileTag"] = tmpTags
- tmpOtherCases.append(tmpOtherCase)
-
- x["StyleProfile"] = tmpOtherCases
- #相同風格的RANDOM 10筆
- #cases = db.query("""SELECT * FROM _hcase c
- #WHERE style = '""" + x["CaseStyle"] + """' and hcase_id <> '""" + x["CaseId"] + """' and sdate < now() AND c.onoff='1'
- #ORDER BY RAND()
- #LIMIT 10
- #""")
- """ tmpOtherCases = []
- for other in cases:
- tmpOtherCase = {}
- tmpOtherCase["designerid"] = str(other["hdesigner_id"])
- tmpOtherCase["casesid"] = str(other["hcase_id"])
- tmpOtherCase["PortfoliolImg"] = other["cover"]
- tmpOtherCase["PortfoliolLink"] = ""
- tmpOtherCase["PortfoliolImgAlt"] = other["caption"]
- tmpOtherCases.append(tmpOtherCase)
- x["OtherStylePortfolio"]= tmpOtherCases """
- # print(x)
- # print(jData)
- """ if not os.path.exists(hhhMBPath):
- os.mkdir(hhhMBPath)
- with open(hhhMBPath+'/json/cases-' + id + '.json', 'w', encoding='utf-8') as f:
- json.dump(jData, f, ensure_ascii=False, indent=4) """
- return jData
- @app.get("/gendesigner")
- def gendesigner(id: str = "447", sort: str = "new", page: str = "1"):
- with dataset.connect(connstr) as db:
- jData = json.load(open(hhhMBPath+'/json/designers.json', encoding='utf8'))
- records = db.query("""SELECT * FROM _hdesigner d
- WHERE d.hdesigner_id = '""" + id + """' AND d.onoff='1' """)
- # print(jData)
- for x in jData:
- tmpCaseDetail = []
- icount = 0
- for c in records:
- icount += 1
- # tmpCaseDetail.append({"CaseDetailImg":c["cimg"]})
- if page == "1":
- x["id"] = c["hdesigner_id"]
- x["BannerImg"] = c["background"]
- x["CompanyName"] = c["title"]
- x["DesignerName"] = c["name"]
- x["Designerimg"] = c["img_path"]
- x["Description"] = c["seo"]
- x["Approve"] = c["position"]
- x["FB_link"] = c["fbpageurl"]
- x["Basics"] = [
- {"title": "免費專線:",
- "link": c["service_phone"], "data": c["service_phone"]},
- {"title": "諮詢專線:", "link": c["phone"], "data": c["phone"]},
- {"title": "諮詢專線:", "link": c["phone"], "data": c["phone"]},
- {"title": "公司傳真:", "link": c["fax"], "data": c["fax"]},
- {"title": "公司地址:", "link": c["address"], "data": c["address"]},
- {"title": "電子信箱:", "link": c["mail"], "data": c["mail"]},
- {"title": "公司網址:", "link": c["website"], "data": c["website"]}
- ]
- x["FreeCall"] = c["service_phone"]
- x["ConsoleCall_1"] = c["phone"]
- x["ConsoleCall_2"] = c["phone"]
- x["Fax"] = c["fax"]
- x["Address"] = c["address"]
- x["Email"] = c["mail"]
- x["Web"] = c["website"]
-
- x["Branches"] = []
- branches = db.query("""SELECT * FROM designer_branch br
- WHERE br.designer_id = '""" + id + """' """)
-
- for branch in branches:
- tmpobj = {
- "title": branch["title"],
- "address": branch["address"],
- "tel": branch["tel"],
- "fax": branch["fax"]
- }
- x["Branches"].append(tmpobj)
-
- x["Budget"] = c["budget"]
- x["Square"] = c["area"]
- x["SpecialCase"] = c["special"]
- x["Charge"] = c["charge"]
- x["Pay"] = c["payment"]
- x["WorkLoc"] = c["region"]
- x["WorkType"] = c["type"]
- x["WorkStyle"] = c["style"]
- x["WorkBudget"] = c["budget"]
- x["Terms"] = [
- {"title": "接案預算:", "data": c["budget"]},
- {"title": "接案坪數:", "data": c["area"]},
- {"title": "特殊接案:", "data": c["special"]},
- {"title": "收費方式:", "data": c["charge"]},
- {"title": "付費方式:", "data": c["payment"]},
- {"title": "接案區域:", "data": c["region"]},
- {"title": "接案類型:", "data": c["type"]},
- {"title": "接案風格:", "data": c["style"]}
- ]
- x["scMedia"] = [
- {"name": "Facebook", "img": "https://hhh.com.tw/assets/images/rv_web/fb.svg",
- "link": c["fbpageurl"]},
- {"name": "Line", "img": "https://hhh.com.tw/assets/images/rv_web/line.svg",
- "link": c["line_link"]},
- {"name": "Wechat", "img": "https://hhh.com.tw/assets/images/rv_web/wechat.svg",
- "link": c["fbpageurl"]},
- {"name": "email", "img": "https://hhh.com.tw/assets/images/rv_web/share.svg",
- "link": c["mail"]},
- {"name": "Like", "img": "https://hhh.com.tw/assets/images/rv_web/like-o.svg", "link": ""}
- ]
- x["Content"] = [
- {
- "Title": "設計師作品",
- "mb_title": "作品",
- "Tabtag": "intro",
- "Display_mb": "true",
- "isActive": "true",
- "Carddata": []
- },
- {
- "Title": "設計師影音",
- "mb_title": "影音",
- "Tabtag": "video",
- "Display_mb": "true",
- "isActive": "true",
- "Carddata": []
- },
- {
- "Title": "設計師專欄",
- "mb_title": "專欄",
- "Tabtag": "columns",
- "Display_mb": "true",
- "isActive": "true",
- "Carddata": []
- },
- {
- "Title": "VR360",
- "mb_title": "",
- "Tabtag": "vr360",
- "Display_mb": "false",
- "isActive": "true",
- "Carddata": []
- },
- {
- "Title": "設計師公司簡介",
- "mb_title": "公司簡介",
- "Tabtag": "company",
- "Display_mb": "true",
- "isActive": "true",
- "Carddata": [],
- "info": [
- {"title": "設計理念", "data": c["idea"]},
- {"title": "公司統編", "data": c["taxid"]},
- {"title": "相關經歷", "data": c["career"]},
- {"title": "專業證照", "data": c["license"]},
- {"title": "獲獎紀錄", "data": c["awards"]}
- ]
- }
- ]
- # 設計師作品
- x["Content"][0]["Carddata"] = []
- cases = db.query("""SELECT tag,cover,caption,hcase_id,viewed,sdate FROM _hcase c
- WHERE c.hdesigner_id = '""" + id + """' AND c.onoff='1'
- ORDER BY """ + ("sdate" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """)
- for case in cases:
- tmpobj = {
- "url": "https://hhh.com.tw/cases/detail/d/"+str(case["hcase_id"]),
- "imgURL": case["cover"],
- "title": case["caption"],
- "tag": [{"name": tag, "link": ""} for tag in case["tag"].split(',')],
- "views": case["viewed"],
- "dateSort": str(case["sdate"])
- }
- x["Content"][0]["Carddata"].append(tmpobj)
- # 設計師影音
- # https://i.ytimg.com/vi/y6VmaLC7O9Y/hqdefault.jpg
- x["Content"][1]["Carddata"] = []
- videos = db.query("""SELECT tag_vpattern,iframe,title,hvideo_id,viewed,display_datetime FROM _hvideo v
- WHERE v.hdesigner_id = '""" + id + """' AND display_datetime < NOW()
- ORDER BY """ + ("display_datetime" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """)
- for video in videos:
- tmpobj = {
- "url": "https://hhh.com.tw/video-post.php?id="+str(video['hvideo_id']),
- "imgURL": "https://i.ytimg.com/vi/" + extract.video_id(str(video['iframe']))+"/hqdefault.jpg",
- "title": video["title"],
- "tag": [{"name": tag, "link": ""} for tag in video["tag_vpattern"].split(',')],
- "views": video["viewed"],
- "dateSort": str(video["display_datetime"])
- }
- x["Content"][1]["Carddata"].append(tmpobj)
- # 設計師專欄
- x["Content"][2]["Carddata"] = []
- columns = db.query("""SELECT tag,clogo,ctitle,hcolumn_id,viewed,sdate FROM _hcolumn c
- WHERE onoff=1 AND (hdesigner_ids LIKE '""" + id + """,%' OR hdesigner_ids LIKE '%,""" + id + """,%' OR hdesigner_ids LIKE '%,""" + id + """' OR hdesigner_ids = '""" + id + """')
- ORDER BY """ + ("sdate" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """)
- for column in columns:
- tmpobj = {
- "url": "https://hhh.com.tw/video-post.php?id="+str(video['hvideo_id']),
- "imgURL": column['clogo'],
- "title": column["ctitle"],
- "tag": [] if not column["tag"] else [{"name": tag, "link": ""} for tag in column["tag"].split(',')],
- "views": column["viewed"],
- "dateSort": str(column["sdate"])
- }
- x["Content"][2]["Carddata"].append(tmpobj)
- # VR360
- x["Content"][3]["Carddata"] = []
- cases = db.query("""SELECT tag,cover,caption,hcase_id,viewed,sdate FROM _hcase c
- WHERE c.hdesigner_id = '""" + id + """' and istaging != '' AND c.onoff='1'
- ORDER BY """ + ("sdate" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """)
- for case in cases:
- tmpobj = {
- "url": "https://hhh.com.tw/cases/detail/d/"+str(case["hcase_id"]),
- "imgURL": case["cover"],
- "title": case["caption"],
- "tag": [{"name": tag, "link": ""} for tag in case["tag"].split(',')],
- "views": case["viewed"],
- "dateSort": str(case["sdate"])
- }
- x["Content"][3]["Carddata"].append(tmpobj)
- # 設計公司簡介
- #x["Content"][4]["Carddata"] = []
- #cases = db.query("")
- """ for case in cases:
- tmpobj = {
- "imgURL":case["cover"],
- "title":case["caption"],
- "tag":[{"name": tag, "link": ""} for tag in case["tag"].split(',')],
- "views":case["viewed"],
- "dateSort":str(case["sdate"])
- }
- x["Content"][4]["Carddata"].append(tmpobj) """
- # print(x)
- # print(jData)
- """ if not os.path.exists(hhhMBPath):
- os.mkdir(hhhMBPath)
- with open(hhhMBPath+'/json/designers-' + id + '.json', 'w', encoding='utf-8') as f:
- json.dump(jData, f, ensure_ascii=False, indent=4) """
- return jData
- @app.get("/gencolumn")
- def gencolumn(id: str = "6392", sort: str = "new", page: str="1"):
- with dataset.connect(connstr) as db:
- jData = json.load(open(hhhMBPath+'/json/Columns.json', encoding='utf8'))
- records = db.query("""SELECT * FROM _hcolumn c
- WHERE c.hcolumn_id = '""" + id + """' AND c.onoff='1' """)
- #print(id)
- for x in jData:
- icount = 0
- c = None
- for c in records:
- icount += 1
- if c != None:
- x["Columnsid"] = str(c["hcolumn_id"])
- x["Columnstitle"] = c["ctitle"]
- x["ColumnsCoverImg"] = c["clogo"]
- x["ColumnsDate"] = str(c["sdate"])
- x["ColumnsViews"] = str(c["viewed"])
- if page == "1":
- x["ColumnsContent"] = c["page_content"]
- tmpTags = []
- for tag in c["ctag"].split(','):
- tmpTags.append({"Tag": tag})
- x["ColumnsTag"] = tmpTags
- x["author_inf"] = c["extend_str"]
-
- #相同類別的最新12筆
- sql = """SELECT * FROM _hcolumn c
- WHERE (c.ctype like '%""" + str(c["ctype"]) + """%' and c.ctype_sub like '%""" + str(c["ctype_sub"]) + """%') and hcolumn_id <> '""" + str(c["hcolumn_id"]) + """' and sdate < now() AND c.onoff='1'
- ORDER BY """ + ("sdate" if sort == 'new' else 'viewed') + """ DESC
- LIMIT """ + str((int(page) - 1)*12) + """,12
- """
- ctypes = db.query(sql)
- tmpOtherCols = []
- for other in ctypes:
- tmpOtherCol = {}
- tmpOtherCol["Columnsid"] = str(other["hcolumn_id"])
- tmpOtherCol["ColumnsCoverImg"] = other["clogo"]
- tmpOtherCol["Views"] = other["viewed"]
- tmpOtherCol["Columnstitle"] = other["ctitle"]
- tmpTags = []
- for tag in other["ctag"].split(','):
- tmpTags.append({"Tag": tag})
- tmpOtherCol["ColumnsTag"] = tmpTags
- tmpOtherCols.append(tmpOtherCol)
- x["OtherColumns"] = tmpOtherCols
-
- #print(x)
- # print(jData)
- """ if not os.path.exists(hhhMBPath):
- os.mkdir(hhhMBPath)
- with open(hhhMBPath+'/json/Columns-' + id + '.json', 'w', encoding='utf-8') as f:
- json.dump(jData, f, ensure_ascii=False, indent=4) """
- return jData
- @app.get("/getColumnAds")
- def getColumnAds():
- with dataset.connect(connstr) as db:
- ads = []
-
- records = db.query("""SELECT adlogo_mobile imgUrl, adlogo DimgUrl, adhref link
- FROM _had hh
- WHERE adtype LIKE '專欄首大%'
- AND hh.onoff='1'
- and NOW() BETWEEN start_time AND end_time
- ORDER BY cast(SUBSTR(adtype, 4) AS DECIMAL) """)
-
- """ for x in records:
- ads.append(x) """
- ads.extend(records)
- rData = json.loads(json.dumps(ads))
- #print(json.dumps(ads))
- return rData
-
-
- #print(getColumnAds())
- """ if __name__ == "__main__":
- uvicorn.run(app, host="0.0.0.0", port=8000) """
|