Pythonxlwt导出excel完整版
有一组任务数据,要把它excel下载下来,如果仅仅导出,用csv导出就很方便。
10年的全州网站建设经验,针对设计、前端、开发、售后、文案、推广等六对一服务,响应快,48小时及时工作处理。营销型网站的优势是能够根据用户设备显示端的尺寸不同,自动调整全州建站的显示方式,使网站能够适用不同显示终端,在浏览器中调整网站的宽度,无论在任何一种浏览器上浏览网站,都能展现优雅布局与设计,从而大程度地提升浏览体验。成都创新互联从事“全州网站设计”,“全州网站推广”以来,每个客户项目都认真落实执行。
但是要导出漂亮的样式,重复的地方要合并单元格,设置背影颜色,字体,边框等。而CSV是纯文本格式,不支持设置各种样式。
研究了一天,把代码写了出来。
背影颜色参考地址,搜索_colour_map_text
样式设置参考地址
完整代码如下:
import xlwt
book = xlwt.Workbook(encoding='utf-8')
sheet = book.add_sheet('sheetname')
def set_style(color='white', height=260, bold=False):
#: 设置单元格字体
style = xlwt.XFStyle()
font = xlwt.Font()
# font.name = name
font.bold = bold
font.color_index = 4
font.height = height
style.font = font
# 设置单元格背景颜色
pattern = xlwt.Pattern()
pattern.pattern = xlwt.Pattern.SOLID_PATTERN
pattern.pattern_fore_colour = xlwt.Style.colour_map[color]
style.pattern = pattern
#:设置单元格边框线条
borders = xlwt.Borders() # Create Borders
# May be: NO_LINE, THIN, MEDIUM, DASHED, DOTTED, THICK, DOUBLE, HAIR,
# MEDIUM_DASHED, THIN_DASH_DOTTED, MEDIUM_DASH_DOTTED,
# THIN_DASH_DOT_DOTTED, MEDIUM_DASH_DOT_DOTTED,
# SLANTED_MEDIUM_DASH_DOTTED, or 0x00 through 0x0D.
borders.left = xlwt.Borders.THIN
borders.right = xlwt.Borders.THIN
borders.top = xlwt.Borders.THIN
borders.bottom = xlwt.Borders.THIN
style.borders = borders # Add Borders to Style
#: 设置单元格居中格式
alignment = xlwt.Alignment() # Create Alignment
# May be: HORZ_GENERAL, HORZ_LEFT, HORZ_CENTER, HORZ_RIGHT, HORZ_FILLED,
# HORZ_JUSTIFIED, HORZ_CENTER_ACROSS_SEL, HORZ_DISTRIBUTED
alignment.horz = xlwt.Alignment.HORZ_CENTER # 横向居中
alignment.vert = xlwt.Alignment.VERT_CENTER # 纵向居中 May be: VERT_TOP, VERT_CENTER, VERT_BOTTOM, VERT_JUSTIFIED, VERT_DISTRIBUTED
alignment.wrap = xlwt.Alignment.WRAP_AT_RIGHT # :自动换行
style.alignment = alignment
return style
#:模拟数据
data = [
{
"goal_id": 17,
"goal_name": "\u5b9e\u73b0OKR\u4e0e\u4efb\u52a1\u7ba1\u7406\u7cfb\u7edf\u4e0a\u7ebf\u4e14\u672c\u7ec4\u5185\u65e5\u5e38\u5316\u4f7f\u7528",
"goal_progress": 0,
"name": "\u589e\u52a0\u76ee\u6807\u3001\u4efb\u52a1\u7684\u5ef6\u671f\u63d0\u9192\u4e0e\u5c55\u793a\u3002",
"owner": "aaron.yang",
"progress": 0,
"result_id": 26,
"result_name": "\u5b8c\u6210\u4efb\u52a1\u3001OKR\u3001\u62a5\u8868\u3001\u7ee9\u6548\u529f\u80fd\u6a21\u5757\u5f00\u53d1\u4e0e\u4e0a\u7ebf",
"result_progress": 0,
},
{
"goal_id": 17,
"goal_name": "\u5b9e\u73b0OKR\u4e0e\u4efb\u52a1\u7ba1\u7406\u7cfb\u7edf\u4e0a\u7ebf\u4e14\u672c\u7ec4\u5185\u65e5\u5e38\u5316\u4f7f\u7528",
"goal_progress": 0,
"name": "task2",
"owner": "aaron.yang",
"progress": 0,
"result_id": 26,
"result_name": "\u5b8c\u6210\u4efb\u52a1\u3001OKR\u3001\u62a5\u8868\u3001\u7ee9\u6548\u529f\u80fd\u6a21\u5757\u5f00\u53d1\u4e0e\u4e0a\u7ebf",
"result_progress": 0,
},
{
"goal_id": 17,
"goal_name": "\u5b9e\u73b0OKR\u4e0e\u4efb\u52a1\u7ba1\u7406\u7cfb\u7edf\u4e0a\u7ebf\u4e14\u672c\u7ec4\u5185\u65e5\u5e38\u5316\u4f7f\u7528",
"goal_progress": 0,
"name": "task3",
"owner": "aaron.yang",
"progress": 0,
"result_id": 26,
"result_name": "\u5b8c\u6210\u4efb\u52a1\u3001OKR\u3001\u62a5\u8868\u3001\u7ee9\u6548\u529f\u80fd\u6a21\u5757\u5f00\u53d1\u4e0e\u4e0a\u7ebf",
"result_progress": 0,
},
]
# 写入第一行数据
row0 = ['目标名称', '目标进度', '关键结果名称', '关键结果进度', '任务名称', '任务负责人', '任务进度']
sheet.write_merge(0, 0, 0, 3, 'OKR', set_style('purple_ega', height=500))
sheet.write_merge(0, 0, 4, 6, '任务', set_style('cyan_ega', height=500))
# 写入第二行数据
for i in range(0, len(row0)):
sheet.write(1, i, row0[i], set_style('sky_blue'))
# 把需要合并的单元格,统计要合并多少行,以key,value形式缓存起来。
goal_tmp = {}
result_tmp = {}
for task in data:
if task['goal_id'] not in goal_tmp:
goal_tmp[task['goal_id']] = 1
else:
goal_tmp[task['goal_id']] += 1
if task['result_id'] not in result_tmp:
result_tmp[task['result_id']] = 1
else:
result_tmp[task['result_id']] += 1
goal_ids = []
result_ids = []
for index, task in enumerate(data):
index = index + 1
row_start = 1 + index
#: 如果已合并单元格,把id追加到列表,避免重复合并
if task['goal_id'] not in goal_ids:
row_end = index + goal_tmp[task['goal_id']]
sheet.write_merge(row_start, row_end, 0, 0, task['goal_name'], set_style())
sheet.write_merge(row_start, row_end, 1, 1, task['goal_progress'], set_style())
goal_ids.append(task['goal_id'])
if task['result_id'] not in result_ids:
row_end = index + result_tmp[task['result_id']]
sheet.write_merge(row_start, row_end, 2, 2, task['result_name'], set_style())
sheet.write_merge(row_start, row_end, 3, 3, task['result_progress'], set_style('yellow'))
result_ids.append(task['result_id'])
# 写入接下来的几列数据,并根据关键的列设置背景颜色
sheet.write(row_start, 4, task['name'], set_style())
sheet.write(row_start, 5, task['owner'], set_style())
sheet.write(row_start, 6, task['progress'], set_style('yellow'))
#: 设置单元格宽度
sheet.col(0).width = 10000
sheet.col(2).width = 10000
sheet.col(3).width = 4000
sheet.col(4).width = 10000
sheet.col(5).width = 5000
#:写入到文件
book.save('/tmp/bookname.xls')
网站栏目:Pythonxlwt导出excel完整版
本文网址:http://lswzjz.com/article/peocds.html