这是一个简化后的问题，我在macOS下使用Python和NetworkX时，遇到了一个奇怪的情况——在循环里给图的边属性赋值后，到了循环外面就没法正常访问这个属性的值了。以下是我的代码和运行输出：

import networkx as nx
import matplotlib.pyplot as plt
import pandas as pd
import matplotlib.pyplot as plt
from pyvis.network import Network
from io import StringIO

csv_cities= """City,Location,Lon,Lat,x,y,,LonDat,LatDat
Los Angeles,"Los Angeles, California, United States",-118.254190,34.048050,140,328,q,-123.114980,25.761681
New York,"New York, New York, United States",-74.005994,40.712749,1415,591,q,-71.056800,51.045700
Atlanta,"Atlanta, Georgia, United States",-84.389854,33.750800,1116,316,q,-94.971032,39.002842
Chicago,"Chicago, Illinois, United States",-87.632360,41.881954,1022,638,q,52.05818,25.284019 """

csv_connections= """City1,City2,numTracks,NumCarSpots,Colors,SegmentTotalCarSpots
Chicago,Los Angeles,1,4,Blue,4
New York,Chicago,2,2,"Orange, Black",4
Atlanta,Chicago,1,3,Blue,3"""

cities = pd.read_csv(StringIO(csv_cities))
interCityConnections = pd.read_csv(StringIO(csv_connections))

G = nx.Graph()          

#Add cities to graph
coords={}
for index, row in cities.iterrows():
    print(f"{row['City']:<17} {int(row['x']):<6} {int(row['y']):<6}")
    G.add_node(row['City'],city=True, x=row['x'], y=row['y'], physics=False, 
    label=row['City'], title=f'{row['x']},{row['y']}')
    coords[row['City']]=int(row['x']),int(row['y']) 
    
print("- 1 - ")

#Add intercity connections
for index, row in interCityConnections.iterrows():    
    print(f"{row['City1']:>17} <-{row['NumCarSpots']} {row['Colors']}-> {row['City2']:<17} ")
    G.add_edge(row['City1'],row['City2'],interCity=True,numTracks=row['numTracks'])
    G[row['City1']][row['City2']]['taxes']="Major"  # This works
    shortPaths=nx.all_shortest_paths(G,row['City1'], row['City2'])
    G[row['City1']][row['City2']]['shortPaths']=shortPaths  #some assignment happens here
                                                            # but value not usable outside loop?
    for p in shortPaths:    #This works
        print(f"Path:{p}")  #This works

print("- 2 - ")
    
print(G['Atlanta']['Chicago'])
print(f"type:{type(G['Atlanta']['Chicago']['shortPaths'])}")
for p in G['Atlanta']['Chicago']['shortPaths']: 
    print("For looping...") # This never executes
    print(f"Path:{p}")      # This never executes
print("Done. There should be a path displayed above this line.")

运行输出：

q@q-mbp ttr % py simplebug.py
Los Angeles       140    328   
New York          1415   591   
Atlanta           1116   316   
Chicago           1022   638   
- 1 - 
          Chicago <-4 Blue-> Los Angeles       
Path:['Chicago', 'Los Angeles']
         New York <-2 Orange, Black-> Chicago           
Path:['New York', 'Chicago']
          Atlanta <-3 Blue-> Chicago           
Path:['Atlanta', 'Chicago']
- 2 - 
{'interCity': True, 'numTracks': 1, 'taxes': 'Major', 'shortPaths': <generator object _build_paths_from_predecessors at 0x10d107480>}
type:<class 'generator'>
Done. There should be a path displayed above this line.

我想知道，怎么修改才能让循环内对边属性的修改在循环外也能生效？怎么才能在循环外访问G[row['City1']][row['City2']]['shortPaths']的赋值结果呢？我感觉这像是作用域的问题，但我刚接触Python和NetworkX，完全摸不着头绪，难道要把图对象G传入循环里吗？

问题解答 #

别急，这根本不是作用域的bug，也完全不用把图对象传入循环！问题的根源出在你赋值的生成器对象上。

你用nx.all_shortest_paths()得到的shortPaths是一个生成器，它的特性是：只能被迭代一次，迭代完成后就会被“耗尽”，里面的元素就全部消失了。

你在循环里已经用for p in shortPaths:遍历过这个生成器了，等你把它赋值给边属性再拿到循环外的时候，这个生成器已经是空的了，自然遍历不出任何内容。

解决办法超简单，把生成器转换成列表保存就行——列表是可多次迭代的容器，能把所有路径都存下来，随时都能访问：

# 把生成器转成列表再赋值
shortPaths = list(nx.all_shortest_paths(G, row['City1'], row['City2']))
G[row['City1']][row['City2']]['shortPaths'] = shortPaths

这样修改之后，不管你在循环里遍历（如果需要的话），还是到循环外面去访问，都能正常拿到所有最短路径。另外，如果你不需要在循环里打印路径，也可以直接转成列表赋值，避免提前耗尽生成器。

再补充一句：NetworkX的图对象本身就是全局的，你在循环里对它的修改本来就是持久化的，这次的问题纯粹是生成器的特性导致的，和图的作用域完全没关系哦～

备注：内容来源于stack exchange，提问作者Perry Horwich