1
00:00:05,239 --> 00:00:07,259
[music]

2
00:00:09,825 --> 00:00:11,845
[music]

3
00:00:15,040 --> 00:00:19,880
Can I get a warm welcome for Peter?

4
00:00:20,305 --> 00:00:22,325
[applause]

5
00:00:26,960 --> 00:00:28,820
Thank you. Thank you very much. Um,

6
00:00:28,820 --> 00:00:29,439
[snorts]

7
00:00:29,439 --> 00:00:30,960
I was going to write this talk, but then

8
00:00:30,960 --> 00:00:35,600
I uh I didn't get around to it. Sorry.

9
00:00:35,600 --> 00:00:38,879
That was a joke.

10
00:00:38,879 --> 00:00:40,640
For those that don't know me, my name is

11
00:00:40,640 --> 00:00:42,559
Peter. Peter love it. I have been a

12
00:00:42,559 --> 00:00:44,399
programmer for longer than most of you

13
00:00:44,399 --> 00:00:47,280
have been alive.

14
00:00:47,280 --> 00:00:49,520
Um, I run uh computer programming

15
00:00:49,520 --> 00:00:51,920
language training courses uh in Python,

16
00:00:51,920 --> 00:00:55,039
of course, um, and also C++, hence the

17
00:00:55,039 --> 00:00:57,760
name of my company. And I've been doing

18
00:00:57,760 --> 00:01:01,120
that for a long time. The outline for

19
00:01:01,120 --> 00:01:03,520
the talk today will be in three parts.

20
00:01:03,520 --> 00:01:05,519
Firstly, looking at being a lazy

21
00:01:05,519 --> 00:01:07,680
programmer.

22
00:01:07,680 --> 00:01:10,640
Secondly, lazy evaluation.

23
00:01:10,640 --> 00:01:12,479
But about half of the talk is going to

24
00:01:12,479 --> 00:01:15,600
be on the new 3.15

25
00:01:15,600 --> 00:01:18,720
new keyword, the lazy keyword. Of

26
00:01:18,720 --> 00:01:20,560
course, I think that I'm a very brave

27
00:01:20,560 --> 00:01:22,080
person

28
00:01:22,080 --> 00:01:24,080
because I'm going to be doing live demo

29
00:01:24,080 --> 00:01:27,920
with pre-release software.

30
00:01:29,360 --> 00:01:33,680
Let's start on being a lazy programmer.

31
00:01:33,680 --> 00:01:36,159
According to Larry Wall, anyone familiar

32
00:01:36,159 --> 00:01:38,400
with Larry Wall? Oh, good. Lots of you.

33
00:01:38,400 --> 00:01:40,000
For those that aren't, Larry Wall is

34
00:01:40,000 --> 00:01:42,000
quite a famous programmer, he wrote the

35
00:01:42,000 --> 00:01:44,000
Pearl programming language. If you're

36
00:01:44,000 --> 00:01:46,159
not familiar with Pearl, it um at one

37
00:01:46,159 --> 00:01:48,240
stage was known as the duct tape of the

38
00:01:48,240 --> 00:01:51,600
internet and um uh was is a crucial

39
00:01:51,600 --> 00:01:54,880
language. and uh he came up with the

40
00:01:54,880 --> 00:01:57,280
three great virtues of a programmer.

41
00:01:57,280 --> 00:01:59,439
Although in my head it goes the three

42
00:01:59,439 --> 00:02:01,920
virtues of a great programmer. They

43
00:02:01,920 --> 00:02:04,799
being number one

44
00:02:04,799 --> 00:02:06,560
laziness,

45
00:02:06,560 --> 00:02:09,599
number two, impatience

46
00:02:09,599 --> 00:02:13,520
and number three hubris or pride in your

47
00:02:13,520 --> 00:02:18,319
work. Let's focus on laziness.

48
00:02:18,319 --> 00:02:20,400
Lazy Oh, there was a link to that.

49
00:02:20,400 --> 00:02:22,560
Laziness makes you write laborsaving

50
00:02:22,560 --> 00:02:24,160
programs that other people will find

51
00:02:24,160 --> 00:02:25,920
useful.

52
00:02:25,920 --> 00:02:28,640
Part one. Part two, document what you

53
00:02:28,640 --> 00:02:30,400
wrote so you don't have to answer so

54
00:02:30,400 --> 00:02:34,480
many in my mind bracket stupid questions

55
00:02:34,480 --> 00:02:37,680
about it. Um, this isn't a talk about

56
00:02:37,680 --> 00:02:41,040
AI, but I would make a mention about AI

57
00:02:41,040 --> 00:02:42,239
is not going to help you with the first

58
00:02:42,239 --> 00:02:44,080
part. [laughter]

59
00:02:44,080 --> 00:02:47,920
That's why we still need programmers.

60
00:02:47,920 --> 00:02:50,319
So, I'm going to do an exercise here, a

61
00:02:50,319 --> 00:02:52,560
live thing where I'm going to break the

62
00:02:52,560 --> 00:02:55,040
uh the the the group up into two. We'll

63
00:02:55,040 --> 00:02:57,120
have the lefthand side and we'll have

64
00:02:57,120 --> 00:02:58,879
the right hand side and we'll see who's

65
00:02:58,879 --> 00:03:02,319
better. The um I'm going to show you uh

66
00:03:02,319 --> 00:03:04,000
the left hand side and then the right

67
00:03:04,000 --> 00:03:06,560
hand side, a small snippet of code and

68
00:03:06,560 --> 00:03:08,480
I'm going to show it to you for about

69
00:03:08,480 --> 00:03:12,080
one to one and a half seconds.

70
00:03:12,080 --> 00:03:14,319
And uh it's going to be similar but not

71
00:03:14,319 --> 00:03:16,959
identical code. and I'm going to see

72
00:03:16,959 --> 00:03:19,519
which of the two sides can get the

73
00:03:19,519 --> 00:03:22,480
answer for what this code actually does.

74
00:03:22,480 --> 00:03:25,280
Who's going to get it quickest?

75
00:03:25,280 --> 00:03:27,440
When you've got it, don't say it out.

76
00:03:27,440 --> 00:03:29,840
Let the uh let the other side see if

77
00:03:29,840 --> 00:03:35,560
they can get it. So, are you ready?

78
00:03:36,959 --> 00:03:38,879
H the right hand side seems to know that

79
00:03:38,879 --> 00:03:40,480
they're going to be the ones that win.

80
00:03:40,480 --> 00:03:42,239
But [laughter]

81
00:03:42,239 --> 00:03:44,400
let's try it. Left hand side, are you

82
00:03:44,400 --> 00:03:47,360
ready?

83
00:03:48,597 --> 00:03:50,560
[laughter]

84
00:03:50,560 --> 00:03:52,720
How'd you go? Don't Don't say it out

85
00:03:52,720 --> 00:03:54,560
because uh let's try the right hand

86
00:03:54,560 --> 00:03:57,112
side. Are you ready? Right hand side.

87
00:03:57,112 --> 00:03:58,080
[cheering]

88
00:03:58,080 --> 00:04:01,879
They know they're going to win.

89
00:04:04,080 --> 00:04:08,680
Let's have a look at how that was.

90
00:04:09,439 --> 00:04:11,760
As a trainer, I run training courses and

91
00:04:11,760 --> 00:04:14,000
people say, "We want self-documenting

92
00:04:14,000 --> 00:04:16,560
code. we don't need comments. I'm like,

93
00:04:16,560 --> 00:04:18,799
yeah, nah.

94
00:04:18,799 --> 00:04:21,919
Yeah, comments still have a place. And

95
00:04:21,919 --> 00:04:24,479
um I think you will all agree, even the

96
00:04:24,479 --> 00:04:27,440
left-hand side people will agree that um

97
00:04:27,440 --> 00:04:30,320
a comment there makes a big difference.

98
00:04:30,320 --> 00:04:34,080
Now um one thing about comments

99
00:04:34,080 --> 00:04:36,720
is Pepe makes a very strong point and I

100
00:04:36,720 --> 00:04:38,960
like it that inline comments are

101
00:04:38,960 --> 00:04:41,600
unnecessary and in fact distracting if

102
00:04:41,600 --> 00:04:43,840
they state the obvious. If you're doing

103
00:04:43,840 --> 00:04:45,600
an x gets x plus one and I don't know

104
00:04:45,600 --> 00:04:48,960
why pep 8 doesn't have plus equals 1 but

105
00:04:48,960 --> 00:04:51,840
uh that will increment x. We know I want

106
00:04:51,840 --> 00:04:55,040
to know why. I want the summary. I want

107
00:04:55,040 --> 00:04:57,520
a I want a oneliner.

108
00:04:57,520 --> 00:05:00,800
So I threw this into AI. Can you please

109
00:05:00,800 --> 00:05:03,919
comment this code?

110
00:05:03,919 --> 00:05:06,720
And to be honest, this is a great

111
00:05:06,720 --> 00:05:10,080
comment in a training context or in a

112
00:05:10,080 --> 00:05:13,039
detail context, but it's um and to be

113
00:05:13,039 --> 00:05:14,880
honest, it's pretty much how I would

114
00:05:14,880 --> 00:05:17,039
write a comment if I was deeply

115
00:05:17,039 --> 00:05:21,440
commenting that particular reject

116
00:05:21,440 --> 00:05:22,960
um

117
00:05:22,960 --> 00:05:25,280
intentionally complicated. So I asked

118
00:05:25,280 --> 00:05:28,160
AI, please uh summarize that, make it a

119
00:05:28,160 --> 00:05:32,160
oneliner. and for some reason it

120
00:05:32,160 --> 00:05:35,680
realized it was a date and gave me an XY

121
00:05:35,680 --> 00:05:38,320
Z.

122
00:05:38,320 --> 00:05:43,120
So, thank you AI. But um so to be clear,

123
00:05:43,120 --> 00:05:45,759
I do use AI. AI has a lot of value, but

124
00:05:45,759 --> 00:05:50,000
I also like um poking fun at AI,

125
00:05:50,000 --> 00:05:52,560
like asking um any African countries

126
00:05:52,560 --> 00:05:54,320
that begin with K. And as most of you

127
00:05:54,320 --> 00:05:56,080
know, the closest is Kenya, which starts

128
00:05:56,080 --> 00:05:58,880
with a K, but is actually spelled with a

129
00:05:58,880 --> 00:06:01,039
K.

130
00:06:01,039 --> 00:06:04,080
Or I could ask it for uh the ancient

131
00:06:04,080 --> 00:06:08,169
Greek merchant cargo ship. [snorts]

132
00:06:08,639 --> 00:06:11,120
Well, the one that I love,

133
00:06:11,120 --> 00:06:15,880
what sound does a fox make?

134
00:06:17,939 --> 00:06:19,520
[laughter] It's just that's just

135
00:06:19,520 --> 00:06:21,039
beautiful.

136
00:06:21,039 --> 00:06:23,360
I could I I feel like just doing a

137
00:06:23,360 --> 00:06:27,039
25minut talk just on AI fails because I

138
00:06:27,039 --> 00:06:29,919
also love that one. [laughter]

139
00:06:29,919 --> 00:06:32,880
I need me a laptop like that.

140
00:06:32,880 --> 00:06:38,400
Um and uh of course watch out for AI

141
00:06:38,400 --> 00:06:41,840
inside a spreadsheet.

142
00:06:41,840 --> 00:06:43,600
And I love how if you change the

143
00:06:43,600 --> 00:06:46,000
question from sum the numbers above to

144
00:06:46,000 --> 00:06:49,280
sum all the numbers above, you get a

145
00:06:49,280 --> 00:06:52,800
completely different answer. Um AI,

146
00:06:52,800 --> 00:06:54,880
again, this isn't an AI talk, but I kind

147
00:06:54,880 --> 00:06:56,319
of sort of have to talk about it a

148
00:06:56,319 --> 00:06:58,160
little bit. If you haven't come across

149
00:06:58,160 --> 00:07:01,280
the meta study, um this is one that you

150
00:07:01,280 --> 00:07:04,960
really need to be exposed to. First up,

151
00:07:04,960 --> 00:07:07,280
um it is an old study. It's from last

152
00:07:07,280 --> 00:07:11,039
year. And uh I like the way that um a

153
00:07:11,039 --> 00:07:13,280
programmer Steve Gibson of Gibson

154
00:07:13,280 --> 00:07:14,960
Research Corporation, some of you might

155
00:07:14,960 --> 00:07:17,039
know him, doesn't talk about AI, he

156
00:07:17,039 --> 00:07:19,520
talks about today's AI. [snorts] So

157
00:07:19,520 --> 00:07:22,319
whenever I say AI, I mean today's AI

158
00:07:22,319 --> 00:07:24,400
because next week, next month, next

159
00:07:24,400 --> 00:07:26,400
year, it's all going to be different.

160
00:07:26,400 --> 00:07:30,319
The meta study actually looked at um

161
00:07:30,319 --> 00:07:33,199
programmers using AI for code solving

162
00:07:33,199 --> 00:07:36,720
and these were real life problems and

163
00:07:36,720 --> 00:07:39,680
measuring the perceptional difference

164
00:07:39,680 --> 00:07:41,440
between

165
00:07:41,440 --> 00:07:44,240
how much faster I think it made me and

166
00:07:44,240 --> 00:07:46,880
how much faster it actually made me. And

167
00:07:46,880 --> 00:07:49,759
there's a big disparity between how much

168
00:07:49,759 --> 00:07:51,680
the people thought they were how much

169
00:07:51,680 --> 00:07:53,520
the programmers thought they were going

170
00:07:53,520 --> 00:07:56,160
faster and how much they uh were

171
00:07:56,160 --> 00:07:58,000
actually slower.

172
00:07:58,000 --> 00:08:01,919
So um it is a study that should be taken

173
00:08:01,919 --> 00:08:03,919
with not [snorts] a grain of salt but

174
00:08:03,919 --> 00:08:07,199
with caution because it's um measuring a

175
00:08:07,199 --> 00:08:09,759
perceptional difference. But um if

176
00:08:09,759 --> 00:08:11,199
you're not familiar with that study, I'd

177
00:08:11,199 --> 00:08:14,879
recommend looking that up afterwards.

178
00:08:14,879 --> 00:08:17,520
too long, didn't listen. Document your

179
00:08:17,520 --> 00:08:22,160
work for your own sake. And I'm all for

180
00:08:22,160 --> 00:08:23,520
if I'm going to be lazy, I'm going to

181
00:08:23,520 --> 00:08:26,240
use AI, but I'm going to use it

182
00:08:26,240 --> 00:08:29,039
carefully.

183
00:08:29,039 --> 00:08:32,479
That's part one.

184
00:08:32,479 --> 00:08:37,399
Part two, lazy evaluation.

185
00:08:38,880 --> 00:08:41,839
Lazy evaluation. For those of you that

186
00:08:41,839 --> 00:08:43,680
are familiar with Python 2, who are my

187
00:08:43,680 --> 00:08:46,000
Python 2 users? First started using

188
00:08:46,000 --> 00:08:49,920
Python in the Python 2 era. Good. Good.

189
00:08:49,920 --> 00:08:52,080
You'll remember that range gave you a

190
00:08:52,080 --> 00:08:55,080
list.

191
00:08:55,440 --> 00:08:58,000
And one of the big shifts in Python 3

192
00:08:58,000 --> 00:09:00,880
was not that print got PNS.

193
00:09:00,880 --> 00:09:02,959
That was a big shift. The biggest shift

194
00:09:02,959 --> 00:09:05,200
most people don't realize was the um the

195
00:09:05,200 --> 00:09:06,959
shift from text to uniode. And that's

196
00:09:06,959 --> 00:09:09,680
the bit that created the biggest drama.

197
00:09:09,680 --> 00:09:12,560
But there was also a big shift in lazy

198
00:09:12,560 --> 00:09:15,760
evaluation so that in Python 3 when you

199
00:09:15,760 --> 00:09:18,160
ask for a range it gives you a range

200
00:09:18,160 --> 00:09:20,480
object.

201
00:09:20,480 --> 00:09:24,279
Time for a live demo.

202
00:09:25,120 --> 00:09:30,240
So in Python 2 and Python this is 314

203
00:09:30,240 --> 00:09:33,120
which is um up to date. So here if I ask

204
00:09:33,120 --> 00:09:36,959
for a range of a range of 10 numbers I

205
00:09:36,959 --> 00:09:39,760
get a list

206
00:09:39,760 --> 00:09:44,480
and in Python 3 3 point anything color

207
00:09:44,480 --> 00:09:49,440
ooh I get a range object from zero up to

208
00:09:49,440 --> 00:09:53,040
but not including 10. Now this has to

209
00:09:53,040 --> 00:09:58,640
build the list. So if I built a list of

210
00:09:58,640 --> 00:10:02,000
actually I might not do that

211
00:10:02,000 --> 00:10:03,839
but if I did something like I'm going to

212
00:10:03,839 --> 00:10:07,440
iterate through that list for each I ps

213
00:10:07,440 --> 00:10:10,880
in my brain four always says for each

214
00:10:10,880 --> 00:10:13,680
I'm a pearl programmer

215
00:10:13,680 --> 00:10:19,839
for each I in a range oops not capitals

216
00:10:19,839 --> 00:10:23,040
from uh one one two three one two three

217
00:10:23,040 --> 00:10:25,600
maybe another

218
00:10:25,839 --> 00:10:29,839
And uh it didn't indent for me. If I is

219
00:10:29,839 --> 00:10:33,360
greater than three.

220
00:10:33,360 --> 00:10:36,360
Oops,

221
00:10:36,480 --> 00:10:39,600
then break.

222
00:10:39,600 --> 00:10:42,000
And when that runs,

223
00:10:42,000 --> 00:10:43,600
went too quickly. Let's try another

224
00:10:43,600 --> 00:10:46,560
bigger Let's try a bigger one.

225
00:10:46,560 --> 00:10:51,360
for each I um break

226
00:10:51,360 --> 00:10:54,480
it has to build that list

227
00:10:54,480 --> 00:10:57,279
and that took too long. Whereas in

228
00:10:57,279 --> 00:11:01,040
Python 3 anything for each I in range of

229
00:11:01,040 --> 00:11:03,120
numbers one I don't know pick anything

230
00:11:03,120 --> 00:11:05,920
doesn't matter integers how big an

231
00:11:05,920 --> 00:11:08,959
integer can you have

232
00:11:08,959 --> 00:11:12,037
arbitrary how much RAM have you got

233
00:11:12,037 --> 00:11:14,720
[snorts] it indents for me thank you and

234
00:11:14,720 --> 00:11:17,920
if I is greater than three it indents

235
00:11:17,920 --> 00:11:20,560
for me thank you then that will always

236
00:11:20,560 --> 00:11:24,160
be instant yay

237
00:11:24,160 --> 00:11:29,240
that's Whoops lazy evaluation.

238
00:11:29,279 --> 00:11:31,760
And uh there is actually a nice little

239
00:11:31,760 --> 00:11:34,000
trick there that happens with in as

240
00:11:34,000 --> 00:11:36,800
well. Uh in is special case for the

241
00:11:36,800 --> 00:11:38,800
range object. So it doesn't matter what

242
00:11:38,800 --> 00:11:41,600
you do. Uh [snorts] side note. Oh, I

243
00:11:41,600 --> 00:11:43,120
just noticed a typo. We didn't get

244
00:11:43,120 --> 00:11:47,120
underscores in Python 2.

245
00:11:47,120 --> 00:11:49,040
PSC got a big number. Put an underscore

246
00:11:49,040 --> 00:11:50,959
in it.

247
00:11:50,959 --> 00:11:53,600
Um so lazy valuation that was a big

248
00:11:53,600 --> 00:11:56,160
shift in two to three and it affected

249
00:11:56,160 --> 00:11:58,480
lots of the method lots of the built-in

250
00:11:58,480 --> 00:12:01,839
functions including range zip filter map

251
00:12:01,839 --> 00:12:05,120
etc etc etc and one of the big benefits

252
00:12:05,120 --> 00:12:08,000
is that it gave us quicker startup it

253
00:12:08,000 --> 00:12:10,160
was all about that startup speed and

254
00:12:10,160 --> 00:12:14,918
also affected the memory overhead

255
00:12:14,918 --> 00:12:17,279
[snorts] um or you could make your own

256
00:12:17,279 --> 00:12:19,440
and uh I'm not going to go into detail

257
00:12:19,440 --> 00:12:21,680
on yield. Uh that's been covered in

258
00:12:21,680 --> 00:12:26,000
detail over um a long time, but uh I

259
00:12:26,000 --> 00:12:28,560
could have a a generator there that uh

260
00:12:28,560 --> 00:12:31,440
is an infinite generator. PS, it would

261
00:12:31,440 --> 00:12:33,680
be easier to use it tools for that. It

262
00:12:33,680 --> 00:12:36,800
tools repeat would be nice.

263
00:12:36,800 --> 00:12:39,200
That's

264
00:12:39,200 --> 00:12:40,800
that

265
00:12:40,800 --> 00:12:44,600
that's lazy evaluation.

266
00:12:44,720 --> 00:12:50,279
Which brings us to lazy imports.

267
00:12:50,720 --> 00:12:53,920
So this has one specific use case. We're

268
00:12:53,920 --> 00:12:57,360
looking at the startup latency. How long

269
00:12:57,360 --> 00:13:00,240
it takes things to start.

270
00:13:00,240 --> 00:13:02,880
So let me give you a use case. This is

271
00:13:02,880 --> 00:13:07,680
my starting code. I'm using pandas

272
00:13:07,680 --> 00:13:09,920
and um I've got some command line

273
00:13:09,920 --> 00:13:13,519
processing. So uh if that's help,

274
00:13:13,519 --> 00:13:16,720
it does something and exits. Side note,

275
00:13:16,720 --> 00:13:18,079
I don't know if a lot of people don't

276
00:13:18,079 --> 00:13:19,920
know this, you should be using cisexit,

277
00:13:19,920 --> 00:13:22,000
not exit. Exit is for interactive use.

278
00:13:22,000 --> 00:13:25,600
CIS exit for use within a program. Um,

279
00:13:25,600 --> 00:13:28,240
if it's a help or something, then go and

280
00:13:28,240 --> 00:13:32,079
do something. And I'm not using pandas.

281
00:13:32,079 --> 00:13:35,120
Now, um, for those that were in, um,

282
00:13:35,120 --> 00:13:38,880
Alyssa's talk 2 days ago,

283
00:13:38,880 --> 00:13:42,320
she gave a, uh, very nice example of a,

284
00:13:42,320 --> 00:13:44,959
uh, a switch that some people don't know

285
00:13:44,959 --> 00:13:49,440
about, which is my friend

286
00:13:49,440 --> 00:13:52,560
Tac X. X for the extras. If you're not

287
00:13:52,560 --> 00:13:54,880
familiar with X, this is a C Python

288
00:13:54,880 --> 00:13:57,920
specific feature that may or may not

289
00:13:57,920 --> 00:13:59,440
stay.

290
00:13:59,440 --> 00:14:02,160
experimental extras

291
00:14:02,160 --> 00:14:04,480
for import time.

292
00:14:04,480 --> 00:14:07,519
I've got a little program there and uh

293
00:14:07,519 --> 00:14:11,040
that goes and import time shows me how

294
00:14:11,040 --> 00:14:13,279
long it takes for things to be imported.

295
00:14:13,279 --> 00:14:16,000
Very useful for your startup process.

296
00:14:16,000 --> 00:14:17,760
Not only does it tell you how much time,

297
00:14:17,760 --> 00:14:19,920
it shows you what's being imported. And

298
00:14:19,920 --> 00:14:21,440
all of my data scientists should be

299
00:14:21,440 --> 00:14:23,299
going pandas

300
00:14:23,299 --> 00:14:24,880
[laughter]

301
00:14:24,880 --> 00:14:28,240
because when you import pandas

302
00:14:28,240 --> 00:14:31,040
you get

303
00:14:31,040 --> 00:14:34,680
you get an error.

304
00:14:35,040 --> 00:14:39,000
Pick the right version.

305
00:14:40,639 --> 00:14:44,920
Try and pick the right version.

306
00:14:46,160 --> 00:14:50,399
Ah sorry. I had hacked my code and

307
00:14:50,399 --> 00:14:52,639
didn't want that.

308
00:14:52,639 --> 00:14:56,360
Let's try that again.

309
00:14:57,680 --> 00:15:00,560
Let's try

310
00:15:00,560 --> 00:15:04,040
saving it.

311
00:15:05,220 --> 00:15:06,079
[laughter]

312
00:15:06,079 --> 00:15:09,120
And that's the modules that pandas goes

313
00:15:09,120 --> 00:15:11,839
and imports.

314
00:15:11,839 --> 00:15:14,880
Now I'm confession. This is a very old

315
00:15:14,880 --> 00:15:17,920
laptop. I inherited it from my son who

316
00:15:17,920 --> 00:15:21,519
made me pay money for it.

317
00:15:21,519 --> 00:15:25,199
So, it's an i7. It's an early i7 and

318
00:15:25,199 --> 00:15:28,079
that took admittedly the printing would

319
00:15:28,079 --> 00:15:29,760
have taken a fair bit of time, but it

320
00:15:29,760 --> 00:15:32,320
took nearly 7 seconds for me to import

321
00:15:32,320 --> 00:15:34,639
pandas. And if I'm not going to be using

322
00:15:34,639 --> 00:15:38,399
pandas, then that was a waste of time.

323
00:15:38,399 --> 00:15:40,320
And if you have a look at what pandas

324
00:15:40,320 --> 00:15:41,839
imports,

325
00:15:41,839 --> 00:15:45,760
you can see that it imports a um am I

326
00:15:45,760 --> 00:15:47,839
allowed to swear

327
00:15:47,839 --> 00:15:49,759
lot [laughter]

328
00:15:49,759 --> 00:15:53,120
a large amount of stuff including numpy

329
00:15:53,120 --> 00:15:56,880
etc. We're familiar with that. So all of

330
00:15:56,880 --> 00:15:59,360
that's to say coming back to the

331
00:15:59,360 --> 00:16:00,880
presentation

332
00:16:00,880 --> 00:16:04,240
is that there is a downside of doing all

333
00:16:04,240 --> 00:16:07,199
my imports but we like doing all our

334
00:16:07,199 --> 00:16:09,680
imports at top at the top but here's uh

335
00:16:09,680 --> 00:16:12,560
here's let's go and explore uh so that

336
00:16:12,560 --> 00:16:14,399
was the performance that's on my

337
00:16:14,399 --> 00:16:17,519
computer and if I use the switch it

338
00:16:17,519 --> 00:16:20,800
takes the same amount of time.

339
00:16:20,800 --> 00:16:22,560
So, let's look at some options for

340
00:16:22,560 --> 00:16:25,839
solving the startup latency issue.

341
00:16:25,839 --> 00:16:28,240
Option one, don't do the import at the

342
00:16:28,240 --> 00:16:31,600
top. Do the import later on when you

343
00:16:31,600 --> 00:16:35,440
need it. Um, that will actually solve my

344
00:16:35,440 --> 00:16:37,920
imports or my startup latency problem,

345
00:16:37,920 --> 00:16:41,279
but is messy. Your code is probably

346
00:16:41,279 --> 00:16:44,480
longer than this sample code. And to

347
00:16:44,480 --> 00:16:46,399
have imports scattered all the way

348
00:16:46,399 --> 00:16:49,839
through it is considered bad style. It's

349
00:16:49,839 --> 00:16:51,920
uh in fact if you're using pilint um

350
00:16:51,920 --> 00:16:54,320
pilint will complain about that saying

351
00:16:54,320 --> 00:16:56,240
you should be doing all of the imports

352
00:16:56,240 --> 00:16:58,639
at the top.

353
00:16:58,639 --> 00:17:00,800
Option two

354
00:17:00,800 --> 00:17:02,720
is that you could do the import in the

355
00:17:02,720 --> 00:17:05,439
middle of a function. Wrap it up into a

356
00:17:05,439 --> 00:17:08,559
function put it where it's needed. Again

357
00:17:08,559 --> 00:17:10,640
um pilint will complain about that.

358
00:17:10,640 --> 00:17:12,640
Again we're going to be scattering the

359
00:17:12,640 --> 00:17:15,199
imports the dependencies throughout the

360
00:17:15,199 --> 00:17:18,720
code and that's not desirable.

361
00:17:18,720 --> 00:17:21,839
So, what's a better solution?

362
00:17:21,839 --> 00:17:23,600
Do that.

363
00:17:23,600 --> 00:17:26,799
Use the new keyword lazy

364
00:17:26,799 --> 00:17:30,240
and uh that will lazily import. It'll

365
00:17:30,240 --> 00:17:32,720
import it when it needs it, but you've

366
00:17:32,720 --> 00:17:35,360
still got it at the top. That's the end

367
00:17:35,360 --> 00:17:37,360
of the end of the talk. I hope you found

368
00:17:37,360 --> 00:17:39,039
that useful. That's really all you need

369
00:17:39,039 --> 00:17:41,840
to know.

370
00:17:41,840 --> 00:17:44,400
And of course then I get a much better

371
00:17:44,400 --> 00:17:47,520
performance because if I use my switch

372
00:17:47,520 --> 00:17:50,400
and it doesn't go and use anything out

373
00:17:50,400 --> 00:17:54,720
of pandas then it won't import pandas.

374
00:17:54,720 --> 00:17:56,720
And then the last thing that was

375
00:17:56,720 --> 00:18:00,160
imported was site. Nice and quick.

376
00:18:00,160 --> 00:18:01,840
Maybe there's a bit more I can tell you

377
00:18:01,840 --> 00:18:03,840
about lazy.

378
00:18:03,840 --> 00:18:05,760
Um if you want to know more um about

379
00:18:05,760 --> 00:18:09,200
this check out Python 810.

380
00:18:09,200 --> 00:18:13,039
Um as mentioned um this is uh a new

381
00:18:13,039 --> 00:18:17,120
feature in 315 in October. We should be

382
00:18:17,120 --> 00:18:19,600
getting it in October.

383
00:18:19,600 --> 00:18:23,440
Um the opposite of the keyword of of of

384
00:18:23,440 --> 00:18:27,440
the concept of lazy is eager. Um but

385
00:18:27,440 --> 00:18:31,360
I've also seen it called non-lazy.

386
00:18:31,360 --> 00:18:33,919
I haven't seen it called unlazy. And the

387
00:18:33,919 --> 00:18:37,520
basic syntax is that you lazy import the

388
00:18:37,520 --> 00:18:40,480
module. And again, that gives me the

389
00:18:40,480 --> 00:18:42,799
quicker startup and a lower memory

390
00:18:42,799 --> 00:18:46,240
overhead if I don't need it.

391
00:18:46,240 --> 00:18:50,720
Um, lazy is a soft keyword.

392
00:18:50,720 --> 00:18:52,080
Who's familiar with the difference

393
00:18:52,080 --> 00:18:56,720
between soft and hard keywords?

394
00:18:56,720 --> 00:18:58,320
Three of you. [laughter]

395
00:18:58,320 --> 00:19:02,160
Four of you. Toby, who I um quizzed on

396
00:19:02,160 --> 00:19:04,799
earlier,

397
00:19:04,799 --> 00:19:08,240
lazy is a soft keyword. Now, what does

398
00:19:08,240 --> 00:19:10,880
that mean? In introducing a new keyword

399
00:19:10,880 --> 00:19:12,720
into a language such as Python, there's

400
00:19:12,720 --> 00:19:14,960
a great danger that people are already

401
00:19:14,960 --> 00:19:18,640
using that word for something. I might

402
00:19:18,640 --> 00:19:21,919
have a variable called lazy.

403
00:19:21,919 --> 00:19:23,919
And if I've got a variable called lazy,

404
00:19:23,919 --> 00:19:28,400
315 will now break that. So a l a soft

405
00:19:28,400 --> 00:19:30,799
keyword is a keyword that you can still

406
00:19:30,799 --> 00:19:33,280
use as a variable. You could think of it

407
00:19:33,280 --> 00:19:35,760
as a contextual

408
00:19:35,760 --> 00:19:38,960
um a con contextual keyword. It only

409
00:19:38,960 --> 00:19:41,919
makes sense. So you can do a lazy equals

410
00:19:41,919 --> 00:19:45,679
true. That's fine. And as well you can

411
00:19:45,679 --> 00:19:49,679
also do a lazy import of a module. The

412
00:19:49,679 --> 00:19:53,840
lazy is just uh the first lazy that lazy

413
00:19:53,840 --> 00:19:57,280
is just an ordinary variable and doesn't

414
00:19:57,280 --> 00:20:00,799
conflict with the lazy import. Um this

415
00:20:00,799 --> 00:20:03,200
all happened when the paser changed in

416
00:20:03,200 --> 00:20:05,039
Python

417
00:20:05,039 --> 00:20:07,679
can't remember

418
00:20:07,679 --> 00:20:10,480
when we got the peg 311 312 somewhere

419
00:20:10,480 --> 00:20:12,799
around there.

420
00:20:12,799 --> 00:20:17,520
So lazy is one of the soft keywords.

421
00:20:17,520 --> 00:20:20,720
What are the other soft keywords in 315?

422
00:20:20,720 --> 00:20:23,440
There's five of them. Lazy is one, the

423
00:20:23,440 --> 00:20:28,240
other four. Anyone shout it out?

424
00:20:28,240 --> 00:20:29,360
Async.

425
00:20:29,360 --> 00:20:31,520
Async. Oh, that's a very interesting

426
00:20:31,520 --> 00:20:33,600
question. Yeah, that was because that

427
00:20:33,600 --> 00:20:37,280
actually varied. Um, it it wasn't uh it

428
00:20:37,280 --> 00:20:39,760
was [sighs]

429
00:20:39,760 --> 00:20:42,159
not for a while. That's right. What else

430
00:20:42,159 --> 00:20:44,000
have we got?

431
00:20:44,000 --> 00:20:46,880
Match. Match and case.

432
00:20:46,880 --> 00:20:48,799
uh came out when we got the uh the match

433
00:20:48,799 --> 00:20:53,120
and case. Um one um

434
00:20:53,120 --> 00:20:55,840
two more.

435
00:20:55,840 --> 00:20:56,720
What was that?

436
00:20:56,720 --> 00:20:57,440
Type.

437
00:20:57,440 --> 00:21:00,080
Type. Yes. Nice. And uh even that's an

438
00:21:00,080 --> 00:21:01,360
interesting one because we've even got

439
00:21:01,360 --> 00:21:04,960
type as a built-in function and uh type

440
00:21:04,960 --> 00:21:07,600
as a type aliasing keyword. And the

441
00:21:07,600 --> 00:21:10,320
other one is one that I um often forget

442
00:21:10,320 --> 00:21:13,600
which is underscore which only applies

443
00:21:13,600 --> 00:21:18,880
in a um within a case label.

444
00:21:18,880 --> 00:21:21,280
That's soft keywords. That means that

445
00:21:21,280 --> 00:21:25,440
lazy won't break your uh your um current

446
00:21:25,440 --> 00:21:29,760
programs as you move to 315.

447
00:21:29,760 --> 00:21:31,919
Lazy imports are optin. You don't have

448
00:21:31,919 --> 00:21:34,240
to use them. If you don't use them,

449
00:21:34,240 --> 00:21:36,880
they're um it'll just be eager by

450
00:21:36,880 --> 00:21:39,520
default, how it always was. Lazy is

451
00:21:39,520 --> 00:21:41,520
thread safe, so we don't have to worry

452
00:21:41,520 --> 00:21:44,960
about different um uh threads doing the

453
00:21:44,960 --> 00:21:47,360
imports. The first of the threads that

454
00:21:47,360 --> 00:21:51,120
does the actual use will be the one that

455
00:21:51,120 --> 00:21:54,559
um uh causes the invocation to um to

456
00:21:54,559 --> 00:21:57,760
load it. Uh there's no measurable

457
00:21:57,760 --> 00:21:59,600
performance overhead. So there's no real

458
00:21:59,600 --> 00:22:02,080
penalty in making them lazy from a

459
00:22:02,080 --> 00:22:04,559
performance perspective. And once

460
00:22:04,559 --> 00:22:06,960
they're brought in then there is

461
00:22:06,960 --> 00:22:09,520
literally zero overhead. There is no

462
00:22:09,520 --> 00:22:13,360
overhead once they have been

463
00:22:13,360 --> 00:22:15,679
that word. [laughter]

464
00:22:15,679 --> 00:22:18,640
Uh I've read that as reified but um my

465
00:22:18,640 --> 00:22:21,679
wife corrected me that it's raified. um

466
00:22:21,679 --> 00:22:24,799
making real once it's actually been

467
00:22:24,799 --> 00:22:27,679
brought in once one of the threads once

468
00:22:27,679 --> 00:22:30,480
somebody has used one of those fields

469
00:22:30,480 --> 00:22:33,679
then there is zero overhead

470
00:22:33,679 --> 00:22:35,120
and of course this is going to give us

471
00:22:35,120 --> 00:22:37,039
the quicker startup and lower memory

472
00:22:37,039 --> 00:22:38,799
overhead.

473
00:22:38,799 --> 00:22:41,840
Um, you can use it with a from.

474
00:22:41,840 --> 00:22:44,480
That's a little um

475
00:22:44,480 --> 00:22:46,320
there was much debate when this was

476
00:22:46,320 --> 00:22:48,880
proposed about where is lazy with from

477
00:22:48,880 --> 00:22:51,520
going to happen. And that reminds me to

478
00:22:51,520 --> 00:22:55,120
mention that um there are other ways of

479
00:22:55,120 --> 00:22:57,840
doing lazy imports. There's the lazy

480
00:22:57,840 --> 00:23:00,080
loader in the standard library, but they

481
00:23:00,080 --> 00:23:02,320
don't work with from.

482
00:23:02,320 --> 00:23:04,720
So, we needed a new keyword to make it

483
00:23:04,720 --> 00:23:07,840
work with from.

484
00:23:07,840 --> 00:23:10,559
Um but you can't use lazy on a star

485
00:23:10,559 --> 00:23:13,200
import.

486
00:23:13,200 --> 00:23:16,720
Uh imports can uh inside functions class

487
00:23:16,720 --> 00:23:19,039
bodies try accept blocks uh can't be

488
00:23:19,039 --> 00:23:21,200
lazy.

489
00:23:21,200 --> 00:23:24,880
It doesn't cascade recursively. So if

490
00:23:24,880 --> 00:23:27,039
this one's lazy that one may or may not

491
00:23:27,039 --> 00:23:30,200
be lazy.

492
00:23:30,240 --> 00:23:32,640
And uh the important point here is that

493
00:23:32,640 --> 00:23:35,520
they're resolved at access.

494
00:23:35,520 --> 00:23:37,360
So the first time you use something out

495
00:23:37,360 --> 00:23:39,600
of that module that's when it gets

496
00:23:39,600 --> 00:23:41,760
imported

497
00:23:41,760 --> 00:23:44,320
or you can call resolve if you want to

498
00:23:44,320 --> 00:23:47,840
force it to happen. Now

499
00:23:47,840 --> 00:23:52,640
uh more notes it occurs at use time not

500
00:23:52,640 --> 00:23:56,400
at the lazy import time which means uh

501
00:23:56,400 --> 00:23:58,159
one of the downsides one of the things

502
00:23:58,159 --> 00:24:00,080
that you need to be careful of is that

503
00:24:00,080 --> 00:24:03,440
module not found error could happen

504
00:24:03,440 --> 00:24:05,600
somewhere else.

505
00:24:05,600 --> 00:24:08,000
For example, if I do a lazy import of

506
00:24:08,000 --> 00:24:10,480
not there, no problem.

507
00:24:10,480 --> 00:24:15,440
Well, not yet. But it won't crash there.

508
00:24:15,440 --> 00:24:18,159
It won't module not found there. It'll

509
00:24:18,159 --> 00:24:21,520
module not found once you go to actually

510
00:24:21,520 --> 00:24:24,000
use something out of the module. For my

511
00:24:24,000 --> 00:24:25,840
money, that's probably the most um

512
00:24:25,840 --> 00:24:29,360
significant aspect of um lazy loading

513
00:24:29,360 --> 00:24:31,840
and why I might not choose to use lazy

514
00:24:31,840 --> 00:24:34,400
loading everywhere.

515
00:24:34,400 --> 00:24:38,400
So it fails there, not there. And that

516
00:24:38,400 --> 00:24:40,480
could have some um error catching

517
00:24:40,480 --> 00:24:42,080
issues.

518
00:24:42,080 --> 00:24:45,600
It's very nicely suited to typeing type

519
00:24:45,600 --> 00:24:47,120
checking

520
00:24:47,120 --> 00:24:50,000
because um in this case it doesn't even

521
00:24:50,000 --> 00:24:52,720
have to go and import.

522
00:24:52,720 --> 00:24:55,120
So the um obviously the type checker

523
00:24:55,120 --> 00:24:58,000
will need to but my program that's using

524
00:24:58,000 --> 00:25:00,480
those types now doesn't get any overhead

525
00:25:00,480 --> 00:25:03,200
from that. Although annotations have

526
00:25:03,200 --> 00:25:05,919
been um lazy for a while. The downside

527
00:25:05,919 --> 00:25:07,679
of lazy is that you're going to only be

528
00:25:07,679 --> 00:25:09,792
able to use it if you move to 315.

529
00:25:09,792 --> 00:25:11,279
[laughter]

530
00:25:11,279 --> 00:25:13,360
Some of you might be stuck on older

531
00:25:13,360 --> 00:25:16,799
versions of Python.

532
00:25:16,799 --> 00:25:20,480
He says um knowing it's everybody.

533
00:25:20,480 --> 00:25:23,600
But um so what can we do? Well uh

534
00:25:23,600 --> 00:25:27,760
upgrade to 315. Easy. Or you could go

535
00:25:27,760 --> 00:25:31,120
and use the lazy modules.

536
00:25:31,120 --> 00:25:34,000
Do that before your import. Setting that

537
00:25:34,000 --> 00:25:37,279
global variable will then um make all of

538
00:25:37,279 --> 00:25:40,559
these ones now lazy. That's a nice

539
00:25:40,559 --> 00:25:43,279
solution for backwards compatibility.

540
00:25:43,279 --> 00:25:48,120
Back to old versions like 313.

541
00:25:48,240 --> 00:25:51,120
Or you can do global control.

542
00:25:51,120 --> 00:25:53,919
You got the tac x again for CPython to

543
00:25:53,919 --> 00:25:56,240
set lazy imports to a mode. I'll show

544
00:25:56,240 --> 00:25:58,320
you what that is. You could set the

545
00:25:58,320 --> 00:26:02,240
environmental variable or using cis you

546
00:26:02,240 --> 00:26:04,000
could set lazy imports. In all of those

547
00:26:04,000 --> 00:26:07,520
cases, mode is normal as in behave like

548
00:26:07,520 --> 00:26:11,840
lazy tells it to or everybody is lazy or

549
00:26:11,840 --> 00:26:14,880
nobody is lazy.

550
00:26:14,880 --> 00:26:16,799
You can if you need to. I haven't

551
00:26:16,799 --> 00:26:18,960
explored this at depth but uh there are

552
00:26:18,960 --> 00:26:22,320
some more advanced feature filters. Um

553
00:26:22,320 --> 00:26:25,120
you can set it so that uh you get very

554
00:26:25,120 --> 00:26:28,159
close control over how the um lazy

555
00:26:28,159 --> 00:26:30,960
import works.

556
00:26:30,960 --> 00:26:32,559
Some fancy words you'll come across.

557
00:26:32,559 --> 00:26:37,360
Lazy I mentioned eager is the opposite.

558
00:26:37,360 --> 00:26:40,799
Elleading um is a very common word in

559
00:26:40,799 --> 00:26:44,720
C++ actually but um means leaving it out

560
00:26:44,720 --> 00:26:48,159
and that's what the lazy is doing and

561
00:26:48,159 --> 00:26:50,960
raify to make it concrete to make it

562
00:26:50,960 --> 00:26:56,520
real in our case to actually load it

563
00:26:56,559 --> 00:26:59,919
or don't use AI. After I wrote this talk

564
00:26:59,919 --> 00:27:01,520
I thought maybe I should just run it

565
00:27:01,520 --> 00:27:03,919
through um some AI check that I've

566
00:27:03,919 --> 00:27:05,919
covered everything. So I asked the

567
00:27:05,919 --> 00:27:07,200
question, what are the most important

568
00:27:07,200 --> 00:27:11,960
points with Python's new lazy keyword?

569
00:27:14,898 --> 00:27:18,159
[laughter] And of course, it's not 314.

570
00:27:18,159 --> 00:27:22,758
So I asked it, are you sure it's in 314?

571
00:27:22,758 --> 00:27:24,778
[laughter]

572
00:27:25,919 --> 00:27:28,080
And cudos, they admitted they were

573
00:27:28,080 --> 00:27:31,600
wrong. Too long didn't listen. Use lazy

574
00:27:31,600 --> 00:27:33,440
where it's relevant, particularly for

575
00:27:33,440 --> 00:27:36,400
command line processing.

576
00:27:36,400 --> 00:27:38,640
Um, particularly for startup. Watch out

577
00:27:38,640 --> 00:27:40,240
for import errors. You're going to have

578
00:27:40,240 --> 00:27:42,400
to be careful about that. And, uh, that

579
00:27:42,400 --> 00:27:44,480
will give you quicker startup and lower

580
00:27:44,480 --> 00:27:47,200
memory overhead.

581
00:27:47,200 --> 00:27:49,120
Too long, didn't listen. Document your

582
00:27:49,120 --> 00:27:52,880
own work. Use AI carefully. Use lazy

583
00:27:52,880 --> 00:27:55,919
evaluation yield where it's relevant.

584
00:27:55,919 --> 00:27:58,880
And use lazy imports where they're

585
00:27:58,880 --> 00:28:01,200
relevant.

586
00:28:01,200 --> 00:28:03,440
There's some details. This was done on

587
00:28:03,440 --> 00:28:05,840
pre-release software. Although by the

588
00:28:05,840 --> 00:28:07,919
time any of the video goes out, it might

589
00:28:07,919 --> 00:28:10,960
be uh it might be live benchmarks were

590
00:28:10,960 --> 00:28:13,679
done there. Your mileage may vary, but I

591
00:28:13,679 --> 00:28:16,480
got good uh good good um performance

592
00:28:16,480 --> 00:28:19,520
startup using lazy. There's some of the

593
00:28:19,520 --> 00:28:22,559
um benchmarks

594
00:28:22,559 --> 00:28:26,399
for more information. Oops.

595
00:28:26,399 --> 00:28:29,039
Uh 810. And it's worthwhile noting that

596
00:28:29,039 --> 00:28:32,399
the earlier PEP 690 which was to make

597
00:28:32,399 --> 00:28:35,039
lazy which was to make imports always

598
00:28:35,039 --> 00:28:39,200
lazy um by default was in fact rejected

599
00:28:39,200 --> 00:28:44,120
for good reasons that were listed there.

600
00:28:45,760 --> 00:28:50,584
Thanks for coming to my talk. [applause]

601
00:28:55,120 --> 00:28:57,679
Thanks to the uh Python AU team, the AV

602
00:28:57,679 --> 00:29:00,240
team, the um the venue I also meant to

603
00:29:00,240 --> 00:29:02,240
add in there as well, volunteers and the

604
00:29:02,240 --> 00:29:04,880
session chairs and uh of course to the

605
00:29:04,880 --> 00:29:06,880
Python core team and all the

606
00:29:06,880 --> 00:29:10,320
contributors. Um you can contact me at

607
00:29:10,320 --> 00:29:14,120
my company plus.com.au/cont.

608
00:29:14,399 --> 00:29:16,080
I think we're out of time.

609
00:29:16,080 --> 00:29:17,279
Yeah, I don't think we'd have time for

610
00:29:17,279 --> 00:29:18,799
questions unfortunately. I can take uh

611
00:29:18,799 --> 00:29:20,960
questions um offline after

612
00:29:20,960 --> 00:29:23,520
out out in the corridor. Um here is your

613
00:29:23,520 --> 00:29:25,279
delightful mug. Thank you. There's a

614
00:29:25,279 --> 00:29:28,575
cracker of speech. [applause]