Compiling Regexp only once

[singsang]

Dear all,

Writing some httpd logfile pre-processing (splitting it up, getting
already some basic numbers), I think that I should compile the Regexp
for the logfile entry only once.

So my guess is that I should have perhaps a class LogFormat that holds
this as a class variable or a class constant. Below I use a non-tested
regular expression that is not complete yet.

So the idea is to have:

class LogFormat
@@RegEx = Regexp.new( '(\S+) (\S+) (\S+) \[(\d+)/(\w+)/(\d+)
[+\-]\d+?\]' )
def LogFormat.regex
@@RegEx
end
end

If now from a class LogLine (instanciated for each line in the logfile)
I use something like

class LogLine
# ...
ip, rfc931, user, day, month, year, offset =
line.match(LogFormat.regex)
# ...
end

My question: How often is the Regexp compiled? When?
When the definition of LogFormat is read first?

Btw: If anybody has a ready-to-use regex for the common log format this
would be great, but I will get that done as far as I need by myself.
;-)
Other question: Does anybody know a "Webalizer" sort of thing written
in Ruby?

Thank you all so much
Herm (using a strange google account, I know)

 

[Jan Svitok]

Dear all,

Writing some httpd logfile pre-processing (splitting it up, getting
already some basic numbers), I think that I should compile the Regexp
for the logfile entry only once.

So my guess is that I should have perhaps a class LogFormat that holds
this as a class variable or a class constant. Below I use a non-tested
regular expression that is not complete yet.

So the idea is to have:

class LogFormat
@@RegEx = Regexp.new( '(\S+) (\S+) (\S+) \[(\d+)/(\w+)/(\d+)
[+\-]\d+?\]' )
def LogFormat.regex
@@RegEx
end
end

If now from a class LogLine (instanciated for each line in the logfile)
I use something like

class LogLine
# ...
ip, rfc931, user, day, month, year, offset =
line.match(LogFormat.regex)
# ...
end

My question: How often is the Regexp compiled? When?
When the definition of LogFormat is read first?

Btw: If anybody has a ready-to-use regex for the common log format this
would be great, but I will get that done as far as I need by myself.
;-)
Other question: Does anybody know a "Webalizer" sort of thing written
in Ruby?

Hi,

<non-authoritative answer follows ;-) >

- =~ seems to be (by an order) faster than String#match
- maybe you can make a constant REGEX from @@RegEx, eliminating the
need for #regex
- it seems that when the definition is constant (i.e. no #{xxx}), only
one object is created.

 

[Robert Klemme]

Dear all,

Writing some httpd logfile pre-processing (splitting it up, getting
already some basic numbers), I think that I should compile the Regexp
for the logfile entry only once.

So my guess is that I should have perhaps a class LogFormat that holds
this as a class variable or a class constant. Below I use a non-tested
regular expression that is not complete yet.

So the idea is to have:

class LogFormat
@@RegEx = Regexp.new( '(\S+) (\S+) (\S+) \[(\d+)/(\w+)/(\d+)
[+\-]\d+?\]' )
def LogFormat.regex
@@RegEx
end
end

You don't need a class variable for that. A simple class instance
variable or a constance is sufficient.

If now from a class LogLine (instanciated for each line in the logfile)
I use something like

class LogLine
# ...
ip, rfc931, user, day, month, year, offset =
line.match(LogFormat.regex)
# ...
end

My question: How often is the Regexp compiled? When?
When the definition of LogFormat is read first?

It's compiled every time the line "@@RegEx = ..." is evaluated - so most
likely only once.

Note though that it's usually faster to have a regexp in line. So in
your case you might have a method that does the line parsing (or
multiple line parsing) and that's where you can put the inline regexp
for max efficiency.

Kind regards

robert

 

[Eric Hodel]

Writing some httpd logfile pre-processing (splitting it up, getting
already some basic numbers), I think that I should compile the Regexp
for the logfile entry only once.

You should benchmark, I expect you will find you made matching slower.

So my guess is that I should have perhaps a class LogFormat that holds
this as a class variable or a class constant. Below I use a non-tested
regular expression that is not complete yet.

class LogLine
# ...
ip, rfc931, user, day, month, year, offset =
line.match(LogFormat.regex)
# ...
end

#match is the slowest way to use regular expressions.

[ruby-talk:204747]

 

[Rick DeNatale]

Note though that it's usually faster to have a regexp in line. So in
your case you might have a method that does the line parsing (or
multiple line parsing) and that's where you can put the inline regexp
for max efficiency.

Robert,

I'm not sure what you mean by inline, do you mean a regexp literal like
/[ABC]/ vs. RegEx.new('/[ABC]/') if so that's not clear.

See my benchmarks on the thread about "Ordered contrast for String or
Array," where changing a literal /./ to a constant set to
RegEx.new('/./') caused a 200 fold performance increase.

Not that you always want to do that, benchmarking always beats rules of thumb.

 

[Eric Hodel]

Note though that it's usually faster to have a regexp in line. So in
your case you might have a method that does the line parsing (or
multiple line parsing) and that's where you can put the inline regexp
for max efficiency.

Robert,

I'm not sure what you mean by inline, do you mean a regexp literal
like
/[ABC]/ vs. RegEx.new('/[ABC]/') if so that's not clear.

See my benchmarks on the thread about "Ordered contrast for String or
Array," where changing a literal /./ to a constant set to
RegEx.new('/./') caused a 200 fold performance increase.

Those aren't the same regular expression. I expect that your
benchmark is flawed.

Looking at it:

$ ruby
str = "abcdefghijklmnopqrstuvwxyz" * 5
RE = Regexp.new('/./')
p str.scan RE
-:3: warning: parenthesize argument(s) for future version
p str.scan /./
-:4: warning: parenthesize argument(s) for future version
[]
["a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m",
"n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "a",
"b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o",
"p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "a", "b", "c",
"d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q",
"r", "s", "t", "u", "v", "w", "x", "y", "z", "a", "b", "c", "d", "e",
"f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s",
"t", "u", "v", "w", "x", "y", "z", "a", "b", "c", "d", "e", "f", "g",
"h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u",
"v", "w", "x", "y", "z"]

Also, you forgot about using String#split (but it is slower than
unpack and scan).

$ ruby -e 'p Regexp.new("/./").source, /./.source'
"/./"
"."

The correct way to construct a Regexp from Regexp.new is to omit //
in the pattern when it is a String.

Not that you always want to do that, benchmarking always beats
rules of thumb.

Benchmarking and testing beat rules of thumb.

 

[Rick DeNatale]

Note though that it's usually faster to have a regexp in line. So in
your case you might have a method that does the line parsing (or
multiple line parsing) and that's where you can put the inline regexp
for max efficiency.

Robert,

I'm not sure what you mean by inline, do you mean a regexp literal
like
/[ABC]/ vs. RegEx.new('/[ABC]/') if so that's not clear.

See my benchmarks on the thread about "Ordered contrast for String or
Array," where changing a literal /./ to a constant set to
RegEx.new('/./') caused a 200 fold performance increase.

Those aren't the same regular expression. I expect that your
benchmark is flawed.

Oops, right you are.

 

[Rick DeNatale]

So, I've corrected the benchmark. I've also added another method
using String#split as Eric suggested.

Here's the code being benchmarked:
rick@frodo:~/rubyscripts$ cat stringsplit.rb
class String
To_chars_regex = Regexp.new('.')
To_chars_regex2 = Regexp.new(/./)
To_chars_regex3 = /./

def to_chars_array_with_unpack
unpack('a'*length)
end

def to_chars_array_with_split
split(//)
end

def to_chars_array_with_scan
scan /./
end

def to_chars_array_with_scan_precomp
scan To_chars_regex
end

def to_chars_array_with_scan_precomp2
scan To_chars_regex2
end

def to_chars_array_with_scan_precomp3
scan To_chars_regex3
end
end

Note that I made tested three different "pre-compiled" regex's one
with a Regex.new('.'), one with Regex.new(/./), and one a constant
with the literal regex /./.

Now here's the benchmark:
rick@frodo:~/rubyscripts$ cat benchstringsplit.rb
require 'benchmark'
include Benchmark
load 'stringsplit.rb'

iters = 100
bmbm do | x |
str = "abcdefghijklmnopqrstuvwxyz" * 5
5.times do
x.report("unpack #{str.length} character string") do
iters.times do
str.to_chars_array_with_unpack
end
end
str += str
end

str = "abcdefghijklmnopqrstuvwxyz" * 5
5.times do
x.report("scan-precomp #{str.length} character string") do
iters.times do
str.to_chars_array_with_scan_precomp
end
end
str += str
end

str = "abcdefghijklmnopqrstuvwxyz" * 5
5.times do
x.report("scan-precomp2 #{str.length} character string") do
iters.times do
str.to_chars_array_with_scan_precomp2
end
end
str += str
end

str = "abcdefghijklmnopqrstuvwxyz" * 5
5.times do
x.report("scan-precomp3 #{str.length} character string") do
iters.times do
str.to_chars_array_with_scan_precomp3
end
end
str += str
end

str = "abcdefghijklmnopqrstuvwxyz" * 5
5.times do
x.report("scan #{str.length} character string") do
iters.times do
str.to_chars_array_with_scan
end
end
str += str
end

str = "abcdefghijklmnopqrstuvwxyz" * 5
5.times do
x.report("split #{str.length} character string") do
iters.times do
str.to_chars_array_with_split
end
end
str += str
end
end

And here are the results:
rick@frodo:~/rubyscripts$ ruby benchstringsplit.rb
Rehearsal -----------------------------------------------------------------------
unpack 130 character string 0.950000 0.000000 0.950000 (
1.225834)
unpack 260 character string 0.920000 0.000000 0.920000 (
1.736296)
unpack 520 character string 0.950000 0.010000 0.960000 (
2.474714)
unpack 1040 character string 1.000000 0.000000 1.000000 (
2.142817)
unpack 2080 character string 0.940000 0.000000 0.940000 (
2.772912)
scan-precomp 130 character string 2.210000 0.000000 2.210000 (
5.032766)
scan-precomp 260 character string 2.190000 0.000000 2.190000 (
7.615967)
scan-precomp 520 character string 2.160000 0.010000 2.170000 (
5.815543)
scan-precomp 1040 character string 2.110000 0.000000 2.110000 (
5.582919)
scan-precomp 2080 character string 2.220000 0.000000 2.220000 (
4.290547)
scan-precomp2 130 character string 2.190000 0.020000 2.210000 (
4.990034)
scan-precomp2 260 character string 2.160000 0.020000 2.180000 (
4.715208)
scan-precomp2 520 character string 2.150000 0.040000 2.190000 (
3.891652)
scan-precomp2 1040 character string 2.180000 0.010000 2.190000 (
3.757304)
scan-precomp2 2080 character string 2.130000 0.010000 2.140000 (
3.308557)
scan-precomp3 130 character string 2.100000 0.000000 2.100000 (
2.134875)
scan-precomp3 260 character string 2.210000 0.040000 2.250000 (
2.793601)
scan-precomp3 520 character string 2.140000 0.000000 2.140000 (
2.504348)
scan-precomp3 1040 character string 2.090000 0.040000 2.130000 (
2.872274)
scan-precomp3 2080 character string 2.040000 0.000000 2.040000 (
2.113700)
scan 130 character string 2.240000 0.020000 2.260000 (
3.108414)
scan 260 character string 2.230000 0.010000 2.240000 (
3.767771)
scan 520 character string 2.170000 0.010000 2.180000 (
3.892586)
scan 1040 character string 2.280000 0.000000 2.280000 (
2.540717)
scan 2080 character string 2.190000 0.000000 2.190000 (
2.454891)
split 130 character string 2.350000 0.010000 2.360000 (
2.749489)
split 260 character string 2.330000 0.000000 2.330000 (
2.620174)
split 520 character string 2.370000 0.020000 2.390000 (
3.186766)
split 1040 character string 2.310000 0.040000 2.350000 (
2.848300)
split 2080 character string 2.460000 0.000000 2.460000 (
2.930465)
------------------------------------------------------------- total:
60.280000sec

user system total real
unpack 130 character string 1.280000 0.040000 1.320000 (
1.665192)
unpack 260 character string 1.340000 0.040000 1.380000 (
1.492881)
unpack 520 character string 1.170000 0.010000 1.180000 (
1.204762)
unpack 1040 character string 1.190000 0.000000 1.190000 (
1.855012)
unpack 2080 character string 1.250000 0.000000 1.250000 (
1.677772)
scan-precomp 130 character string 2.500000 0.030000 2.530000 (
3.720339)
scan-precomp 260 character string 2.460000 0.010000 2.470000 (
3.895554)
scan-precomp 520 character string 2.470000 0.020000 2.490000 (
3.777175)
scan-precomp 1040 character string 2.430000 0.050000 2.480000 (
2.955399)
scan-precomp 2080 character string 2.520000 0.000000 2.520000 (
2.828614)
scan-precomp2 130 character string 2.330000 0.010000 2.340000 (
2.668416)
scan-precomp2 260 character string 2.280000 0.000000 2.280000 (
2.322685)
scan-precomp2 520 character string 2.310000 0.000000 2.310000 (
2.335216)
scan-precomp2 1040 character string 2.300000 0.000000 2.300000 (
2.331397)
scan-precomp2 2080 character string 2.370000 0.010000 2.380000 (
2.477581)
scan-precomp3 130 character string 2.400000 0.000000 2.400000 (
2.456566)
scan-precomp3 260 character string 2.340000 0.010000 2.350000 (
2.416143)
scan-precomp3 520 character string 2.550000 0.010000 2.560000 (
3.359391)
scan-precomp3 1040 character string 2.350000 0.040000 2.390000 (
2.822663)
scan-precomp3 2080 character string 2.350000 0.010000 2.360000 (
2.649131)
scan 130 character string 2.550000 0.050000 2.600000 (
3.186714)
scan 260 character string 2.490000 0.040000 2.530000 (
3.098105)
scan 520 character string 2.370000 0.000000 2.370000 (
2.442281)
scan 1040 character string 2.330000 0.000000 2.330000 (
2.593032)
scan 2080 character string 2.360000 0.010000 2.370000 (
3.467826)
split 130 character string 2.610000 0.030000 2.640000 (
5.929269)
split 260 character string 2.730000 0.040000 2.770000 (
3.155784)
split 520 character string 2.670000 0.030000 2.700000 (
3.625916)
split 1040 character string 2.730000 0.020000 2.750000 (
8.130285)
split 2080 character string 2.750000 0.010000 2.760000 (
6.109507)

So there are slight differences between using:
PCR1 - A constant set to Regex('.')
PCR2 - A constant set to Regex(/./)
PCR3 A constant set to /./
and
LIT a literal in-line /./

For this benchmark, for each string length tested, the techniques from
fastest to slowest are;
130 Characters:
Rehearsal: PCR3(2.10), PCR1(2.21), PCR2(2.21), LIT(2.26)
"Live": PCR2(2.34), PCR3(2.40), PCR1(2.53), LIT(2.60)
260 Characters:
Rehearsal: PCR2(2.18), PCR1(2.19), PCR3(2.25), LIT(2.25)
"Live": PCR2(2.28), PCR3(2.40), PCR1(2.47), LIT(2.60)
520 Characters:
Rehearsal: PCR3(2.14), PCR1(2.17), LIT(2.18), PCR2(2.19)
"Live": PCR2(2.31), LIT(2.37), PCR1(2.49), PCR3(2.56)
1040 Characters:
Rehearsal: PCR1(2.11), PCR3(2.13), PCR2(2.19), LIT(2.28)
"Live": PCR2(2.30), LIT(2.33), PCR3(2.39), PCR1(2.48)
2080 Characters:
Rehearsal: PCR3(2.04), PCR2(2.14), LIT(2.19), PCR1(2.22)
"Live": PCR3(2.36), LIT(2.37), PCR2(2.38), PCR1(2.52)

So, at least from this benchmark it doesn't seem that in-line literal
regular expressions are faster than pre-compiled ones. In fact they
never came in first, although PCR3 which was a constant set to a
literal regex did win 4 times.

Is this significant? Who knows, and with a new RegExp engine coming
the numbers will be different in future.

So as usual the right approach is test, profile to find out what needs
improvement, and benchmark.

 

[khaines]

So there are slight differences between using:
PCR1 - A constant set to Regex('.')
PCR2 - A constant set to Regex(/./)
PCR3 A constant set to /./
and
LIT a literal in-line /./

For this benchmark, for each string length tested, the techniques from
fastest to slowest are;
130 Characters:
Rehearsal: PCR3(2.10), PCR1(2.21), PCR2(2.21), LIT(2.26)
"Live": PCR2(2.34), PCR3(2.40), PCR1(2.53), LIT(2.60)
260 Characters:
Rehearsal: PCR2(2.18), PCR1(2.19), PCR3(2.25), LIT(2.25)
"Live": PCR2(2.28), PCR3(2.40), PCR1(2.47), LIT(2.60)
520 Characters:
Rehearsal: PCR3(2.14), PCR1(2.17), LIT(2.18), PCR2(2.19)
"Live": PCR2(2.31), LIT(2.37), PCR1(2.49), PCR3(2.56)
1040 Characters:
Rehearsal: PCR1(2.11), PCR3(2.13), PCR2(2.19), LIT(2.28)
"Live": PCR2(2.30), LIT(2.33), PCR3(2.39), PCR1(2.48)
2080 Characters:
Rehearsal: PCR3(2.04), PCR2(2.14), LIT(2.19), PCR1(2.22)
"Live": PCR3(2.36), LIT(2.37), PCR2(2.38), PCR1(2.52)

So, at least from this benchmark it doesn't seem that in-line literal
regular expressions are faster than pre-compiled ones. In fact they
never came in first, although PCR3 which was a constant set to a
literal regex did win 4 times.

Those numbers seem funny, like there is a lot of background noise
affecting them.

a = Regexp.new('.')
b = Regexp.new(/./)
c = /./

irb(main):015:0> a == b
=> true
irb(main):016:0> b == c
=> true
irb(main):017:0> a == c
=> true

They all produce equivalent Regexp objects.

Further, you are benchmarking scan, which isn't the same as comparing
matching with regular expressions.

In a simple matching operation, an inline regexp, in the form of:

foo =~ /bar/

is the fastest way to match.

Ruby specially optimizes that style. Eric Hodel explained it in a post
from...some time this summer, I think.

bar = /bar/
bar = Regexp.new('bar')
bar = Regexp/new(/bar/)

are all equivalent, and

foo =~ bar

Will be slower.

Bar = /bar/
foo =~ Bar

will be slower yet.

Any variation that calls match() will be even slower yet, by a substantial
margin.

So, for regular expression matching, the fastest, if not prettiest,
approach is to use the '=~ /expression/' syntax.


Kirk Haines

 

[Eric Hodel]

So, I've corrected the benchmark. I've also added another method
using String#split as Eric suggested.

Here's the code being benchmarked:
rick@frodo:~/rubyscripts$ cat stringsplit.rb

Putting the code here involves an extra method call. While they are
all wrapped in a method call equally, doing less is always better.

class String
[methods]
end

Note that I made tested three different "pre-compiled" regex's one
with a Regex.new('.'), one with Regex.new(/./), and one a constant
with the literal regex /./.

You don't need three different benchmarks for this. Its easy to
determine that these are the same with irb.

Now here's the benchmark:
rick@frodo:~/rubyscripts$ cat benchstringsplit.rb
require 'benchmark'
include Benchmark
load 'stringsplit.rb'

iters = 100

100 iterations is never enough. GC behavior, other processes waking
up, etc. will all cause fluctuations in the benchmark.

Use 100_000 or 1_000_000.

[benchmark code]

Here's a more-meaningful version of yours:

$ cat stringsbench.rb
require 'benchmark'

N = 100_000
RE = /./
str = "abcdefghijklmnopqrstuvwxyz" * 5

Benchmark.bmbm do | x |
x.report 'empty' do N.times do end end
x.report 'str' do N.times do str end end
x.report 'unpack' do N.times do str.unpack('a' * str.length) end
end
x.report 'scan RE' do N.times do str.scan RE end end
x.report 'scan /./' do N.times do str.scan /./ end end
x.report 'split //' do N.times do str.split // end end
end

$ ruby stringsbench.rb
Rehearsal --------------------------------------------
empty 0.030000 0.000000 0.030000 ( 0.088456)
str 0.050000 0.000000 0.050000 ( 0.064515)
unpack 14.370000 0.210000 14.580000 ( 20.264274)
scan RE 24.240000 0.380000 24.620000 ( 36.296017)
scan /./ 24.100000 0.310000 24.410000 ( 32.661832)
split // 29.660000 0.360000 30.020000 ( 40.596177)
---------------------------------- total: 93.710000sec

user system total real
empty 0.030000 0.000000 0.030000 ( 0.045199)
str 0.050000 0.010000 0.060000 ( 0.065739)
unpack 14.500000 0.190000 14.690000 ( 19.813677)
scan RE 23.850000 0.350000 24.200000 ( 33.028376)
scan /./ 23.910000 0.320000 24.230000 ( 35.219831)
split // 29.720000 0.470000 30.190000 ( 42.712222)

with 'abc...xyz' * 25:

Rehearsal --------------------------------------------
empty 0.050000 0.000000 0.050000 ( 0.217969)
str 0.050000 0.000000 0.050000 ( 0.087123)
unpack 65.370000 0.710000 66.080000 ( 80.341765)
scan RE 107.840000 1.030000 108.870000 (129.723449)
scan /./ 110.320000 1.250000 111.570000 (142.540348)
split // 135.220000 1.330000 136.550000 (164.418453)
--------------------------------- total: 423.170000sec

user system total real
empty 0.030000 0.000000 0.030000 ( 0.040596)
str 0.050000 0.000000 0.050000 ( 0.066506)
unpack 62.130000 0.590000 62.720000 ( 70.183396)
scan RE 107.730000 0.940000 108.670000 (125.308283)
scan /./ 107.550000 0.950000 108.500000 (127.365719)
split // 133.700000 1.020000 134.720000 (150.798611)

So, at least from this benchmark it doesn't seem that in-line literal
regular expressions are faster than pre-compiled ones.

inline regular expressions are no less "pre-compiled" than regular
expressions in a variable or constant.

$ ruby -e '2.times do puts /./.object_id end'
938970
938970

('=~ /./' is faster than '=~ var' is faster than 'match anything',
but for other reasons)

If you want to test inline vs "pre-compiled" regular expressions you
need to throw away all the parts that are the same and focus on what
is different. In your benchmark it was what was on the right hand
side of String#scan. Since the string was the same and the scan was
the same, just throw those away.

You end up with a benchmark like this:

$ cat vs.rb
require 'benchmark'

N = 100_000_000

RE = /./
re = /./
$re = /./
@re = /./
@@re = /./

Benchmark.bmbm do |bm|
bm.report 'empty' do N.times do end end
bm.report 'lit' do N.times do /./ end end
bm.report 'local' do N.times do re end end
bm.report 'global' do N.times do $re end end
bm.report 'instance' do N.times do @re end end
bm.report 'class' do N.times do @@re end end
bm.report 'constant' do N.times do RE end end
end

$ ruby vs.rb
Rehearsal --------------------------------------------
empty 27.930000 0.120000 28.050000 ( 31.550709)
lit 48.240000 0.320000 48.560000 ( 57.422011)
local 48.820000 0.290000 49.110000 ( 60.650110)
global 49.920000 0.420000 50.340000 ( 62.156780)
instance 55.740000 0.180000 55.920000 ( 61.027820)
class 57.800000 0.190000 57.990000 ( 63.733099)
constant 59.240000 0.330000 59.570000 (119.118487)
--------------------------------- total: 349.540000sec

user system total real
empty 27.820000 0.160000 27.980000 ( 58.584698)
lit 48.170000 0.170000 48.340000 ( 59.557450)
local 48.720000 0.170000 48.890000 ( 53.775692)
global 49.830000 0.180000 50.010000 ( 55.743586)
instance 55.880000 0.370000 56.250000 ( 72.011443)
class 57.910000 0.390000 58.300000 ( 67.037158)
constant 59.230000 0.410000 59.640000 ( 77.993887)

So literal regular expressions are "faster" by about 56% (after
discounting loop overhead) (when you're performing 100 million
retrievals only) (actual speedup in real code will probably be
swallowed elsewhere or completely irrelevant).

This follows from reading eval.c. Constant lookup involves a bunch
of C function calls, but literal lookup just returns an Object stored
in the parse tree.

In fact they never came in first, although PCR3 which was a
constant set to a
literal regex did win 4 times.

In fact, when you adjust the stringsbench.rb you'll see /./ winning
over RE. Set N to 10_000_000 (or more) and str to "abcde" (or just N
high enough).

Is this significant? Who knows, and with a new RegExp engine coming
the numbers will be different in future.

The difference was probably due to processes waking up, garbage
collection and similar unevenly distributed events. With more
iterations and more focused benchmarks you'll get better results.

But ultimately, these types of microbenchmarks are not very useful.
Yes, you will get a speedup using /./ over RE, but will your program
really run long enough where it matters to make a difference? I bet
not.

So as usual the right approach is test, profile to find out what needs
improvement, and benchmark.

Be careful with your benchmarks, they are most useful when they are
as small and simple as possible. Be sure to throw away all the
irrelevant parts.

Be careful with your benchmarks, they are most useless when they are
as small and simple as possible. Be sure to understand how little
speedup you'll get.

 

[Michal Suchanek]

Dear all,

Writing some httpd logfile pre-processing (splitting it up, getting
already some basic numbers), I think that I should compile the Regexp
for the logfile entry only once.

So my guess is that I should have perhaps a class LogFormat that holds
this as a class variable or a class constant. Below I use a non-tested
regular expression that is not complete yet.

So the idea is to have:

class LogFormat
@@RegEx = Regexp.new( '(\S+) (\S+) (\S+) \[(\d+)/(\w+)/(\d+)
[+\-]\d+?\]' )
def LogFormat.regex
@@RegEx
end
end

If now from a class LogLine (instanciated for each line in the logfile)
I use something like

class LogLine
# ...
ip, rfc931, user, day, month, year, offset =
line.match(LogFormat.regex)
# ...
end

My question: How often is the Regexp compiled? When?
When the definition of LogFormat is read first?

Btw: If anybody has a ready-to-use regex for the common log format this
would be great, but I will get that done as far as I need by myself.
;-)
Other question: Does anybody know a "Webalizer" sort of thing written
in Ruby?

I tried to write something like that once.

But It was very slow (especially because it did DNS resolving
synchronously), and I never finished it.

Thanks

Michal

 

[Rick DeNatale]

So, I've corrected the benchmark. I've also added another method
using String#split as Eric suggested.

Here's the code being benchmarked:
rick@frodo:~/rubyscripts$ cat stringsplit.rb

Putting the code here involves an extra method call. While they are
all wrapped in a method call equally, doing less is always better.

class String
[methods]
end

Note that I made tested three different "pre-compiled" regex's one
with a Regex.new('.'), one with Regex.new(/./), and one a constant
with the literal regex /./.

You don't need three different benchmarks for this. Its easy to
determine that these are the same with irb.

Now here's the benchmark:
rick@frodo:~/rubyscripts$ cat benchstringsplit.rb
require 'benchmark'
include Benchmark
load 'stringsplit.rb'

iters = 100

100 iterations is never enough. GC behavior, other processes waking
up, etc. will all cause fluctuations in the benchmark.

Use 100_000 or 1_000_000.

[benchmark code]

Here's a more-meaningful version of yours:

[ Eric's benchmark code]

then a lot of other insightful stuff.

But ultimately, these types of microbenchmarks are not very useful.
Yes, you will get a speedup using /./ over RE, but will your program
really run long enough where it matters to make a difference? I bet
not.


Be careful with your benchmarks, they are most useful when they are
as small and simple as possible. Be sure to throw away all the
irrelevant parts.

Be careful with your benchmarks, they are most useless when they are
as small and simple as possible. Be sure to understand how little
speedup you'll get.

Which is the point after all. I think that we are actually in violent
agreement about the role of benchmarking.

The key thing is a useful benchmark is crafted to show the performance
at a functional level which affects the particular application.

My benchmarks came out of another thread in which I proposed some code
and someone made the observation that my use of unpack was a quick way
to split up a string into individual 1-character strings. I did a
benchmark to see if I could find a faster way, and when the results
seemed interesting to the "Compiling Regexp only once" thread, I
mentioned there, when it was pointed out that I had a typo which
invalidated that benchmark and re-did it.

I benchmarked several different approaches to splitting a string into
single character strings, a slightly higher level function than just a
regexp match. And the results indicate that at that level using
literal REs doesn't really make much difference.

Doing something a particular way by rote, because you heard X, and
assume that X both applies to your situation and still applies, might
get your code written, but might over time, turn out to be folk
wisdom.

I'm old enough to remember when some folks programmed in PL/I, the
early PL/I compiler did a very poor job of generating code for
subroutine calls, which led to PL/I "best practice" documents
recommending that subroutines be avoided at all costs! Subsequent
compilers made this harmful advice unnecessary. Of course avoiding
subroutines is a much bigger deal than exactly how best to represent a
regexp in source code.

The bottom line is that as is often said "premature optimization is
the root of all evil." Better to first write clearly, then test, and
then, if there is a performance problem, fix it by benchmarking,
profiling and re-coding/re-factoring.

I hope that we agree on that.

 

[Devin Mullins]

So literal regular expressions are "faster" by about 56% (after
discounting loop overhead) (when you're performing 100 million
retrievals only) (actual speedup in real code will probably be
swallowed elsewhere or completely irrelevant).

This follows from reading eval.c. Constant lookup involves a bunch of
C function calls, but literal lookup just returns an Object stored in
the parse tree.

Interesting. Note that this is different from your benchmark in
ruby-talk:204747, wherein much of the savings seems to come from not
having to instantiate & populate a MatchData. Compare:

require 'benchmark'

N = 1_000_000

string = 'a b c d e'

Benchmark.bmbm do |bm|

bm.report 'empty' do
N.times do end
end

bm.report 'none' do
N.times do string =~ /a (.) c\s. e/ end
end

bm.report '5' do
N.times do string =~ /a (.) c\s. e/;5 end
end

bm.report '$1' do
N.times do string =~ /a (.) c\s. e/;$1 end
end

bm.report '$&' do
N.times do string =~ /a (.) c\s. e/;$& end
end

bm.report '$~' do
N.times do string =~ /a (.) c\s. e/;$~ end
end

bm.report 'match' do
N.times do string.match /a (.) c\s. e/ end
end
end
__END__

Rehearsal -----------------------------------------
empty 0.141000 0.000000 0.141000 ( 0.204000)
none 1.453000 0.000000 1.453000 ( 1.500000)
5 1.578000 0.000000 1.578000 ( 1.625000)
$1 3.500000 0.015000 3.515000 ( 3.546000)
$& 3.640000 0.000000 3.640000 ( 3.875000)
$~ 4.641000 0.016000 4.657000 ( 4.922000)
match 5.313000 0.015000 5.328000 ( 5.422000)
------------------------------- total: 20.312000sec

user system total real
empty 0.156000 0.000000 0.156000 ( 0.156000)
none 1.484000 0.000000 1.484000 ( 1.531000)
5 1.563000 0.000000 1.563000 ( 1.610000)
$1 3.468000 0.000000 3.468000 ( 3.937000)
$& 3.484000 0.000000 3.484000 ( 3.875000)
$~ 4.609000 0.016000 4.625000 ( 4.641000)
match 5.219000 0.016000 5.235000 ( 5.313000)

Devin

 

[Ryan Davis]

Interesting. Note that this is different from your benchmark in
ruby-talk:204747, wherein much of the savings seems to come from
not having to instantiate & populate a MatchData. Compare:

Vastly different. He's calling scan/split instead of s =~ //. s =~ //
is the fastest way to invoke a match as you avoid the method dispatch
altogether.

 

[Devin Mullins]

Interesting. Note that this is different from your benchmark in

Vastly different. He's calling scan/split instead of s =~ //. s =~ //
is the fastest way to invoke a match as you avoid the method dispatch
altogether.

Hurh?

I'll admit up front that I'm absolutely clueless about benchmarking, but
didn't my last bench show that most of the savings with s =~ // doesn't
have to do with the method dispatch, but with MatchData?

(Sadly, in retrospect, it lacked any comparisons of the various ways to
*reference* a regex -- literal, constant, etc. -- which was the meat of
the bench to which I was responding. Oh well. Sleepyhead..)

Devin

 

[Robert Klemme]

Note though that it's usually faster to have a regexp in line. So in
your case you might have a method that does the line parsing (or
multiple line parsing) and that's where you can put the inline regexp
for max efficiency.

Robert,

I'm not sure what you mean by inline, do you mean a regexp literal like
/[ABC]/ vs. RegEx.new('/[ABC]/') if so that's not clear.

No, I meant that

if /foo/ =~ something

is faster than

FOO = /foo/
....
if FOO =~ something

Maybe the term "inline" was confusing. It refers to the fact that the
RX is defined at the place where it matches.

$ ruby rxbm.rb
Rehearsal ----------------------------------------------------
inline match 0.781000 0.000000 0.781000 ( 0.781000)
inline non match 1.500000 0.000000 1.500000 ( 1.500000)
const match 0.969000 0.000000 0.969000 ( 0.985000)
const non match 1.719000 0.000000 1.719000 ( 1.718000)
------------------------------------------- total: 4.969000sec

user system total real
inline match 0.797000 0.000000 0.797000 ( 0.797000)
inline non match 1.484000 0.000000 1.484000 ( 1.485000)
const match 1.000000 0.000000 1.000000 ( 1.000000)
const non match 1.719000 0.000000 1.719000 ( 1.719000)


(BM attached)

Kind regards

robert


require 'benchmark'
include Benchmark

REP = 1_000_000
RX = /foo/
MATCH = ("foo" * 100).freeze
NON_MATCH = ("fo" * 100).freeze

bmbm do |b|
b.report "inline match" do
REP.times do
/foo/ =~ MATCH
end
end

b.report "inline non match" do
REP.times do
/foo/ =~ NON_MATCH
end
end

b.report "const match" do
REP.times do
RX =~ MATCH
end
end

b.report "const non match" do
REP.times do
RX =~ NON_MATCH
end
end

end