[ruby-core:81210] [Ruby trunk Bug#12373] Optimize CSV#shift

Issue #12373 has been updated by ksss (Yuki Kurihara).


@hsbt

Thank you for responding.
csv-shift-hsbt.patch looks good to me.

> Should we use @double_quote_char instead of @quote_char ?
> your patch named csv-shift-3.patch is only assign @double_quote_char. It's not used in csv.rb

@double_quote_char and @quote_char are different values. It can not be used instead.
@double_quote_char looks like using in CSV#shift method on csv-shift-3.patch and csv-shift-hsbt.patch.


----------------------------------------
Bug #12373: Optimize CSV#shift
https://bugs.ruby-lang.org/issues/12373#change-64863

* Author: ksss (Yuki Kurihara)
* Status: Assigned
* Priority: Normal
* Assignee: hsbt (Hiroshi SHIBATA)
* Target version: 2.5
* ruby -v: ruby 2.4.0dev (2016-05-11 trunk 54976) [x86_64-darwin15]
* Backport: 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN
----------------------------------------
I think that **`str1.start_with?(str2)`** is faster than **`str1[0] == str2`**.
Because **`str1.start_with?(str2)`** just call **`String#start_with?`**, But **`str1[0] == str2`** call **`String#[]`**, make new `String` object and call **`String#==`**.

(The patch is csv-shift.patch)

Benchmark results.

csv-benchmark.rb make temp CSV file and call **`CSV#each`** method(inner call **`CSV#shift`**)

~~~
$ ruby csv-benchmark.rb
Warming up --------------------------------------
       old_csv_shift     1.000  i/100ms
       new_csv_shift     1.000  i/100ms
Calculating -------------------------------------
       old_csv_shift      0.444  (賊 0.0%) i/s -      3.000  in   6.759200s
       new_csv_shift      0.479  (賊 0.0%) i/s -      3.000  in   6.264069s

Comparison:
       new_csv_shift:        0.5 i/s
       old_csv_shift:        0.4 i/s - 1.08x slower
~~~

string-start_with.rb is a micro benchmark for **`str1[0] == str2`** and **`str1.start_with?(str2)`**

~~~
$ ruby string-start_with.rb
Warming up --------------------------------------
           a[0] == b    90.881k i/100ms
    a.start_with?(b)   115.557k i/100ms
Calculating -------------------------------------
           a[0] == b      1.836M (賊 3.8%) i/s -      9.179M in   5.006568s
    a.start_with?(b)      3.183M (賊 4.2%) i/s -     15.947M in   5.018654s

Comparison:
    a.start_with?(b):  3183386.0 i/s
           a[0] == b:  1836263.5 i/s - 1.73x slower
~~~

Of course $ make test-all TESTS="test/csv/*" passed


---Files--------------------------------
csv-benchmark.rb (561 Bytes)
old_csv_shift.rb (5.1 KB)
new_csv_shift.rb (5.11 KB)
string-start_with.rb (215 Bytes)
csv-shift.patch (434 Bytes)
csv-shift-1.patch (1.15 KB)
csv-shift-2.patch (1.2 KB)
csv-shift-3.patch (2.13 KB)
csv-shift-hsbt.path (2.05 KB)


-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>