Feature #14410: Reduce allocations and retained objects in the uri common module

• Status: Closed
• Assignee: -
• Target version: -

[sam.saffron (Sam Saffron)]

Patch at:

https://github.com/ruby/ruby/pull/1801

require 'memory_profiler'

MemoryProfiler.report do
require 'uri'
end.pretty_print

Before:

Total allocated: 986643 bytes (15159 objects)
Total retained: 246370 bytes (2532 objects)

After:

Total allocated: 926903 bytes (13665 objects)
Total retained: 208570 bytes (1587 objects)

---

Quite a significant change, almost 1000 less RVALUEs around when you require 'uri' which almost every web app does.

See also: https://ruby-talk.trydiscourse.com/t/psa-string-memory-use-reduction-techniques/74477/13?u=samsaffron

To explain [-i.chr]

[sam.saffron (Sam Saffron)]

• File deleted (diff.txt)

[sam.saffron (Sam Saffron)]

• File diff.txt diff.txt added

[normalperson (Eric Wong)]

[e-mail] wrote:

https://bugs.ruby-lang.org/issues/14410
https://github.com/ruby/ruby/pull/1801
diff.txt (1.03 KB)

Careful with "frozen_string_literal: true".

Even with tired eyes I see self.decode_www_form uses literals
which get modified, and maybe other places.

Did you try "make test-all"?

[sam.saffron (Sam Saffron)]

Yes, tests appear to be passing.

I went with the atomic version here cause there is a giant string Hash and it is way cleaner to apply it to the entire file then work through every string in the Hash

• Status changed from Open to Closed

Applied in changeset trunk|r62081.

---

uri/common: reduce allocations and retained objects

Thanks to Sam Saffron for this patch, it shows
a nice reduction which affects many web applications:

require 'memory_profiler'

MemoryProfiler.report do
require 'uri'
end.pretty_print

Before:

Total allocated: 986643 bytes (15159 objects)
Total retained: 246370 bytes (2532 objects)

After:

Total allocated: 926903 bytes (13665 objects)
Total retained: 208570 bytes (1587 objects)

• lib/uri/common.rb: reduce allocations and retained objects
  [ruby-core:85161] [Feature #14410]

[normalperson (Eric Wong)]

[e-mail] wrote:

I went with the atomic version here cause there is a giant string Hash and it is way cleaner to apply it to the entire file then work through every string in the Hash

Thanks, I looked over it again and committed as r62081