Reduce allocations from URLFilters

[ashmaroli]

Summary

Many temporary objects are repeatedly allocated within relative_url and absolute_url.
This PR attempts to reduce allocating such temporary objects.

In the following lines:

jekyll/lib/jekyll/filters/url_filters.rb

Lines 35 to 39 in 8673cdc

	return input if Addressable::URI.parse(input.to_s).absolute?
	parts = [sanitized_baseurl, input]
	Addressable::URI.parse(
	parts.compact.map { |part| ensure_leading_slash(part.to_s) }.join

• ln#35 calls input.to_s but does not store the result
• ln#37 initializes a two-member array on each call to #relative_url
• ln#39 duplicates above array via non-destructive compact and non-destructive map with each member being subject to to_s calls.

With the proposed changes, the intermediate array objects and the numerous calls to to_s have been eliminated.

API Changes

• Removed
  • private method URLFilters#sanitized_baseurl removed. Classes mixing-in the Jekyll::Filters module may need to update to using site.baseurl directly.
• Added
  • site now responds to :sanitized_baseurl and :sanitized_url (getters)
  • Utility method Utils.strip_leading_or_trailing_slashes
• Changed
  • relative_url and absolute_url now handles multiple leading or trailing slashes in both config["baseurl"] and config["url"]

Context

Additionally resolves #6947

[mattr-]

I do not want to introduce another breaking change in the 4.0 release cycle. Let's write a custom baseurl= that does the sanitization for us when the new baseurl is set, keeping us from removing that bit of public API. The rest I'm happy to keep, since folks shouldn't be using private methods anyways.

[ashmaroli]

Profiler summary

--- master branch https://travis-ci.org/jekyll/jekyll/jobs/532959217
+++ PR branch     https://travis-ci.org/jekyll/jekyll/jobs/532976771

- Total allocated: 532.65 MB (5301707 objects)
- Total retained:  18.34 MB (96237 objects)
+ Total allocated: 532.22 MB (5295029 objects)
+ Total retained:  18.34 MB (96242 objects)

[mattr-]

I'm not sure how I feel about adding new public attr_readers to Jekyll::Site, since they'll become part of the public API.

Are there any disadvantages to always returning a sanitized baseurl and a sanitized url?

[ashmaroli]

The disadvantage is unnecessary allocations. Technically, I'm only using site.sanitized_baseurl. The sanitized_url is just being provided for completeness.

Since you're on the fence about this, let us wait and see if the optimizations make a difference with a future state of master

[DirtyF]

Do we need to revisit this after #7793 ?

[ashmaroli]

Do we need to revisit this after #7793 ?

No. I don't think it's worth the added complexity.
Closing..