[Harry Brundage & Xavier Noria]

Possible thanks to Zeitwerk 1.3.0.

[ci skip]

Rails.autoloader and Rails.once_autoloader was just tentative API good
enough for a first patch. Rails.autoloader is singular and does not
convey in its name that there is another autoloader. That might be
confusing, for example if you set a logger and miss traces. On the other
hand, the name `once_autoloader` is very close to being horrible.

Rails.autoloaders.main and Rails.autoloaders.once read better for my
taste, and have a nice symmetry. Also, both "main" and "once" are four
letters long, short and same length.

They are tagged as "rails.main" and "rails.once", respectively.

References #35235.

Both methods were introduced in Ruby 2.5 and fit this use case very
nicely: listing a directory's entries excluding the "." and ".." nodes.

The private method #exclude_from was removed as it no longer serves
its original purpose.

I believe the current style does not clearly communicate that we are
ignoring non-existing autoload paths altogether. Your eyes may even be
looking for an else clause that is easy to miss but that does not exist.

With the early `next` and code comment the loop reads better for my
taste.

`object.transform_values!` returns `object` itself.

I've found a few places in Rails code base where I think it makes sense
to calculate elapsed time more precisely by using
`Concurrent.monotonic_time`:

- Fix calculation of elapsed time in `ActiveSupport::Cache::MemoryStore#prune`
- Fix calculation of elapsed time in
  `ActiveRecord::ConnectionAdapters::ConnectionPool::Queue#wait_poll`
- Fix calculation of elapsed time in
  `ActiveRecord::ConnectionAdapters::ConnectionPool#attempt_to_checkout_all_existing_connections`
- Fix calculation of elapsed time in `ActiveRecord::ConnectionAdapters::Mysql2Adapter#explain`

See
https://docs.ruby-lang.org/en/2.5.0/Process.html#method-c-clock_gettime
https://blog.dnsimple.com/2018/03/elapsed-time-with-ruby-the-right-way

Related to 7c454214

When adding/removing a subscription with a string pattern, it isn't
necessary to clear the entire cache, only the cache for the key being
added.

When adding/removing subscriptions for a regex or to match all events,
the full cache is still cleared.

Previously we stored all subscribers in an array, and every time a new
message name came in asked each subscriber if they responded to the
message.

This commit changes Fanout to hash subscribers with a String pattern by
their pattern. This way we can look them up directly and only do the
O(N) scan over the non-string (Regex or any) patterns.

ActiveSupport overrides `` Kernel#` `` so that it would not raise
`Errno::ENOENT` but return `nil` instead (due to the last statement
`STDERR.puts` returning nil) if a given command were not found.
Because of this, you cannot safely say somthing like
`` `command`.chomp `` when ActiveSupport is loaded.

It turns out that this is an outdated monkey patch for Windows
platforms to emulate Unix behavior on an ancient version of Ruby, and
it should be removed by now.

This is useful when we need to do some work associated to `Current.reset`
but that work depends on the values of the current attributes themselves.
This cannot be done in the supported `resets` callback because when the
block is executed, CurrentAttributes's instance has already been reset.

For symmetry, `after_reset` is defined as alias of `resets`.

Memory comparison:

Options: {:years=>1, :months=>1, :weeks=>1, :days=>1}
Calculating -------------------------------------
              master   576.000  memsize (     0.000  retained)
                         5.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)
      advance_no_dup   288.000  memsize (     0.000  retained)
                         4.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)

Comparison:
      advance_no_dup:        288 allocated
              master:        576 allocated - 2.00x more

Options: {:years=>1}
Calculating -------------------------------------
              master   264.000  memsize (     0.000  retained)
                         2.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)
      advance_no_dup    72.000  memsize (     0.000  retained)
                         1.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)

Comparison:
      advance_no_dup:         72 allocated
              master:        264 allocated - 3.67x more

Options: {:weeks=>1}
Calculating -------------------------------------
              master   264.000  memsize (     0.000  retained)
                         2.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)
      advance_no_dup    72.000  memsize (     0.000  retained)
                         1.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)

Comparison:
      advance_no_dup:         72 allocated
              master:        264 allocated - 3.67x more

Peformance comparison:

Options: {:years=>1, :months=>1, :weeks=>1, :days=>1}
Warming up --------------------------------------
              master    27.740k i/100ms
      advance_no_dup    37.705k i/100ms
Calculating -------------------------------------
              master    338.511k (± 5.9%) i/s -      1.692M in   5.020333s
      advance_no_dup    572.980k (± 3.7%) i/s -      2.866M in   5.008680s

Comparison:
      advance_no_dup:   572979.7 i/s
              master:   338510.9 i/s - 1.69x  slower

Options: {:years=>1}
Warming up --------------------------------------
              master    53.313k i/100ms
      advance_no_dup   115.016k i/100ms
Calculating -------------------------------------
              master    639.715k (± 1.7%) i/s -      3.199M in   5.001851s
      advance_no_dup      1.579M (± 6.4%) i/s -      7.936M in   5.053876s

Comparison:
      advance_no_dup:  1579251.7 i/s
              master:   639714.8 i/s - 2.47x  slower

Options: {:weeks=>1}
Warming up --------------------------------------
              master    57.353k i/100ms
      advance_no_dup   129.141k i/100ms
Calculating -------------------------------------
              master    674.113k (± 3.4%) i/s -      3.384M in   5.025973s
      advance_no_dup      1.911M (± 2.5%) i/s -      9.556M in   5.004496s

Comparison:
      advance_no_dup:  1910739.3 i/s
              master:   674112.6 i/s - 2.83x  slower

Now that secret_token was removed all this code is now dead.

* Calculate first month of quarter instead of finding

* Calculate last month of quarter instead of finding

[Krzysztof Rybka + Rafael Mendonça França]

### Summary

There was an issues when using `safe_constantize` on a string that has
the wrong case.

File `em.rb` defines `EM`.
`"Em".safe_constantize` causes a little confusion with the autoloader.
The autoloader finds file "em.rb",
expecting it to define `Em`, but `Em` is not defined.
The autoloader raises a `LoadError`, which is good,
But `safe_constantize` is defined to return `nil` when a class is not found.

### Before

```
"Em".safe_constantize
LoadError: Unable to autoload constant Em, \
expected rails/activesupport/test/autoloading_fixtures/em.rb to define it
```

### After

```
"Em".safe_constantize
# => nil
```

This speeds up Range.new(x, y).step(Duration).each { ... }

Fixes #34888

Since #34864 removed explicit receiver to clarify the
purpose of `delegate_missing_to`, I think it will be
better to do the same a few lines above to easier figure
out that `delegate_missing_to` defines `method_missing`,
`respond_to_missing?` when comparing these examples.

Removing the explicit receiver clarifies the purpose of `delegate_missing_to`.

[ci skip]

Accommodate case-insensitive filesystems and database collations.

fetch_multi(*names) now returns its results in the same order
as the `*names` requested, rather than returning cache hits
followed by cache misses.

`String#to_d` does not raise an exception if an invalid value is specified.
So can remove exception handling.

```
$ bundle exec ruby -v -rbigdecimal -rbigdecimal/util -e 'p "123,003".to_d'
ruby 2.6.0dev (2018-12-21 trunk 66474) [x86_64-linux]
0.123e3
```

Currently we sometimes find a redundant begin block in code review
(e.g. https://github.com/rails/rails/pull/33604#discussion_r209784205).

I'd like to enable `Style/RedundantBegin` cop to avoid that, since
rescue/else/ensure are allowed inside do/end blocks in Ruby 2.5
(https://bugs.ruby-lang.org/issues/12906), so we'd probably meets with
that situation than before.