`ActiveSupport::Dependencies.constantize(const_name)` calls
`Reference.new` which is defined as
`ActiveSupport::Dependencies.constantize(const_name)` meaning this call
is already cached and we're doing caching that isn't necessary.
```ruby
require 'action_pack'
require 'action_dispatch'
require 'benchmark/ips'
route_set = ActionDispatch::Routing::RouteSet.new
routes = ActionDispatch::Routing::Mapper.new route_set
ObjectSpace::AllocationTracer.setup(%i{path line type})
result = ObjectSpace::AllocationTracer.trace do
500.times do
routes.resources :foo
end
end
sorted = ObjectSpace::AllocationTracer.allocated_count_table.sort_by(&:last)
sorted.each do |k,v|
next if v == 0
p k => v
end
__END__
Before:
{:T_SYMBOL=>11}
{:T_REGEXP=>17}
{:T_STRUCT=>6500}
{:T_MATCH=>12004}
{:T_OBJECT=>99009}
{:T_DATA=>100088}
{:T_HASH=>122015}
{:T_STRING=>159637}
{:T_IMEMO=>363134}
{:T_ARRAY=>433056}
After:
{:T_SYMBOL=>11}
{:T_REGEXP=>17}
{:T_STRUCT=>6500}
{:T_MATCH=>12004}
{:T_OBJECT=>91009}
{:T_DATA=>100088}
{:T_HASH=>114013}
{:T_STRING=>159637}
{:T_ARRAY=>321056}
{:T_IMEMO=>351133}
```
Eagerly calculate and cache the name of Symbol objects in the path AST.
This drops about 26 string allocations per resource:
```ruby
require 'action_pack'
require 'action_dispatch'
require 'benchmark/ips'
route_set = ActionDispatch::Routing::RouteSet.new
routes = ActionDispatch::Routing::Mapper.new route_set
ObjectSpace::AllocationTracer.setup(%i{path line type})
result = ObjectSpace::AllocationTracer.trace do
500.times do
routes.resources :foo
end
end
sorted = ObjectSpace::AllocationTracer.allocated_count_table.sort_by(&:last)
sorted.each do |k,v|
next if v == 0
p k => v
end
__END__
Before:
{:T_SYMBOL=>11}
{:T_REGEXP=>17}
{:T_STRUCT=>6500}
{:T_MATCH=>12004}
{:T_OBJECT=>99009}
{:T_DATA=>116084}
{:T_HASH=>122015}
{:T_STRING=>172647}
{:T_IMEMO=>371132}
{:T_ARRAY=>433056}
After:
{:T_SYMBOL=>11}
{:T_REGEXP=>17}
{:T_STRUCT=>6500}
{:T_MATCH=>12004}
{:T_OBJECT=>99009}
{:T_DATA=>100088}
{:T_HASH=>122015}
{:T_STRING=>159637}
{:T_IMEMO=>363134}
{:T_ARRAY=>433056}
```
Rather than building a regexp for every route, lets use the strategy
pattern to select among objects that can match HTTP verbs. This commit
introduces strategy objects for each verb that has a predicate method on
the request object like `get?`, `post?`, etc.
When we build the route object, look up the strategy for the verbs the
user specified. If we can't find it, fall back on string matching.
Using a strategy / null object pattern (the `All` VerbMatcher is our
"null" object in this case) we can:
1) Remove conditionals
2) Drop boot time allocations
2) Drop run time allocations
3) Improve runtime performance
Here is our boot time allocation benchmark:
```ruby
require 'action_pack'
require 'action_dispatch'
route_set = ActionDispatch::Routing::RouteSet.new
routes = ActionDispatch::Routing::Mapper.new route_set
result = ObjectSpace::AllocationTracer.trace do
500.times do
routes.resources :foo
end
end
sorted = ObjectSpace::AllocationTracer.allocated_count_table.sort_by(&:last)
sorted.each do |k,v|
next if v == 0
p k => v
end
__END__
Before:
$ be ruby -rallocation_tracer route_test.rb
{:T_SYMBOL=>11}
{:T_REGEXP=>4017}
{:T_STRUCT=>6500}
{:T_MATCH=>12004}
{:T_DATA=>84092}
{:T_OBJECT=>99009}
{:T_HASH=>122015}
{:T_STRING=>216652}
{:T_IMEMO=>355137}
{:T_ARRAY=>441057}
After:
$ be ruby -rallocation_tracer route_test.rb
{:T_SYMBOL=>11}
{:T_REGEXP=>17}
{:T_STRUCT=>6500}
{:T_MATCH=>12004}
{:T_DATA=>84092}
{:T_OBJECT=>99009}
{:T_HASH=>122015}
{:T_STRING=>172647}
{:T_IMEMO=>355136}
{:T_ARRAY=>433056}
```
This benchmark adds 500 resources. Each resource has 8 routes, so it
adds 4000 routes. You can see from the results that this patch
eliminates 4000 Regexp allocations, ~44000 String allocations, and ~8000
Array allocations. With that, we can figure out that the previous code
would allocate 1 regexp, 11 strings, and 2 arrays per route *more* than
this patch in order to handle verb matching.
Next lets look at runtime allocations:
```ruby
require 'action_pack'
require 'action_dispatch'
require 'benchmark/ips'
route_set = ActionDispatch::Routing::RouteSet.new
routes = ActionDispatch::Routing::Mapper.new route_set
routes.resources :foo
route = route_set.routes.first
request = ActionDispatch::Request.new("REQUEST_METHOD" => "GET")
result = ObjectSpace::AllocationTracer.trace do
500.times do
route.matches? request
end
end
sorted = ObjectSpace::AllocationTracer.allocated_count_table.sort_by(&:last)
sorted.each do |k,v|
next if v == 0
p k => v
end
__END__
Before:
$ be ruby -rallocation_tracer route_test.rb
{:T_MATCH=>500}
{:T_STRING=>501}
{:T_IMEMO=>1501}
After:
$ be ruby -rallocation_tracer route_test.rb
{:T_IMEMO=>1001}
```
This benchmark runs 500 calls against the `matches?` method on the route
object. We check this method in the case that there are two methods
that match the same path, but they are differentiated by the verb (or
other conditionals). For example `POST /users` vs `GET /users`, same
path, different action.
Previously, we were using regexps to match against the verb. You can
see that doing the regexp match would allocate 1 match object and 1
string object each time it was called. This patch eliminates those
allocations.
Next lets look at runtime performance.
```ruby
require 'action_pack'
require 'action_dispatch'
require 'benchmark/ips'
route_set = ActionDispatch::Routing::RouteSet.new
routes = ActionDispatch::Routing::Mapper.new route_set
routes.resources :foo
route = route_set.routes.first
match = ActionDispatch::Request.new("REQUEST_METHOD" => "GET")
no_match = ActionDispatch::Request.new("REQUEST_METHOD" => "POST")
Benchmark.ips do |x|
x.report("match") do
route.matches? match
end
x.report("no match") do
route.matches? no_match
end
end
__END__
Before:
$ be ruby -rallocation_tracer runtime.rb
Calculating -------------------------------------
match 17.145k i/100ms
no match 24.244k i/100ms
-------------------------------------------------
match 259.708k (± 4.3%) i/s - 1.303M
no match 453.376k (± 5.9%) i/s - 2.279M
After:
$ be ruby -rallocation_tracer runtime.rb
Calculating -------------------------------------
match 23.958k i/100ms
no match 29.402k i/100ms
-------------------------------------------------
match 465.063k (± 3.8%) i/s - 2.324M
no match 691.956k (± 4.5%) i/s - 3.469M
```
This tests tries to see how many times it can match a request per
second. Switching to method calls and string comparison makes the
successful match case about 79% faster, and the unsuccessful case about
52% faster.
That was fun!
verb matching is very common (all routes besides rack app endpoints
require one). We will extract verb matching for now, and use a more
efficient method of matching (then regexp) later
This commit introduces a functional Path AST visitor and implements
`each` on the AST in terms of the functional visitor. The functional
visitor doesn't maintain state, so we only need to allocate one of them.
Given this benchmark route file:
```ruby
require 'action_pack'
require 'action_dispatch'
route_set = ActionDispatch::Routing::RouteSet.new
routes = ActionDispatch::Routing::Mapper.new route_set
ObjectSpace::AllocationTracer.setup(%i{path line type})
result = ObjectSpace::AllocationTracer.trace do
500.times{|i|
routes.resource :omglol
}
end
result.find_all { |k,v| k.first =~ /git\/rails/ }.sort_by { |k,v|
v.first
}.each { |k,v|
p k => v
}
```
node.rb line 17 was in our top 3 allocation spot:
```
{["/Users/aaron/git/rails/actionpack/lib/action_dispatch/journey/nodes/node.rb", 17, :T_OBJECT]=>[31526, 0, 28329, 0, 2, 1123160]}
{["/Users/aaron/git/rails/actionpack/lib/action_dispatch/routing/mapper.rb", 2080, :T_IMEMO]=>[34002, 0, 30563, 0, 2, 1211480]}
{["/Users/aaron/git/rails/actionpack/lib/action_dispatch/routing/mapper.rb", 2071, :T_IMEMO]=>[121934, 1, 109608, 0, 7, 4344400]}
```
This commit eliminates allocations at that place.
I would like to change the signature of the Route constructor. Since
the mapping object has all the data required to construct a Route
object, move the allocation to a factory method.
We should build the routes using the user facing API which is `Mapper`.
This frees up the library internals to change as we see fit. IOW we
shouldn't be testing internals.
This is part of a larger refactoring on controller tests. We needed to
move these methods here so that we could get rid of the `|| key ==
:action || key == :controller` in `assign_parameters`. We know this is
ugly and intend to fix it but for now `generate_extras` needs to be used
in the two methods to access the path and the query_string_keys.
We're adding `:controller` and `:action` to the `query_string_keys`
because we always need a controller and action. If someone passed
`action` or `controller` in in there test they are unambigious - we
know they have to go into the query params.
if `to` was initialized, this method would return, so we can eliminate
the to ||= in the conditional. Finally, let's return a nil in the else
block so that it's explicit that this method can return nil
We don't need a method for something like this. I want to pull this up
the stack as well and move the module + :controller ArgumentError up the
stack as well
the same value that is extracted from the options hash earlier is
returned, so we don't need to pass it in in the first place. The caller
already has the data, so stop passing it around.
These three options are stored in the `scope` chain outside of the
options hash. If they are in the options hash, then someone passed them
in to `match` and they don't really do anything. So lets remove the
code.
Remove the `options` reader from `Resource` because nobody needs to see
that hash. Also remove mutations on the options hash in
`apply_common_behavior_for` because leaving the side effects in that
method makes it difficult to understand what is going on in the caller.
these two keys have a different merge strategy, and they also just get
removed from the options hash later in the code. If we store them in a
separate place, then we don't need to remove them later
We're going to try pulling this up further, and check `via` validity
sooner. This way we don't have to do a bunch of processing on `options`
hashes only to find out that the route is incorrect
I don't want `split_constraints` to mutate any instance variables. That
way it's easier to move the method around and understand what it does
(it has no side effects)
we want to try to pull this logic up to where the user actually passed
in "controller" so that it's close to the related call. That way when
we're down the stack, we don't need to wonder "why are we doing this?"
All callers of `action_path` interpolate the return value in to a
string, so there is no need for the method to to_s it. to_sym on a
symbol will return the same symbol, though I think `action_path` may
always be called with a symbol so this might not be necessary.
eliminates calling `scope` in one method, pushes the other calls up one
frame. This goes a little way towards eliminating the internal calls to
`scope`.
we need to get a grip on what `scope` actually does. This commit
removes some of the internal calls to `scope`. Eventually we should add
public facing methods that provide the API that `scope` is trying to
accomplish.
`prepare_params!` would raise an exception if `params` wasn't
initialized, so it must always be available. Remove the existence
conditional from the `controller` method.
`Dispatcher` doesn't need to hold on to the defaults hash. It only used
the hash to determine whether or not it should raise an exception if
there is a name error. We can pass that in further up the stack and
alleviate Dispatcher from knowing about that hash.
We know in advance whether the object is a dispatcher or not, so we can
configure the Constraints object with a strategy that will call the
right method.
The tests and methods were hard to read with `options[:options]` all
over the place. This refactoring makes the code easier to understand.
The change came out of work for moving the underlying code of controller
tests to integraiton tests.
Using the string version of the class reference is now deprecated when
referencing middleware. This should be written as a class not as a string.
Deprecation warning that this change fixes:
```
DEPRECATION WARNING: Passing strings or symbols to the middleware
builder is deprecated, please change
them to actual class references. For example:
"ActionDispatch::ShowExceptions" => ActionDispatch::ShowExceptions
```
`extra_keys` is a confusing variable name because it's not clear what is
"extra". This renames it to `query_string_keys` so it's clear that the
"extra" is just the query string.
We were doing extra work that could be pushed off to Integration test
and SharedRoutes. Creating an extra module isn't necessary when those
are created by their respective classes.
people should be accessing request information through the request
object, not via the env hash. If they really really want at the env
hash, then they can get it off the request.
Actions are processed through `dispatch`, so they should have the
request set on them before any user land code can be executed. Lets
stop setting _env on the controller, and give access to it through the
`env` method.
We should do the hard work outside the constructor. Also fix the tests
to not directly construct middleware objects, but to go through the
stack object.
Proc.new will pick up the passed in block, but since it's a default
param, it won't get evaluated unless someone doesn't pass in an app. It
will raise an exception if no block is provided.
again, we want to hide the contents of `env` from the implementation.
Allocate a request object to access the contents of env, but save
allocations due to string literal allocations when accessing the env
hash.
ExceptionWrapper only cares about the backtrace cleaner, so lets just
pass the cleaner to the wrapper. It does not need to know that env
exists or what key the backtrace cleaner is stored in
Implement `serve` on the middleware. Nothing can be placed between the
instance of FileHandler and Static because Static instantiates an
instance of FileHandler. IOW there is no reason to implement the `call`
API in this case.
The cookie jar can just ask the request object for the information it
needs. This allows us to stop allocating hashes for options, and also
allows us to delay calculating values in advance. Generating the
options hash forced us to calculate values that we may never have needed
at runtime
Accessing a request object has nice advantages over accessing a hash.
If you use a missing method name, you'll get an exception rather than a
`nil` (is one nice feature)
While the readability may be slightly worse, the speed improvement is
significant: Twice as fast when there's no leading "/" to remove, and
over 4 times as fast when there is a leading "/".
Benchmark:
require 'benchmark/ips'
def match(controller)
if controller
if m = controller.match(/\A\/(?<controller_without_leading_slash>.*)/)
m[:controller_without_leading_slash]
else
controller
end
end
end
def start_with(controller)
if controller
if controller.start_with?('/'.freeze)
controller[1..-1]
else
controller
end
end
end
Benchmark.ips do |x|
x.report("match") { match("no_leading_slash") }
x.report("start_with") { start_with("no_leading_slash") }
x.compare!
end
Benchmark.ips do |x|
x.report("match") { match("/a_leading_slash") }
x.report("start_with") { start_with("/a_leading_slash") }
x.compare!
end
Result (Ruby 2.2.2):
Calculating -------------------------------------
match 70.324k i/100ms
start_with 111.264k i/100ms
-------------------------------------------------
match 1.468M (± 7.1%) i/s - 7.314M
start_with 3.787M (± 3.5%) i/s - 18.915M
Comparison:
start_with: 3787389.4 i/s
match: 1467636.4 i/s - 2.58x slower
Calculating -------------------------------------
match 36.694k i/100ms
start_with 86.071k i/100ms
-------------------------------------------------
match 532.795k (± 4.7%) i/s - 2.679M
start_with 2.518M (± 5.8%) i/s - 12.566M
Comparison:
start_with: 2518366.8 i/s
match: 532794.5 i/s - 4.73x slower
Add descriptions about `ActiveRecord::Base#to_param` to
* `ActionDispatch::Routing::Base#match`
* Overriding Named Route Parameters (guide)
When passes `:param` to route definision, always `to_param` method of
related model is overridden to constructe an URL by passing these
model instance to named_helper.
We don't always need an array when generating a url with the formatter. We can be lazy about allocating the `missing_keys` array. This saves us:
35,606 bytes and 889 objects per request
THe only reason we were allocating an array is to get the "missing_keys" variable in scope of the error message generator. Guess what? Arrays kinda take up a lot of memory, so by replacing that with a nil, we save:
35,303 bytes and 886 objects per request
If we don't mutate the `recall` hash, then there's no reason to duplicate it. While this change doesn't get rid of that many objects, each hash object it gets rid of was massive.
Saves 888 string objects per request, 206,013 bytes (thats 0.2 mb which is kinda a lot).
Instead of calling `sub` on every link_to call for controller, we can detect when the string __needs__ to be allocated and only then create a new string (without the leading slash), otherwise, use the string that is given to us.
Saves 888 string objects per request, 35,524 bytes.
When `defaults[key]` in `generate` in the journey formatter is called, it often returns a `nil` when we call `to_s` on a nil, it allocates an empty string. We can skip this check when the default value is nil.
This change buys us 35,431 bytes of memory and 887 fewer objects per request.
Thanks to @matthewd for help with the readability
When generating a url with `url_for` the hash of arguments passed in, is dup-d and merged a TON. I wish I could clean this up better, and might be able to do it in the future. This change removes one dup, since it's literally right after we just dup-d the hash to pass into this constructor.
This may be a breaking, change but the tests pass...so :shipit: we can revert if it causes problems
This change buys us 205,933 bytes of memory and 887 fewer objects per request.
In handle_positional_args `Array#-=` is used which allocates a new array. Instead we can iterate through and delete elements, modifying the array in place.
Also `Array#take` allocates a new array. We can build the same by iterating over the other element.
This change buys us 106,470 bytes of memory and 2,663 fewer objects per request.
Most routes have a `route.path.requirements[key]` of `/[-_.a-zA-Z0-9]+\/[-_.a-zA-Z0-9]+/` yet every time this method is called a new regex is generated on the fly with `/\A#{DEFAULT_INPUT}\Z/`. OBJECT ALLOCATIONS BLERG!
This change uses a special module that implements `===` so it can be used in a case statement to pull out the default input. When this happens, we use a pre-generated regex.
This change buys us 1,643,465 bytes of memory and 7,990 fewer objects per request.
Micro optimization: `reverse.drop_while` is slower than `reverse_each.drop_while`. This doesn't save any object allocations.
Second, `keys_to_keep` is typically a very small array. The operation `parameterized_parts.keys - keys_to_keep` actually allocates two arrays. It is quicker (I benchmarked) to iterate over each and check inclusion in array manually.
This change buys us 1774 fewer objects per request
The request.script_name is dup-d which allocates an extra string. It is most commonly an empty string "". We can save a ton of string allocations by checking first if the string is empty, if so we can use a frozen empty string instead of duplicating an empty string.
This change buys us 35,714 bytes of memory and 893 fewer objects per request.
This feature also works with `PUT`, `PATCH` and `DELETE` requests.
Also developers can add `:url_encoded_form` and `:multipart_form`
into the `:format` for wrapping url encoded or multipart form data.
there is no reason to `convert_hashes_to_parameters` with an assignemt
flag. The caller knows whether or not it wants the value assigned. We
should just change the uncommon case (not writing to the underlying
hash) to just call the conversion method and return that value.
only hashes are converted to parameter objects, so lets add a branch for
them. This also removes a is_a? test for Parameters so we can be
abstracted from the class.
This clears the transaction record state when the transaction finishes
with a `:committed` status.
Considering the following example where `name` is a required attribute.
Before we had `new_record?` returning `true` for a persisted record:
```ruby
author = Author.create! name: 'foo'
author.name = nil
author.save # => false
author.new_record? # => true
```
I wrote a utility that helps find areas where you could optimize your program using a frozen string instead of a string literal, it's called [let_it_go](https://github.com/schneems/let_it_go). After going through the output and adding `.freeze` I was able to eliminate the creation of 1,114 string objects on EVERY request to [codetriage](codetriage.com). How does this impact execution?
To look at memory:
```ruby
require 'get_process_mem'
mem = GetProcessMem.new
GC.start
GC.disable
1_114.times { " " }
before = mem.mb
after = mem.mb
GC.enable
puts "Diff: #{after - before} mb"
```
Creating 1,114 string objects results in `Diff: 0.03125 mb` of RAM allocated on every request. Or 1mb every 32 requests.
To look at raw speed:
```ruby
require 'benchmark/ips'
number_of_objects_reduced = 1_114
Benchmark.ips do |x|
x.report("freeze") { number_of_objects_reduced.times { " ".freeze } }
x.report("no-freeze") { number_of_objects_reduced.times { " " } }
end
```
We get the results
```
Calculating -------------------------------------
freeze 1.428k i/100ms
no-freeze 609.000 i/100ms
-------------------------------------------------
freeze 14.363k (± 8.5%) i/s - 71.400k
no-freeze 6.084k (± 8.1%) i/s - 30.450k
```
Now we can do some maths:
```ruby
ips = 6_226k # iterations / 1 second
call_time_before = 1.0 / ips # seconds per iteration
ips = 15_254 # iterations / 1 second
call_time_after = 1.0 / ips # seconds per iteration
diff = call_time_before - call_time_after
number_of_objects_reduced * diff * 100
# => 0.4530373333993266 miliseconds saved per request
```
So we're shaving off 1 second of execution time for every 220 requests.
Is this going to be an insane speed boost to any Rails app: nope. Should we merge it: yep.
p.s. If you know of a method call that doesn't modify a string input such as [String#gsub](b0e2da69f0/lib/let_it_go/core_ext/string.rb (L37)) please [give me a pull request to the appropriate file](b0e2da69f0/lib/let_it_go/core_ext/string.rb (L37)), or open an issue in LetItGo so we can track and freeze more strings.
Keep those strings Frozen
![](https://www.dropbox.com/s/z4dj9fdsv213r4v/let-it-go.gif?dl=1)
When executing an `ActionController::Parameters#fetch` with a block
that raises a `KeyError` the raised `KeyError` will be rescued and
converted to an `ActionController::ParameterMissing` exception,
covering up the original exception.
[Jonas Schubert Erlandsson & Roque Pinel]
This will silence deprecation warnings.
Most of the test can be changed from `render :text` to render `:plain`
or `render :body` right away. However, there are some tests that needed
to be fixed by hand as they actually assert the default Content-Type
returned from `render :body`.
We've started on discouraging the usage of `render :text` in #12374.
This is a follow-up commit to make sure that we print out the
deprecation warning.
Now that the value is cached on the stack,
`array_of_permitted_scalars_filter` is exactly the same as
`array_of_permitted_scalars?`, so lets just have one
this way the method doesn't have to know what the new params object is,
it just yields to a block. This change also caches the value of
`self[key]` on the stack
We should disconnect `array_of_permitted_scalars_filter` from the
instance so that we can make hash filtering functional. For now, pull
the conditional up out of that method
`ActionController::Parameters#to_h` returns a hash, so lets have
`ActionController::Parameters#to_unsafe_h` return a hash instead of
an `ActiveSupport::HashWithIndifferentAccess` for consistency.
This is another take at #14384 as we decided to wait until `master` is
targeting Rails 5.0. This commit is implementation-complete, as it
guarantees that all the public methods on the hash-inherited Parameters
are still working (based on test case). We can decide to follow-up later
if we want to remove some methods out from Parameters.