1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
|
# frozen_string_literal: true
# Description of the structure
# ----------------------------
#
# This class emulates a hash table whose keys are of type Zeitwerk::Cref.
#
# It is a synchronized 2-level hash.
#
# The keys of the top one, stored in `@map`, are class and module objects, but
# their hash code is forced to be their object IDs because class and module
# objects may not be hashable (https://github.com/fxn/zeitwerk/issues/188).
#
# Then, each one of them stores a hash table with their constants and values.
# Constants are stored as symbols.
#
# For example, if we store values 0, 1, and 2 for the crefs that would
# correspond to `M::X`, `M::Y`, and `N::Z`, the map will look like this:
#
# { M => { X: 0, :Y => 1 }, N => { Z: 2 } }
#
# This structure is internal, so only the needed interface is implemented.
#
# Alternative approaches
# -----------------------
#
# 1. We could also use a 1-level hash whose keys are constant paths. In the
# example above it would be:
#
# { "M::X" => 0, "M::Y" => 1, "N::Z" => 2 }
#
# The gem used this approach for several years.
#
# 2. Write a custom `hash`/`eql?` in Zeitwerk::Cref. Hash code would be
#
# real_mod_hash(@mod) ^ @cname.hash
#
# where `real_mod_hash(@mod)` would actually be a call to the real `hash`
# method in Module. Like what we do for module names to bypass overrides.
#
# 3. Similar to 2, but use
#
# @mod.object_id ^ @cname.object_id
#
# as hash code instead.
#
# Benchamrks
# ----------
#
# Writing:
#
# map - baseline
# (3) - 1.74x slower
# (2) - 2.91x slower
# (1) - 3.87x slower
#
# Reading:
#
# map - baseline
# (3) - 1.99x slower
# (2) - 2.80x slower
# (1) - 3.48x slower
#
# Extra ball
# ----------
#
# In addition to that, the map is synchronized and provides `delete_mod_cname`,
# which is ad-hoc for the hot path in `const_added`, we do not need to create
# unnecessary cref objects for constants we do not manage (but we do not know in
# advance there).
#: [Value]
class Zeitwerk::Cref::Map # :nodoc: all
#: () -> void
def initialize
@map = {}
@map.compare_by_identity
@mutex = Mutex.new
end
#: (Zeitwerk::Cref, Value) -> Value
def []=(cref, value)
@mutex.synchronize do
cnames = (@map[cref.mod] ||= {})
cnames[cref.cname] = value
end
end
#: (Zeitwerk::Cref) -> Value?
def [](cref)
@mutex.synchronize do
@map[cref.mod]&.[](cref.cname)
end
end
#: (Zeitwerk::Cref, { () -> Value }) -> Value
def get_or_set(cref, &block)
@mutex.synchronize do
cnames = (@map[cref.mod] ||= {})
cnames.fetch(cref.cname) { cnames[cref.cname] = block.call }
end
end
#: (Zeitwerk::Cref) -> Value?
def delete(cref)
delete_mod_cname(cref.mod, cref.cname)
end
# Ad-hoc for loader_for, called from const_added. That is a hot path, I prefer
# to not create a cref in every call, since that is global.
#
#: (Module, Symbol) -> Value?
def delete_mod_cname(mod, cname)
@mutex.synchronize do
if cnames = @map[mod]
value = cnames.delete(cname)
@map.delete(mod) if cnames.empty?
value
end
end
end
#: (Value) -> void
def delete_by_value(value)
@mutex.synchronize do
@map.delete_if do |mod, cnames|
cnames.delete_if { _2 == value }
cnames.empty?
end
end
end
# Order of yielded crefs is undefined.
#
#: () { (Zeitwerk::Cref) -> void } -> void
def each_key
@mutex.synchronize do
@map.each do |mod, cnames|
cnames.each_key do |cname|
yield Zeitwerk::Cref.new(mod, cname)
end
end
end
end
#: () -> void
def clear
@mutex.synchronize do
@map.clear
end
end
#: () -> bool
def empty? # for tests
@mutex.synchronize do
@map.empty?
end
end
end
|
