per-ractor Random::DEFAULT #3813
Conversation
| @@ -829,7 +839,7 @@ static VALUE | |||
| rb_f_srand(int argc, VALUE *argv, VALUE obj) | |||
| { | |||
| VALUE seed, old; | |||
| rb_random_mt_t *r = &default_rand; | |||
| rb_random_mt_t *r = rand_mt_start(default_rand()); | |||
There was a problem hiding this comment.
If we call rand_mt_start here, will rand_init be called twice?
There was a problem hiding this comment.
Without it, the previous seed will be false (Qfalse) and srand() returns false if there is no usages of default_rand like rand() and so on.
If returning false by srand(), we can delete it.
There was a problem hiding this comment.
Since rand_init() should initialize r->mt, genrand_initialized should be true and it will not be called again.
There was a problem hiding this comment.
default_rand allocates the memory of the default random object at the first srand call in each ractor.
Then random_mt_start calls rand_init for the default random object.
The second rand_init is called below.
There was a problem hiding this comment.
I see.
Seems OK to return 0 when old == Qfalse, instead of initialization here.
| static VALUE | ||
| random_s_seed(VALUE obj) | ||
| { | ||
| rb_random_mt_t *rnd = rand_mt_start(default_rand()); |
There was a problem hiding this comment.
def test_default_seed
assert_separately([], "#{<<~"begin;"}\n#{<<~'end;'}")
begin;
seed = Random::DEFAULT::seed
rand1 = Random::DEFAULT::rand
rand2 = Random.new(seed).rand
assert_equal(rand1, rand2)
srand seed
rand3 = rand
assert_equal(rand1, rand3)
end;
end
This test doesn't work...
|
|
||
| ruby_version_is ''...'3.0' do | ||
| it "returns a Random instance" do | ||
| Random::DEFAULT.should be_an_instance_of(Random) | ||
| end | ||
| end | ||
|
|
||
| ruby_version_is '3.0' do | ||
| it "returns a Random instance" do | ||
| Random::DEFAULT.should be_an_instance_of(Class) | ||
| end |
There was a problem hiding this comment.
| ruby_version_is ''...'3.0' do | |
| it "returns a Random instance" do | |
| Random::DEFAULT.should be_an_instance_of(Random) | |
| end | |
| end | |
| ruby_version_is '3.0' do | |
| it "returns a Random instance" do | |
| Random::DEFAULT.should be_an_instance_of(Class) | |
| end | |
| it "returns a random number generator" do | |
| Random::DEFAULT.should respond_to(:rand) | |
| end |
There was a problem hiding this comment.
It seems nice, but OTOH the current diff has the advantage that other Ruby implementations will notice the change and that they need to change what Random::DEFAULT points it when updating to Ruby 3.
Maybe have both?
There was a problem hiding this comment.
It can be anything which generates (pseudo) random numbers.
There was a problem hiding this comment.
It doesn't feel reasonable that "the PRNG should be a class", but rather nonsense.
"It should respond to rand" would be fine as duck-typing.
There was a problem hiding this comment.
It is an implementation detail which is useless to be shared.
In fact, the constant might be a proxy object, and could be changed.
What we consider is just that "it works as a RNG".
There was a problem hiding this comment.
Till now, no code uses that "Random::DEFAULT equals Random", of course.
If we publish this statement, users may write such code and we (including other implementors) will have to accept it.
If we don't, we can reject such code.
This is chicken-and-egg.
There was a problem hiding this comment.
Because Random::DEFAULT is Random will be in 3.0, users might depend on it. It would be nicer if they don't , but they might and there is nothing we can do about it, except if we deprecate the Random::DEFAULT constant.
The fact there is a test in ruby/spec doesn't mean some code is supported, recommended or anything. It just tries to test the behavior, in as much details as possible to improve compatibility between Ruby implementations.
Probably lots of tests in ruby/spec are not what user code should do, but they are useful tests nevertheless.
BTW, I think we should mention this change in NEWS because it could cause some incompatibility.
There was a problem hiding this comment.
The fact there is a test in ruby/spec doesn't mean some code is supported, recommended or anything. It just tries to test the behavior, in as much details as possible to improve compatibility between Ruby implementations.
Probably lots of tests in ruby/spec are not what user code should do, but they are useful tests nevertheless.
Probably I've misunderstood your intention, ruby/spec is only for other implementors, but the other users should not rely on it?
There was a problem hiding this comment.
Looks like I forgot to reply to this.
I think ruby/spec is primarily for Ruby implementors yes, similar to the CRuby test suite.
I don't think many Ruby users look at ruby/spec, although of course they enjoy better-tested Ruby implementations.
Random generators are not Ractor-safe, so we need to prepare per-ractor default random genearators. This patch set `Random::DEFAULT = Randm` (not a Random instance, but the Random class) and singleton methods like `Random.rand()` use a per-ractor random generator. [Feature #17322]
Random generators are not Ractor-safe, so we need to prepare
per-ractor default random genearators. This patch set
Random::DEFAULT = Randm(not a Random instance, but the Randomclass) and singleton methods like
Random.rand()use a per-ractorrandom generator.