Bug #16950: Stop nonsense keyword argument warnings in 2.6

• Status: Closed
• Assignee: usa (Usaku NAKAMURA)
• Target version: -
• ruby -v: ruby 2.6.7p147 (2020-03-31 revision 67883) [x86_64-linux]
• Backport: 2.5: DONTNEED, 2.6: DONE, 2.7: DONTNEED

[mame (Yusuke Endoh)]

Ruby 2.6 warns the following code with -w option:

def foo(x)    # warning: in `foo': the last argument was passed as a single Hash
end
h = { k: 42 }
foo(**h)      # warning: although a splat keyword arguments here

This warning had been introduced by @nobu (Nobuyoshi Nakada) based on the original proposal of #14183 "Real" keyword argument. However, the design was changed so that automatic conversion from keywords to positional arguments will be kept in 3.0. In other words, the code above will work even in 3.0, so the warning no longer makes sense.

Rather, according to @matsuda (Akira Matsuda) and @kamipo (Ryuta Kamizono), this warning sometimes hindered the fix work for the keyword argument change in Rails. This issue is possible to work around by using always non-keyword argument (and they actually did so), but they say it is very annoying.

So, I propose stopping the warning in 2.6. This is a change for 2.6, but matz already accepted this. @usa (Usaku NAKAMURA), do you accept this?

A patch is quite simple:

diff --git a/vm_args.c b/vm_args.c
index 299ed16aa8..2e70ed3fcf 100644
--- a/vm_args.c
+++ b/vm_args.c
@@ -685,17 +685,6 @@ setup_parameters_complex(rb_execution_context_t * const ec, const rb_iseq_t * co
     else if (!NIL_P(keyword_hash) && RHASH_SIZE(keyword_hash) > 0) {
 	argument_kw_error(ec, iseq, "unknown", rb_hash_keys(keyword_hash));
     }
-    else if (kw_splat && NIL_P(keyword_hash)) {
-	if (RTEST(ruby_verbose)) {
-	    VALUE path = rb_iseq_path(iseq);
-	    VALUE line = rb_iseq_first_lineno(iseq);
-	    VALUE label = rb_iseq_label(iseq);
-	    rb_compile_warning(NIL_P(path) ? NULL : RSTRING_PTR(path), FIX2INT(line),
-			       "in `%s': the last argument was passed as a single Hash",
-			       NIL_P(label) ? NULL : RSTRING_PTR(label));
-	    rb_warning("although a splat keyword arguments here");
-	}
-    }
 
     if (iseq->body->param.flags.has_block) {
 	if (iseq->body->local_iseq == iseq) {

[kamipo (Ryuta Kamizono)]

Using non-keyword argument is not always workaround for that.

For example, redis 4.2.0 gem has changed options hash to kwargs.

To address kwargs warnings, we need to add ** to the method call, it works for both redis 4.2.0 and 4.1.4 with using Ruby 2.7.

But if we use Ruby 2.6, adding ** works for redis 4.2.0, but raise the warning for redis 4.1.4.

https://github.com/redis/redis-rb/blob/v4.2.0/lib/redis.rb#L816
https://github.com/redis/redis-rb/blob/v4.1.4/lib/redis.rb#L790

Strictly speaking, there is a way to avoid the warning by checking RUBY_VERSION (and gem version).

But I've finally given up to address the warning in the Rails codebase.

https://github.com/rails/rails/commit/37c19f7ebcf542af968a4983b3296b9fa283a0dc#diff-60aba42516f795d80ed6f18ad14c1f6aR401

[jeremyevans0 (Jeremy Evans)]

• Is duplicate of Bug #16632: Remove verbose warning on treating keyword splat as positional argument in Ruby 2.6 added

[usa (Usaku NAKAMURA)]

I see, I'll patch later.

[jeremyevans0 (Jeremy Evans)]

@usa (Usaku NAKAMURA) It looks like this hasn't yet been fixed in Ruby 2.6. Is it possible to fix this in Ruby 2.6 before Ruby 2.6 goes into security maintenance mode?

[usa (Usaku NAKAMURA)]

• Status changed from Open to Closed
• Backport changed from 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN to 2.5: DONTNEED, 2.6: DONE, 2.7: DONTNEED

patched to ruby_2_6 at r67899.