Commit 4f24255 introduced a bug which allows a length to be passed to
rb_ary_new4 which is too large, resulting in invalid memory access.
For example:
(1..1000).to_a.slice!(-2, 1000)
Force recycled objects could create a freelist for the page. At the
start of sweeping we should append to the freelist to avoid
permanently losing the slots on the freelist.
There were two issues:
1. When an IO object is waiting for writablility only (as in test_tcp_accept) the selected hash is empty.
Therefore selected[fiber] returns nil but needs to default to 0 in order to be or'ed with IO::WRITABLE.
2. When an IO object is waiting for read- or writability (as in test_tcp_connect), but only one of these
two events arrive, the Fiber and IO object need to be removed from the other `@readable` or `@writable` list.
We don't need to increment/decrement iteration level for frozen Hash
because frozen Hash can't be modified. We can assume that nobody
changes the target Hash while calling #each family.
How to reproduce:
a = {}
100.times do |i|
a[i] = true
end
Ractor.make_shareable(a)
4.times.collect do
Ractor.new(a) do |b|
100.times do
b.each_value do
end
end
end
end.each(&:take)
Example output:
internal:ractor>:267: warning: Ractor is experimental, and the behavior may change in future versions of Ruby! Also there are many implementation issues.
#<Thread:0x00007fcfb087bb30 run> terminated with exception (report_on_exception is true):
#<Thread:0x00007fcfb087b8d8 run> terminated with exception (report_on_exception is true):
#<Thread:0x00007fcfb088d678 run> terminated with exception (report_on_exception is true):
#<Thread:0x00007fcfb087bd88 run> terminated with exception (report_on_exception is true):
/tmp/h.rb:10:in `each_value'/tmp/h.rb:10:in `each_value': : /tmp/h.rb:10:in `each_value'no implicit conversion from nil to integer/tmp/h.rb:10:in `each_value'no implicit conversion from nil to integer (: : (TypeErrorTypeError)no implicit conversion from nil to integer)no implicit conversion from nil to integer (
(TypeErrorTypeError from /tmp/h.rb:10:in `block (3 levels) in <main>'
from /tmp/h.rb:10:in `block (3 levels) in <main>'
)) from /tmp/h.rb:9:in `times'
from /tmp/h.rb:9:in `times'
from /tmp/h.rb:9:in `block (2 levels) in <main>'
from /tmp/h.rb:10:in `block (3 levels) in <main>'
from /tmp/h.rb:10:in `block (3 levels) in <main>'
from /tmp/h.rb:9:in `block (2 levels) in <main>'
from /tmp/h.rb:9:in `times'
from /tmp/h.rb:9:in `times'
from /tmp/h.rb:9:in `block (2 levels) in <main>'
from /tmp/h.rb:9:in `block (2 levels) in <main>'
<internal:ractor>:694:in `take': thrown by remote Ractor. (Ractor::RemoteError)
from /tmp/h.rb:14:in `each'
from /tmp/h.rb:14:in `<main>'
/tmp/h.rb:10:in `each_value': no implicit conversion from nil to integer (TypeError)
from /tmp/h.rb:10:in `block (3 levels) in <main>'
from /tmp/h.rb:9:in `times'
from /tmp/h.rb:9:in `block (2 levels) in <main>'
The `lambda_body` grammar rule has a `"}"`, which is throwing off the `exyacc.rb` regular expressions. This changes the regular expression to account for `"}"` as well, which makes the output of `ruby sample/exyacc.rb < parse.y` change by the following diff:
```diff
632,634d631
< ", &@3);
< $$ = $2;
< }
```
Which makes it closer to a valid EBNF.
This commits implements size classes in the GC for the Variable Width
Allocation feature. Unless `USE_RVARGC` compile flag is set, only a
single size class is created, maintaining current behaviour. See the
redmine ticket for more details.
Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
This commit removes T_PAYLOAD since the new VWA implementation no longer
requires T_PAYLOAD types.
Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
./spec/ruby/optional/capi/ext/array_spec.c can match with spec/%/ if
using GNU Make under version 3.81. make command installed on macOS is
3.81, so ruby can't be built with default make on macOS with
--enable-shared option since bda56a03a6
Formerly, TypeProf was tested with the latest RBS code during
`make test-bundled-gems`. However, when a new version of rbs is
released, and if it is incompatible with TypeProf,
`make test-bundled-gems` starts failing, which was annoying.
By this change, TypeProf is tested with the bundled version of RBS.
rb_objspace_reachable_objects_from requires that the GC not be active.
Since the Ractor barrier is not executed for incremental sweeping,
Ractor may call rb_objspace_reachable_objects_from after sweeping
has started to share objects. This causes a crash that looks like
the following:
```
<internal:ractor>:627: [BUG] rb_objspace_reachable_objects_from() is not supported while during_gc == true
```
Co-authored-by: Vinicius Stock <vinicius.stock@shopify.com>
Explicitly specify the root as a user to run tests in the containers.
Coming new ruby/ruby-ci-image images are required to run the container as
a regular user by default, while the root user is required to run the
compilers.yml. Add `id` command to print the user info.
Co-authored-by: fedor <fedor@cirruslabs.org>
Formerly, TypeProf was tested with the latest RBS code during
`make test-bundled-gems`. However, when a new version of rbs is
released, and if it is incompatible with TypeProf,
`make test-bundled-gems` starts failing, which was annoying.
By this change, TypeProf is tested with the bundled version of RBS.
