Repository navigation
Nondeterministic failures on version 3.2.1 #337
Description
Activity
I'm also seeing the same, but with array:
Array can't be coerced into BigDecimal. I'm not aware of any place where I'd coerce an array into BigDecimal.Sorry I can't reproduce it.
Does it happen frequently, rarely or always? Could you provide a reproduction code andruby -vresult?I can reproduce this with 100% reliability on
ruby=3.4.4and bigdecimal 3.2.0 and 3.2.1. The only thing that changes is the "what can't be coerced". This is highly sus and likely points to memory corruption related to GC compaction. I've also seen this result in segfaults, presumably if the memory it points to is not a validVALUE.#!/usr/bin/env ruby require "bundler/inline" puts "Running with Ruby version: #{RUBY_VERSION}" gemfile do source "https://rubygems.org" gem "bigdecimal", "3.2.1" end require "bigdecimal" ZERO = BigDecimal(0).freeze GC.verify_compaction_references(expand_heap: true, toward: :empty) ZERO / BigDecimal(100)
❯ ./bigdecimal.rb Running with Ruby version: 3.4.4 bigdecimal.rb:19:in 'BigDecimal#/': #<Class:0x00000001210dde38> can't be coerced into BigDecimal (TypeError) ZERO / BigDecimal(100) ^^^^^^^^^^^^^^^Thank you. I can reproduce it.
X = BigDecimal('2.5') GC.verify_compaction_references(expand_heap: true, toward: :empty) # Bug existed before v3.1.9 X**X # ??? can't be coerced into BigDecimal BigMath.log(X, 10) # ??? can't be coerced into BigDecimal BigMath.exp(X, 10) # Segmentation fault # Bug from v3.2.0 X / 2 # ??? can't be coerced into BigDecimal
Reacted by Mike DiScala, Jacob Evelyn and Robin BrandtThank you for the quick attention here & for the help reproducing!
Hi there,
When we bumped our application from version 3.1.9 to version 3.2.1 we started getting nondeterministic failures in our test suite.
The problematic code does:
and then uses this when calculating an average of an array of number:
The error returned has been:
but we've also seen it error out with a string type:
I initially assumed that we were somehow passing a string into
x_values, but since the issue is non-deterministic, I'm not sure how that could be the case. I tried to reproduce by running a similar sum operation in a loop but wasn't able to reproduce there.Do you have any ideas on what might be causing this after the upgrade? Is there any other debugging information I can provide?
Thank you!
--Mike