JSON.dump returns NULL bytes for large HashWithIndifferentAccess objects (Linux x86_64)

[dalibor]

JSON.dump() intermittently returns all NULL bytes when serializing large ActiveSupport::HashWithIndifferentAccess objects on Linux x86_64. The same code works correctly on macOS ARM with identical Ruby/gem versions.

Details

• Platform: x86_64-linux
• Ruby: 4.0.1
• JSON: 2.19.0
• ActiveSupport: 8.1.2
• CPU: .25 vCPU
• Memory: 1 GiB

I wasn't able to reproduce it on arm64-darwin25.

Reproduction (x86_64-linux only)

Try multiple serialization attempts with larger payloads as it's an intermittent bug, though I'm able to reproduce it consistently.

require "bundler/inline"

gemfile do
  gem "activesupport", "8.1.2"
  gem "json", "2.19.0"
end

require 'active_support/core_ext/hash'

large_string = "x" * (35 * 1024 * 1024); nil
data = { "content" => large_string }.with_indifferent_access; nil

100.times do |i|
  json_data = JSON.dump(data)

  if json_data.getbyte(0) == 0
    puts "BUG: JSON.dump returned NULL bytes on attempt #{i}"
    puts "Size: #{json_data.bytesize} bytes"
    break
  end
end

The workaround is to convert to plain Hash before serialization using .to_h.

[byroot]

Thank you for the repro, I actually get a SEGV on Linux ARM64. Git bisect points to fff25c9

[dalibor]

Thank you for looking into it! I haven't tested on Linux ARM64, but on Linux x86_64 it shows up consistently in more resource-constrained environments even with smaller payloads. It took us some time until we isolated it since in production it was showing up intermittently. Thanks!

[byroot]

The SEGV was likely a red herring, I cleaned up the compilation cache and it no longer repro.

I might need to find an x86 machine because I can't effectively debug with docker. I might try virtual box or something like that.

[byroot]

The SEGV was likely a red herring,

I need to stop posting before finishing my coffee, not a red herring, I know what's going on and have a fix. Can't say for sure it's the same bug though.

[byroot]

Would you mind trying this PR: #952 ?

[dalibor]

I just tested with 2.18.0 and good news is that I cannot reproduce it on Linux x86_64. So, most likely related to the commit you pointed out!

[dalibor]

Sure, will test now with PR: #952

[dalibor]

I can confirm PR: #952 fixes it. I cannot reproduce it anymore on Linux x86_64. Thank you!

[byroot]

Alright, 2.19.1 is out with the fix. THanks again for the repro script and for trying the fix, saved me a lot of time.

[dalibor]

Thank you for your time and contributions!