Implement ARGF.class basics

[edgibbs]

TITLE: Implement ARGF.class basics

---

👋

Taking a run at #5. Same approach as last time, I used Claude Code with Opus 5 as a learning partner to find my way around the codebase and its patterns, but every line here I typed myself in Neovim, along with all the mistakes that entailed. This one took a lot longer than the last two, but figured it might be a challenge as an intermediate ticket. Apologies for the size, but it was a lot adding one thing and then uncovering a new small issue.

Both failures from the issue are gone:

# before
./exe/mspec-run ruby/spec/core/argf/argv_spec.rb ruby/spec/core/argf/each_byte_spec.rb
2 files, 7 examples, 0 expectations, 0 failures, 7 errors

# after
2 files, 7 examples, 9 expectations, 0 failures, 0 errors

The String#each_byte failure listed in the issue was already fixed by #56, so this only covers the ARGF half.

🔧 Implementation Notes

A few things worth calling out:

ARGF.argv has to be ARGV itself, not a copy. argv_spec.rb:10 uses should equal, and checking against MRI 3.3.6 the identity holds for the whole read cycle, even after the array has been emptied. Since it has to stay the same object, Argf holds the ARGV RValue and shifts names off it in place. That's what changed Argf.new's signature and the small exe/ruby-exe.ts diff.

Four of the commits are prerequisites the issue doesn't mention, which is why this touches six files:

• IO#closed? and STDIN — mspec's argf helper needs both in its ensure, so no ARGF spec could get past it
• IO#each_byte — what ARGF delegates to
• Enumerator#size — required by shared/each_byte.rb:49-55
• StdinIO implements the write side only; nothing here reads from STDIN and a real readable one looked like a much bigger job.

🧪 Testing

Beyond the repro above, across the whole ruby/spec/core/argf/ directory expectations passing went from 2 to 15 and errors dropped from 154 to 145. Failures went 1 to 2, but both of those are specs that used to error out in the helper and now run far enough to fail on other missing methods.

📤 Upstream

While checking Enumerator#size against MRI I noticed each_byte_spec.rb only tests non-empty strings, so an implementation returning nil for "".each_byte.size passes — which is exactly the bug I had, since 0 is falsy in JavaScript. MRI can't fail that case. each_char, each_codepoint, and each_line have no size specs at all. Planning a small ruby/spec PR for it.