Two parity divergences: Array.new(2**128) raises the wrong class, and expand_path("~") raises where MRI does not #901

Both found while measuring message quoting for #900, and neither is about quoting — so they are filed rather than folded into that PR.

Measured with rubydiff against /opt/homebrew/opt/ruby@4.0/bin/ruby (ruby 4.0.7 (2026-09-15 revision 229531a6cf) +PRISM [arm64-darwin25]) on main at 6e86fac3.

1. Array.new(2**128) raises the wrong class

Array.new(2**128)

A different class, not just a different message, so rescue RangeError does not catch it. The neighbouring conversions agree — [1][2**128] and "x" * (2**128) both give rbgo's RangeError: bignum too big to convert into 'long' — so Array.new is checking its own size bound before the integer conversion that MRI performs first.

MRI reaches the conversion first because rb_ary_new2 takes a long: NUM2LONG runs on the argument before any size test, and ary_new raises ArgumentError, "array size too big" only for a value that converted successfully and is still out of range. Both errors exist upstream; rbgo reaches the second where MRI reaches the first.

2. File.expand_path("~") raises where MRI does not

ENV.delete("HOME")
File.expand_path("~")

MRI does have that message — file.c:3908 at v4.0.9, which #900 quotes — but it is not reached here, so there is a fallback rbgo does not have. file.c:3889 is the neighbouring couldn't find login name -- expanding '~', which suggests the fallback is a passwd-database lookup for the current user rather than the environment alone.

I have not confirmed which fallback MRI took, only that it did not raise. That is the first thing to establish before changing anything: a fix built on the guess above would be a fix for a cause nobody read.

Note on scope

Neither affects the #799 version move. Both are plain parity defects on main.

🤖 Generated with Claude Code