Skip to content

landing: the largest debt is no longer a hang, and the gap published here is a floor - #42

Merged
tannevaled merged 1 commit into
mainfrom
docs-largest-debt-is-not-a-hang
Nov 15, 2026
Merged

tannevaled merged 1 commit into
mainfrom
docs-largest-debt-is-not-a-hang

Conversation

@tannevaled

Copy link
Copy Markdown
Contributor

landing: the largest debt is no longer a hang, and the gap published here is a floor

This page stated, as its headline reading of ruby/spec's unswept trees, that
"the largest single debt is a hang rather than a missing feature" —
Net::FTP 442 examples behind, 39 of its 56 files killed by the clock. Twelve
merged pull requests in go-embedded-ruby/ruby have made that sentence false in
both halves, and it was the sentence a reader would quote.

The hang is gone

then now
Net::FTP files killed 39 of 56 0
Net::FTP examples behind 442 128
Net::HTTP files killed 27 0
killed across the corpus 75 8

Re-measured 2026-11-13 against MRI 4.0.6, same corpus, 30 s per file.

One symptom, three unrelated causes — which is why the first fix looked like
it had done nothing when it was measured against the other half's specs: a
blocking socket call holding rbgo's emulated GVL, a TCPSocket.open { } that
never closed the socket so the peer waited for an EOF that never came, and a
Thread#kill that could not interrupt a blocking native call.

And it was never the largest debt

library/socket's own spec_helper.rb calls MSpec.enable_feature, which the
conformance shim did not define — so all 190 of those files scored 0 under MRI
as well
. A control of zero is indistinguishable from a module with no gap, and
the same sweep that now reads −1566 read "+217 in rbgo's favour" twenty minutes
earlier.

Repaired, the module measures MRI 1790 / rbgo 224 — about −1560, more
than matrix, bigdecimal, stringscanner, date, zlib, stringio and
openssl put together — and it is Addrinfo, recvmsg and the socket option
accessors rather than anything that hangs.

⛔ The published 2,341 gap is therefore marked a floor, not a total. The
two figures are deliberately not added together: they come from different runs
against different MRI patch releases (4.0.7 and 4.0.6), and a sum across two
measurements is how two errors cancel. A weaker claim that is true beats a
precise one that is not.

The drift gate, and its limit

conformancefigure@v0.8.0 --root . passes on this branch: 6 total claims, 1
version claim and 3 stubbed-helper mentions checked against rbgo's gate (23,672
at Ruby 4.0.9).

It has nothing to say about any of the figures this change touches, and that is
not a fault in it — there is nothing to check them against, because the dated run
that produced them is gone. The gate guards the numbers CI can re-derive; a dated
measurement on a page can only be guarded by someone asking which sentences the
merges made false.

Refs go-embedded-ruby/ruby#916, go-embedded-ruby/ruby#937, go-embedded-ruby/ruby#938

(All three stay open or are handled in their own repository; this pull request
only changes this page. The matching corrections to the repository README and to
the docs site are separate pull requests — one surface each.)

🤖 Generated with Claude Code

…here is a floor

This page stated, as its headline reading of the unswept trees, that **"the
largest single debt is a hang rather than a missing feature"** — `Net::FTP` 442
examples behind with 39 of its 56 files killed by the clock. Twelve merged pull
requests in go-embedded-ruby/ruby have made that sentence false in both halves.

**The hang is gone.** `Net::FTP` is 0 files killed and 128 behind (re-measured
2026-11-13 against MRI 4.0.6, same corpus), `Net::HTTP` is 0 from 27, and 75
killed files across the corpus are down to 8. One symptom, three unrelated
causes, which is why the first fix looked like it had done nothing: a blocking
socket call holding the emulated GVL, a `TCPSocket.open { }` that never closed
the socket, and a `Thread#kill` that could not interrupt a blocking native call.

**And it was never the largest debt.** `library/socket`'s own `spec_helper.rb`
calls `MSpec.enable_feature`, which the conformance shim did not define, so all
190 of those files scored 0 **under MRI as well** — and a control of zero is
indistinguishable from a module with no gap. Repaired, the module measures MRI
1790 / rbgo 224: about -1560, more than matrix, bigdecimal, stringscanner, date,
zlib, stringio and openssl put together, and it is unimplemented surface rather
than anything that hangs.

The published 2,341 gap is therefore marked a **floor**, not a total. The two
figures are deliberately not added together: different runs, different MRI patch
releases, and a sum across two measurements is how two errors cancel.

The page's own drift gate stays green — `conformancefigure@v0.8.0 --root .`
checks 6 total claims, 1 version claim and 3 stubbed-helper mentions against
rbgo's gate (23,672 at Ruby 4.0.9). It has nothing to say about these figures,
because the run that produced them is gone; that is the limit of a drift gate on
a dated measurement, not a fault in it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@tannevaled
tannevaled merged commit d294ba7 into main Nov 15, 2026
3 checks passed
@tannevaled
tannevaled deleted the docs-largest-debt-is-not-a-hang branch November 15, 2026 09:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant