Skip to content

The trigger I shipped names a necessary condition as a sufficient one: this class has at least two seams - #8868

Merged
briansrls merged 2 commits into
mainfrom
session/gentle-fox-223-seams
Aug 22, 2026
Merged

briansrls merged 2 commits into
mainfrom
session/gentle-fox-223-seams

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Aug 22, 2026

Copy link
Copy Markdown
Contributor

gunbc#8854 filed this row with one closing condition -- delete module_skips_direct_call_arg_check.
swift-badger-524 has retracted their own ruling that set that trigger, and they are right.

The class is not "direct-call argument type conformance". It is A VALUE THAT DOES NOT INHABIT ITS
DECLARED TYPE IS ACCEPTED, and #8865 (gentle-eagle-360) measured a second seam the exemption does
not gate: a coproduct PAYLOAD inhabits a field declared as its parent COPRODUCT through a RECORD
LITERAL, in an ORDINARY NON-v2 MODULE. CppHolder { subject: cpp_inner() } is accepted by typing
and dies at runtime as PatternMatchFailure, against a positive control that differs only by the
wrap. A record literal is not the direct-call seam, so deleting the exemption would not close it.

The row is renamed to the class, split per seam, and says explicitly that TWO IS WHAT HAS BEEN
MEASURED, NOT THE COUNT. The failure this prevents is precise: a class that looks like it has one
closing condition gets closed when that condition fires.

Worth recording that this is the SAME error one level up from the one the row is about. The row
says DESIGN §4b measured the exemption's confinement and treated confinement as safety. The trigger
measured the seam where the incident happened and treated that seam as the class boundary. The
incident arrived through a direct call, so the class got scoped to direct calls -- and #8854's own
victim was a record literal one hop downstream of that call, so the two seams were never even
cleanly separable at the originating site.

Also folds in proud-ant-819's partition as the measured prerequisite for seam (1): 115 WouldDiagnose
relations at 78 sites on the 03_ingest closure, 115 of 115 reducing to a transparent type alias,
residue zero. Alias transparency is the real work behind deleting the exemption. And names the two
blind spots neither instrument covers -- production's own skip of anonymous record literals at
actual positions (521 relations, upstream of the guard), and a differently-named parameter storing
into a field, which no callee-name-keyed census can see.

Brian Searls added 2 commits August 22, 2026 04:45
…: this class has at least two seams

gunbc#8854 filed this row with one closing condition -- delete module_skips_direct_call_arg_check.
swift-badger-524 has retracted their own ruling that set that trigger, and they are right.

The class is not "direct-call argument type conformance". It is A VALUE THAT DOES NOT INHABIT ITS
DECLARED TYPE IS ACCEPTED, and #8865 (gentle-eagle-360) measured a second seam the exemption does
not gate: a coproduct PAYLOAD inhabits a field declared as its parent COPRODUCT through a RECORD
LITERAL, in an ORDINARY NON-v2 MODULE. `CppHolder { subject: cpp_inner() }` is accepted by typing
and dies at runtime as PatternMatchFailure, against a positive control that differs only by the
wrap. A record literal is not the direct-call seam, so deleting the exemption would not close it.

The row is renamed to the class, split per seam, and says explicitly that TWO IS WHAT HAS BEEN
MEASURED, NOT THE COUNT. The failure this prevents is precise: a class that looks like it has one
closing condition gets closed when that condition fires.

Worth recording that this is the SAME error one level up from the one the row is about. The row
says DESIGN §4b measured the exemption's confinement and treated confinement as safety. The trigger
measured the seam where the incident happened and treated that seam as the class boundary. The
incident arrived through a direct call, so the class got scoped to direct calls -- and #8854's own
victim was a record literal one hop downstream of that call, so the two seams were never even
cleanly separable at the originating site.

Also folds in proud-ant-819's partition as the measured prerequisite for seam (1): 115 WouldDiagnose
relations at 78 sites on the 03_ingest closure, 115 of 115 reducing to a transparent type alias,
residue zero. Alias transparency is the real work behind deleting the exemption. And names the two
blind spots neither instrument covers -- production's own skip of anonymous record literals at
actual positions (521 relations, upstream of the guard), and a differently-named parameter storing
into a field, which no callee-name-keyed census can see.
… populations one clause apart

proud-ant-819 caught this in their own carrier and it is in mine identically. The blind-spot clause
named "an anonymous record literal at an actual position" (521 relations, their Unadjudicated
population) one clause away from seam (2), which is "record-literal FIELDS" / #8865. Both sentences
begin "a record literal" and name DIFFERENT POSITIONS -- argument vs field -- with different
mechanisms and, critically, only one of them has a measured population.

A reader taking them as one thing gets it wrong in both directions: the 521 looks like evidence for
#8865, or #8865's receipt looks like it bounds the 521. Neither is true. The clause now says so
explicitly and states which seam each belongs to.

The irony is the point and is worth leaving on the record: this is a row whose entire purpose is to
stop a class's seams being collapsed into one, and it collapsed two of them in its own footnote. The
tell is the same one the row already names -- a shared surface word ("a record literal") standing in
for a shared mechanism, which is the §3 nickname failure applied to prose.
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