rpg_eval.runValues: lane Rv finds the failures after the s07 choice token lie outside the choice unit; nothing changes #812

Manually merged
binjovi-bot merged 1 commit from lane-rv-choice-20261117 into trunk 2026-11-17 07:04:18 +00:00
Owner

rpg_eval.runValues: lane Rv finds the failures after the s07 choice token lie outside the choice unit; nothing changes

Negative result. No source, test or emit byte changes in this pull request.

Question: when runValues leaves the s07 choice token and run then
stops, which operation does it stop at, and does that operation belong to a
choice shape (%MAX, %MIN, AND/OR chains) that rpg_choice.zig declines?

Measured on a Debug build of the branch base (SOURCERER_TRACE_REFUSALS=1,
TRACE_OPERAND=1, trace-lib.sh) over every public archive.savf under
src/testdata/savf (LEMONADE included) and the object.savf files up to
SC-ILE/EXPERIMENTS/fmtind-j1; no private, held-out or exam material.
Observable and stripped copies of a probe are counted separately.

Case 1, s07 token and no failure after it. 8560 statements carry the token;
8153 are recovered, so the token only says the statement needs no choice
rewrite. 251 more are not recovered but run does not fail: 218 partial
type_inference/decimal_positions, 14 control_flow block_owner or
closing_destination, 4 call+control_flow remaining_effects, 2
value_flow/remaining_effects and a few display-argument rows. Their cause
is in other stages.

Case 2, s07 token and rpg_eval.run s01 (156 statements). The failure
dependency, by count of failure entries: operation 15 (85), 75 (30), 99 (26,
stages call and runtime_descriptor), 97 (9), procedure calls 254, 255,
272, 275 (14), 33 (4), 141 (4), 27 (2), 67 (1), and a few label ids. Only
operation 15 (a label) can be the tail of a choice shape, so the shapes
behind it were read:

  • %SUBST range guard (decimal-fault-report, decimal-sign-report,
    packed-add-report, packed-compose-report, packed-compose-sign-report,
    OPS/subst): 78 8; 12; 16 L1; 27 a; 27 b; 10; 17 L2; 15 L1 2; 27 c; 35 54; 99 3 0 0 0; 15 L2 2. choice.boolean reaches s08 (pair
    declines the two constants) because the chain does not produce a true/false
    value: both jumps end at an error call. It is a guard, not a choice value.
  • TLICR statement 52 (l_padvar, f = v from VARCHAR to CHAR): the
    compiler's length clamp (27 660; 27 441; 11; 16 9; 27 660; 35 444; 18 10; 15 9 1; 27 441; 35 444; 15 10 1, labels of kind 1). It belongs to the
    varying-copy rules, not to %MIN.
  • FLOW/ccsid, FLOW/dispatch-held, FLOW/return-switch-utf8: the
    boolean s08 token comes from a status guard (9; 16 L; ...; 15 L 1); the
    statement stops at operation 99 or 141, not at the guard.
  • TSELR statement 50 and TRPG3 statement 75: the label is the head of a
    backward scan loop (18 8 back to 15 8 2), a %TRIM-style loop, not a
    choice. The other labels did not reach boolean past s04.

Result: no public statement fails at an operation of a choice shape. The
failures after s07 stop in operations outside the choice unit: stage
control_flow operation 15 (range guard, loop head, varying clamp), stage
type_inference operation 75, stages call and runtime_descriptor
operations 99 and 97, stage value_flow procedures 254, 255, 272, 275 and
operation 33, stage classification operations 141 and 67.

Teaching the choice unit the range guard or the clamp would recover text no
public twin proves (the guard is an error branch, the clamp is a copy rule),
so the unit is left as it is. A follow-up lane would need a public probe
whose source holds an AND/OR condition that choice.boolean declines at
s05 to s13 and that ends in a value; none exists. Counter: 0 recovered, 0
worse.

Gates

{"checks":{"net_statements":0},"census":{"statuses":{"recovered":645,"partial":0,"unresolved":0},"regressions":0,"newly_recovered":0},"golden":"golden floor 173 ceiling 0 <= dump 173/5/0","legend":"current","emit":{"worse":[],"better":0,"newly_clean":0,"clean_delta":0,"base":738,"new":738},"twins":{"groups":1022,"recovered":1015,"regressions":[],"newly":0},"allgroups":{"regressions":[],"newly":[],"worse":[]},"unit":{"exit":0,"last":"ship-unit: no test file changed"},"full_check":{"exit":0,"seconds":1861}}
# rpg_eval.runValues: lane Rv finds the failures after the s07 choice token lie outside the choice unit; nothing changes Negative result. No source, test or emit byte changes in this pull request. Question: when `runValues` leaves the s07 `choice` token and `run` then stops, which operation does it stop at, and does that operation belong to a choice shape (%MAX, %MIN, AND/OR chains) that `rpg_choice.zig` declines? Measured on a Debug build of the branch base (`SOURCERER_TRACE_REFUSALS=1`, `TRACE_OPERAND=1`, `trace-lib.sh`) over every public `archive.savf` under `src/testdata/savf` (LEMONADE included) and the `object.savf` files up to `SC-ILE/EXPERIMENTS/fmtind-j1`; no private, held-out or exam material. Observable and stripped copies of a probe are counted separately. Case 1, s07 token and no failure after it. 8560 statements carry the token; 8153 are recovered, so the token only says the statement needs no choice rewrite. 251 more are not recovered but `run` does not fail: 218 partial `type_inference/decimal_positions`, 14 `control_flow` block_owner or closing_destination, 4 `call`+`control_flow` remaining_effects, 2 `value_flow/remaining_effects` and a few display-argument rows. Their cause is in other stages. Case 2, s07 token and `rpg_eval.run` s01 (156 statements). The failure dependency, by count of failure entries: operation 15 (85), 75 (30), 99 (26, stages `call` and `runtime_descriptor`), 97 (9), procedure calls 254, 255, 272, 275 (14), 33 (4), 141 (4), 27 (2), 67 (1), and a few label ids. Only operation 15 (a label) can be the tail of a choice shape, so the shapes behind it were read: - `%SUBST` range guard (`decimal-fault-report`, `decimal-sign-report`, `packed-add-report`, `packed-compose-report`, `packed-compose-sign-report`, `OPS/subst`): `78 8; 12; 16 L1; 27 a; 27 b; 10; 17 L2; 15 L1 2; 27 c; 35 54; 99 3 0 0 0; 15 L2 2`. `choice.boolean` reaches s08 (`pair` declines the two constants) because the chain does not produce a true/false value: both jumps end at an error call. It is a guard, not a choice value. - `TLICR` statement 52 (`l_padvar`, `f = v` from VARCHAR to CHAR): the compiler's length clamp (`27 660; 27 441; 11; 16 9; 27 660; 35 444; 18 10; 15 9 1; 27 441; 35 444; 15 10 1`, labels of kind 1). It belongs to the varying-copy rules, not to %MIN. - `FLOW/ccsid`, `FLOW/dispatch-held`, `FLOW/return-switch-utf8`: the boolean s08 token comes from a status guard (`9; 16 L; ...; 15 L 1`); the statement stops at operation 99 or 141, not at the guard. - `TSELR` statement 50 and `TRPG3` statement 75: the label is the head of a backward scan loop (`18 8` back to `15 8 2`), a %TRIM-style loop, not a choice. The other labels did not reach `boolean` past s04. Result: no public statement fails at an operation of a choice shape. The failures after s07 stop in operations outside the choice unit: stage `control_flow` operation 15 (range guard, loop head, varying clamp), stage `type_inference` operation 75, stages `call` and `runtime_descriptor` operations 99 and 97, stage `value_flow` procedures 254, 255, 272, 275 and operation 33, stage `classification` operations 141 and 67. Teaching the choice unit the range guard or the clamp would recover text no public twin proves (the guard is an error branch, the clamp is a copy rule), so the unit is left as it is. A follow-up lane would need a public probe whose source holds an AND/OR condition that `choice.boolean` declines at s05 to s13 and that ends in a value; none exists. Counter: 0 recovered, 0 worse. ## Gates ``` {"checks":{"net_statements":0},"census":{"statuses":{"recovered":645,"partial":0,"unresolved":0},"regressions":0,"newly_recovered":0},"golden":"golden floor 173 ceiling 0 <= dump 173/5/0","legend":"current","emit":{"worse":[],"better":0,"newly_clean":0,"clean_delta":0,"base":738,"new":738},"twins":{"groups":1022,"recovered":1015,"regressions":[],"newly":0},"allgroups":{"regressions":[],"newly":[],"worse":[]},"unit":{"exit":0,"last":"ship-unit: no test file changed"},"full_check":{"exit":0,"seconds":1861}} ```
docs/changes: lane Rv finds no choice shape behind the failures after the s07 token
All checks were successful
binjovi/ci Binjovi completed the frozen plan
0673b6dc82
binjovi-bot manually merged commit 0673b6dc82 into trunk 2026-11-17 07:04:18 +00:00
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
sean/sourcerer!812
No description provided.