The status unit of a call in a THREAD(*CONCURRENT) or THREAD(*SERIALIZE) module (regenerates docs/OPERATIONS.md) #779

Manually merged
binjovi-bot merged 3 commits from lext2 into trunk 2026-11-10 17:11:00 +00:00
Owner

The status unit of a call in a THREAD(*CONCURRENT) or THREAD(*SERIALIZE) module (regenerates docs/OPERATIONS.md)

Lane Lext, round two. Round one (v0.1.495) let the statement readers accept
thread-local static storage; the call statements of the L10 development
programs still stopped, every one of them, at the status unit after the call
(type_inference, operation 75 at the head of the statement). The battery
requests/lext/thread-call (24 captures) reproduces two families in public
twins and the status unit is one of them.

What the captures show

A call of a procedure of the module has the frame of rpg_frames.frame before
it and a status unit after it. A module without THREAD tests the status at the
frame: 29 F 96 Q; 32 B 0 I2; 27 1; 7; 16 L; 27 1; 35 R; 18 E; 15 L; 27 0; 35 R; 15 E (lext-call-void-p, lext-call-ret-p). In a thread module the unit is
27 0; 35 R and nothing else: R, the call status, is cleared and not tested
(lext-call-void-c, -const-c, -value-c, -ref-c, -nested-c, -ret-c). The same
unit ends the calls of a THREAD(*SERIALIZE) module (lext-serialize-c), whose
symbols do not have the thread-local bit, so the unit is the proof and not the
bit. A call with callp(e) or inside a monitor keeps its other forms
(lext-call-error-c, lext-call-monitor-c).

The rule

rpg_frames.threadStatus, tried first by rpg_frames.status: the two
operations 27 0 (a 4-byte integer constant 0) and 35 R, with R the call
status of status (the statement status, or a static 4-byte binary). It reads
nothing and has no label, so it has no effect to prove away besides R = 0. The
frame proof before it is unchanged. Eight captures are adopted as twins
(EXPERIMENTS/lext-call-void-c, -void-p, -const-c, -value-c, -ref-c,
-nested-c, -ret-c, lext-serialize-c); src/rpg_lext2_test.zig asserts
their calls (CALLP BUMP(), CALLP ADDTO(5), S254 = TWICE(TWICE(2)), ...),
that the control still reads its status test, and that a store that is not the
call status leaves the statement open. Before the rule the thread twins had 1 to
3 unresolved statements (every call); after it none, and lext-call-void-c
has the 6 recovered statements of lext-call-void-p.

Counter

L10 development data (six programs, Debug build, sourcerer dump): recovered
1,490 of 3,449 before and 1,490 of 3,449 after. No change: every call of those
programs ends with another unit, 29 P 10 7; 37 P 15 7; 27 0; 35 R; 27 T; 37 P 10 7 (P the 429-byte area; the programs declare a program status data
structure, dcl-ds psds psds, and the compiler rolls its status, bytes 11 to 15
into 16 to 20, and clears it after each call). No public twin has it. The
battery requests/lext/psds-call (7 cases, with a plain control) asks for it;
the next rule extends threadStatus with that roll once it is captured.

The other family

export procedures have symbol class 1, read by rpg_statements.recover as the
cycle main (lext-export-main-p, lext-export-cycle-p: the DCL-PROC of the
exported procedure is ? (no operation), its END-PROC and the call of it are
unresolved, without THREAD as well). Not taken this round.

Release gate (coordinator, before this pull request)

Checked on the binary of the shipped commit e6d9d35f; the head differs from it only in the title line of
docs/changes/lext2.md.

  • Development set: the five emits are byte-identical to v0.1.495.
  • Committed public twins at v0.1.495: every emit is byte-identical (emit-bytes-diff.sh).
  • L10 test programs: direction unchanged.
  • L10 development programs (six captures): every emit is byte-identical.
  • Lane captures (lext-thread-call 24, lext-psds-call 7): ten emits change, all in lext-thread-call. Each
    newly emitted statement was read against the probe source and agrees with it: ADDTO(5), GROW(k),
    ADDTO(k), r = TWICE(3), p = NOTHING(), BUMP() in a loop, twice in a row and under THREAD(*SERIALIZE).
    lext-call-monitor-c and lext-call-nested-c change only the text of a failure line.
  • No host round trip decides here: every changed emit still carries failure lines under both binaries (the thread
    keyword has no emit rule), so no case could match before and none can go from match to worse.

Gates (ship.sh, commit e6d9d35f):

{
  "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": "regenerated",
  "emit": {
    "worse": [],
    "better": 0,
    "newly_clean": 0,
    "clean_delta": 0,
    "base": 588,
    "new": 588
  },
  "twins": {
    "groups": 826,
    "recovered": 804,
    "regressions": [],
    "newly": 0
  },
  "allgroups": {
    "regressions": [],
    "newly": [],
    "worse": []
  },
  "full_check": {
    "exit": 0,
    "seconds": 1120
  }
}
# The status unit of a call in a THREAD(*CONCURRENT) or THREAD(*SERIALIZE) module (regenerates docs/OPERATIONS.md) Lane Lext, round two. Round one (v0.1.495) let the statement readers accept thread-local static storage; the call statements of the L10 development programs still stopped, every one of them, at the status unit after the call (`type_inference`, operation 75 at the head of the statement). The battery `requests/lext/thread-call` (24 captures) reproduces two families in public twins and the status unit is one of them. ## What the captures show A call of a procedure of the module has the frame of `rpg_frames.frame` before it and a status unit after it. A module without THREAD tests the status at the frame: `29 F 96 Q; 32 B 0 I2; 27 1; 7; 16 L; 27 1; 35 R; 18 E; 15 L; 27 0; 35 R; 15 E` (lext-call-void-p, lext-call-ret-p). In a thread module the unit is `27 0; 35 R` and nothing else: R, the call status, is cleared and not tested (lext-call-void-c, -const-c, -value-c, -ref-c, -nested-c, -ret-c). The same unit ends the calls of a THREAD(*SERIALIZE) module (lext-serialize-c), whose symbols do not have the thread-local bit, so the unit is the proof and not the bit. A call with `callp(e)` or inside a `monitor` keeps its other forms (lext-call-error-c, lext-call-monitor-c). ## The rule `rpg_frames.threadStatus`, tried first by `rpg_frames.status`: the two operations `27 0` (a 4-byte integer constant 0) and `35 R`, with R the call status of `status` (the statement status, or a static 4-byte binary). It reads nothing and has no label, so it has no effect to prove away besides R = 0. The frame proof before it is unchanged. Eight captures are adopted as twins (`EXPERIMENTS/lext-call-void-c`, `-void-p`, `-const-c`, `-value-c`, `-ref-c`, `-nested-c`, `-ret-c`, `lext-serialize-c`); `src/rpg_lext2_test.zig` asserts their calls (`CALLP BUMP()`, `CALLP ADDTO(5)`, `S254 = TWICE(TWICE(2))`, ...), that the control still reads its status test, and that a store that is not the call status leaves the statement open. Before the rule the thread twins had 1 to 3 unresolved statements (every call); after it none, and `lext-call-void-c` has the 6 recovered statements of `lext-call-void-p`. ## Counter L10 development data (six programs, Debug build, `sourcerer dump`): recovered 1,490 of 3,449 before and 1,490 of 3,449 after. No change: every call of those programs ends with another unit, `29 P 10 7; 37 P 15 7; 27 0; 35 R; 27 T; 37 P 10 7` (P the 429-byte area; the programs declare a program status data structure, `dcl-ds psds psds`, and the compiler rolls its status, bytes 11 to 15 into 16 to 20, and clears it after each call). No public twin has it. The battery `requests/lext/psds-call` (7 cases, with a plain control) asks for it; the next rule extends `threadStatus` with that roll once it is captured. ## The other family `export` procedures have symbol class 1, read by `rpg_statements.recover` as the cycle main (lext-export-main-p, lext-export-cycle-p: the `DCL-PROC` of the exported procedure is `? (no operation)`, its `END-PROC` and the call of it are unresolved, without THREAD as well). Not taken this round. ## Release gate (coordinator, before this pull request) Checked on the binary of the shipped commit e6d9d35f; the head differs from it only in the title line of `docs/changes/lext2.md`. - Development set: the five emits are byte-identical to v0.1.495. - Committed public twins at v0.1.495: every emit is byte-identical (`emit-bytes-diff.sh`). - L10 test programs: direction unchanged. - L10 development programs (six captures): every emit is byte-identical. - Lane captures (`lext-thread-call` 24, `lext-psds-call` 7): ten emits change, all in `lext-thread-call`. Each newly emitted statement was read against the probe source and agrees with it: `ADDTO(5)`, `GROW(k)`, `ADDTO(k)`, `r = TWICE(3)`, `p = NOTHING()`, `BUMP()` in a loop, twice in a row and under THREAD(*SERIALIZE). `lext-call-monitor-c` and `lext-call-nested-c` change only the text of a failure line. - No host round trip decides here: every changed emit still carries failure lines under both binaries (the thread keyword has no emit rule), so no case could match before and none can go from match to worse. Gates (ship.sh, commit e6d9d35f): ```json { "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": "regenerated", "emit": { "worse": [], "better": 0, "newly_clean": 0, "clean_delta": 0, "base": 588, "new": 588 }, "twins": { "groups": 826, "recovered": 804, "regressions": [], "newly": 0 }, "allgroups": { "regressions": [], "newly": [], "worse": [] }, "full_check": { "exit": 0, "seconds": 1120 } } ```
sean force-pushed lext2 from 1dd9aa6bea
Some checks reported errors
binjovi/ci A newer source or plan superseded this Build
to 95b975747b
All checks were successful
binjovi/ci Binjovi completed the frozen plan
2026-11-10 16:58:14 +00:00
Compare
binjovi-bot manually merged commit 95b975747b into trunk 2026-11-10 17:11:00 +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!779
No description provided.