research: Prolog frontend on Prologos post-elaboration IR - #56
Draft
kumavis wants to merge 1 commit into
Draft
Conversation
Stage 1 research note categorizing standard Prolog features into compatible-as-is vs small/medium/large additional work to land them on Prologos's existing relational IR (expr-defr / expr-clause / goal-desc) and BSP-LE propagator runtime. Headline finding: ~95% of Datalog-fragment Prolog and ~70% of ISO-7 Prolog map directly onto existing IR; the residual gap is concentrated in dynamic features (assert/retract, attributed vars, CHR) which are paradigm-level mismatches against CALM-monotone propagator semantics and naturally co-evolve with the PReduce series. Includes architectural caveats (tabling default, fork-not-stack, closed goal-kind set), four-stage staging proposal, and open questions on cut semantics, side-effect ordering, and marketing.
kumavis
marked this pull request as draft
May 7, 2026 06:10
kumavis
pushed a commit
that referenced
this pull request
Aug 1, 2026
#55 and #56 are the two decoder traps that between them produced a handoff which fetched correctly, received a well-formed fulfill, and then did nothing at all -- no error, no log, no failed assertion. #55: a ONE-argument Syrup record's payload is the value itself, not a one-element list, so `record-args-of` reports zero arguments for `<desc:import-object 0>`. It hides because every multi-field record in the port reads fine through the same helper, and because "no reference in this message" is a protocol condition that can legitimately happen. #56: a wire nat arrives as `syrup-nat` OR a non-negative `syrup-int`, and matching only the first silently misses the other half. The codec already owns the conversion (`wire-nat`); re-deriving it is the bug. The rule generalises -- when the codec ships a predicate for a wire-level notion, a local match on one constructor is waiting for the other spelling. The third note is bookkeeping with teeth: the seven Node-spawning interop tests fail on an UNCHANGED tree at e65e0f8, a commit that was CI-green three days earlier, and the CI variant of the same family first failed on a documentation-only commit. Both look like regressions and neither is one. Also records that an `import('@endo/ocapn')` without `@endo/init` fails by design, so that error is not evidence of a broken dependency -- it looked like one for a few minutes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YM6gc3cMNH2Ymor4jdZY8u
kumavis
pushed a commit
that referenced
this pull request
Aug 1, 2026
…built on that The doc-truth review found the premise the whole CI investigation rested on, and it was wrong. Both workflows trigger `on: pull_request` with a bare `actions/checkout@v4`, so GitHub builds refs/pull/N/merge -- the branch merged into main. Main's LET series landed 2026-07-31, between the last green run (07-29) and the first red one (08-01). THE CODE UNDER TEST CHANGED WHILE MY DIFF DID NOT. Every "a docs-only commit failed, so the diff cannot have caused it" argument was reasoning about the wrong tree. So #57's four sections (~250 lines) are replaced by ONE resolved entry rather than appended to, because a wrong investigation left readable as current fact is the exact failure the entry is now about. It records the rule (on pull_request the tested tree is branch + base; check what moved on the base before calling a failure unattributable), the corollary that cost as much (when a failure resists local repro, MERGE THE BASE FIRST -- doing so turned this into a 6-second deterministic local failure), and the four wrong conclusions it produced. Other false claims corrected, all mine: #52's cause statement was FALSE. A forward `defn` reference does not silently fail: under `process-file` residuation RESOLVES it (0 errors); under `process-string` it RAISES. That asymmetry is the real lesson and is precisely what produced the false all-clear the entry warns about -- a process-file probe resolves what the server's process-string path rejects. The verified narrow claim (a stuck term is well-typed) is re-anchored to the shape that actually produces one, a `def` value whose referent never grounds. The "nine of thirty-five" count is unverifiable (the broken state was never committed) and is now stated as such. #53's quoted diagnostic no longer exists at HEAD -- marked stale. #54's "cannot read .prologos" overstated; it is content-dependent, and the transcript quoted did not reproduce. #56's "coin flip" was wrong: the spelling is determined by provenance, not chance. #51 FAMILY's retraction lived 245 lines away; now corrected in place. Gaps doc §1.7 M8's "there is one Syrup parser on a frame again" is literally false -- run-ocapn-test-server.rkt:1140 still parses every frame as a discard-result validation check. Only one implementation ACTS, which is the substantive fix; the parser count was overstated. The § Status section contradicted itself (line 27 promised a "Still open" section that line 111 says does not exist) plus a dangling pointer; both corrected. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YM6gc3cMNH2Ymor4jdZY8u
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stage 1 research note categorizing standard Prolog features into
compatible-as-is vs small/medium/large additional work to land them
on Prologos's existing relational IR (expr-defr / expr-clause /
goal-desc) and BSP-LE propagator runtime.
Headline finding: ~95% of Datalog-fragment Prolog and ~70% of ISO-7
Prolog map directly onto existing IR; the residual gap is concentrated
in dynamic features (assert/retract, attributed vars, CHR) which are
paradigm-level mismatches against CALM-monotone propagator semantics
and naturally co-evolve with the PReduce series.
Includes architectural caveats (tabling default, fork-not-stack,
closed goal-kind set), four-stage staging proposal, and open
questions on cut semantics, side-effect ordering, and marketing.