navicore/patch-prolog
1
0
Fork 0
Code Issues Pull requests Releases 11 Activity Actions

● M5 complete — and with it, all six milestones. Final just ci is green, just diff-test matches the oracle. #3

Merged
navicore merged 1 commit from M5 into main 2026-06-04 21:43:12 +00:00
Conversation 0 Commits 1 Files changed 8 +205 -43
navicore commented 2026-06-04 21:39:39 +00:00
Owner
Copy link

M5 summary

Binary size: 4.4M → 676K. The surprise: the DWARF wasn't coming from -g at all — it rides in from the runtime
archive's object files (Rust std debug info), and the linker copies it unless told otherwise. Release links now
pass -Wl,--strip-debug; --debug builds keep DWARF + -O0. That puts plgc binaries under the patch-seq 730K bar.

Guarded, not just measured: new tests/binary_size.rs in just ci — a 1.3M size ceiling (catches a lost linker
flag or runtime bloat) and an ldd check enforcing the standalone contract: nothing beyond
libc/libm/libgcc/loader, ever.

Script mode: #!/usr/bin/env plgc works end-to-end — ./greet.pl --query "greet(X,Y)" compiles to a temp binary
and execs it. The parser blanks a leading #! line, preserving error line numbers.

Docs trued up: COMPILATION_MODEL.md now honestly states what's inline (dispatch switches, control flow,
constants) vs. runtime calls (head unify, arithmetic eval) with inlining as the named escape hatch; README
rewritten with real numbers; footprint recorded in the ROADMAP.

The project, end to end

┌────────────────┬─────────────────────────┬───────────────────────────────────────────────────────────────┐
│ │ patch-prolog (v1) │ patch-prolog2 │
├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
│ "Compile" │ scaffolds cargo │ LLVM IR → clang, needs nothing but clang │
│ │ project, needs Rust │ │
├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
│ Execution │ tree-walking │ native code, choice points, musttail TCO │
│ │ interpreter, always │ │
├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
│ 5k-chain query │ ~10s │ 3ms (first-arg indexing) │
├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
│ hello-world │ n/a (cargo-built │ 676K, libc/libm only │
│ binary │ interpreter) │ │
├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
│ Semantics │ the spec │ byte-parity (200-assertion ported corpus + 57-goal diff │
│ │ │ harness), ISO where v1 had bugs — each documented │
└────────────────┴─────────────────────────┴───────────────────────────────────────────────────────────────┘

Test totals: 123 runtime + 48 frontend + 54 shared/compiler unit, 89 ported v1-corpus, 14 M3-suite, 14
integration, 12 golden-IR, 2 binary-hygiene, 1 differential — all green in ~12s of CI test time.

What's deliberately left on the table (ROADMAP "Future"): REPL/LSP (v1 had them; they'd drive the same compiled
runtime now), WAM-level codegen and inline fast paths if performance ever demands, copying GC, bagof/setof. The
reserved_builtin machinery is gone entirely — the language surface is exactly v1's, compiled.

M5 summary Binary size: 4.4M → 676K. The surprise: the DWARF wasn't coming from -g at all — it rides in from the runtime archive's object files (Rust std debug info), and the linker copies it unless told otherwise. Release links now pass -Wl,--strip-debug; --debug builds keep DWARF + -O0. That puts plgc binaries under the patch-seq 730K bar. Guarded, not just measured: new tests/binary_size.rs in just ci — a 1.3M size ceiling (catches a lost linker flag or runtime bloat) and an ldd check enforcing the standalone contract: nothing beyond libc/libm/libgcc/loader, ever. Script mode: #!/usr/bin/env plgc works end-to-end — ./greet.pl --query "greet(X,Y)" compiles to a temp binary and execs it. The parser blanks a leading #! line, preserving error line numbers. Docs trued up: COMPILATION_MODEL.md now honestly states what's inline (dispatch switches, control flow, constants) vs. runtime calls (head unify, arithmetic eval) with inlining as the named escape hatch; README rewritten with real numbers; footprint recorded in the ROADMAP. The project, end to end ┌────────────────┬─────────────────────────┬───────────────────────────────────────────────────────────────┐ │ │ patch-prolog (v1) │ patch-prolog2 │ ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤ │ "Compile" │ scaffolds cargo │ LLVM IR → clang, needs nothing but clang │ │ │ project, needs Rust │ │ ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤ │ Execution │ tree-walking │ native code, choice points, musttail TCO │ │ │ interpreter, always │ │ ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤ │ 5k-chain query │ ~10s │ 3ms (first-arg indexing) │ ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤ │ hello-world │ n/a (cargo-built │ 676K, libc/libm only │ │ binary │ interpreter) │ │ ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤ │ Semantics │ the spec │ byte-parity (200-assertion ported corpus + 57-goal diff │ │ │ │ harness), ISO where v1 had bugs — each documented │ └────────────────┴─────────────────────────┴───────────────────────────────────────────────────────────────┘ Test totals: 123 runtime + 48 frontend + 54 shared/compiler unit, 89 ported v1-corpus, 14 M3-suite, 14 integration, 12 golden-IR, 2 binary-hygiene, 1 differential — all green in ~12s of CI test time. What's deliberately left on the table (ROADMAP "Future"): REPL/LSP (v1 had them; they'd drive the same compiled runtime now), WAM-level codegen and inline fast paths if performance ever demands, copying GC, bagof/setof. The reserved_builtin machinery is gone entirely — the language surface is exactly v1's, compiled.
● M5 complete — and with it, all six milestones. Final just ci is green, just diff-test matches the oracle.
All checks were successful
CI - Linux / CI - Linux x86_64 (pull_request) Successful in 47s
Details
f98881bd58 
M5 summary

  Binary size: 4.4M → 676K. The surprise: the DWARF wasn't coming from -g at all — it rides in from the runtime
  archive's object files (Rust std debug info), and the linker copies it unless told otherwise. Release links now
  pass -Wl,--strip-debug; --debug builds keep DWARF + -O0. That puts plgc binaries under the patch-seq 730K bar.

  Guarded, not just measured: new tests/binary_size.rs in just ci — a 1.3M size ceiling (catches a lost linker
  flag or runtime bloat) and an ldd check enforcing the standalone contract: nothing beyond
  libc/libm/libgcc/loader, ever.

  Script mode: #!/usr/bin/env plgc works end-to-end — ./greet.pl --query "greet(X,Y)" compiles to a temp binary
  and execs it. The parser blanks a leading #! line, preserving error line numbers.

  Docs trued up: COMPILATION_MODEL.md now honestly states what's inline (dispatch switches, control flow,
  constants) vs. runtime calls (head unify, arithmetic eval) with inlining as the named escape hatch; README
  rewritten with real numbers; footprint recorded in the ROADMAP.

  The project, end to end

  ┌────────────────┬─────────────────────────┬───────────────────────────────────────────────────────────────┐
  │                │    patch-prolog (v1)    │                         patch-prolog2                         │
  ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
  │ "Compile"      │ scaffolds cargo         │ LLVM IR → clang, needs nothing but clang                      │
  │                │ project, needs Rust     │                                                               │
  ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
  │ Execution      │ tree-walking            │ native code, choice points, musttail TCO                      │
  │                │ interpreter, always     │                                                               │
  ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
  │ 5k-chain query │ ~10s                    │ 3ms (first-arg indexing)                                      │
  ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
  │ hello-world    │ n/a (cargo-built        │ 676K, libc/libm only                                          │
  │ binary         │ interpreter)            │                                                               │
  ├────────────────┼─────────────────────────┼───────────────────────────────────────────────────────────────┤
  │ Semantics      │ the spec                │ byte-parity (200-assertion ported corpus + 57-goal diff       │
  │                │                         │ harness), ISO where v1 had bugs — each documented             │
  └────────────────┴─────────────────────────┴───────────────────────────────────────────────────────────────┘

  Test totals: 123 runtime + 48 frontend + 54 shared/compiler unit, 89 ported v1-corpus, 14 M3-suite, 14
  integration, 12 golden-IR, 2 binary-hygiene, 1 differential — all green in ~12s of CI test time.

  What's deliberately left on the table (ROADMAP "Future"): REPL/LSP (v1 had them; they'd drive the same compiled
  runtime now), WAM-level codegen and inline fast paths if performance ever demands, copying GC, bagof/setof. The
  reserved_builtin machinery is gone entirely — the language surface is exactly v1's, compiled.
navicore referenced this pull request from a commit 2026-06-18 18:09:27 +00:00
⏺ Stage B is built and green.
navicore referenced this pull request from a commit 2026-06-20 22:03:21 +00:00
PR #28 feedback addressed
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
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
navicore/patch-prolog!3
No description provided.