Repository navigation
t27b: glass_window.t27 lowers -- rem_pio2f pointer call fixed, refusals name both types, @sin/@cos f32 wired (Closes #11788) - #11813
Merged
Conversation
…O undefined symbols + BRANCH26 (#8946) A fn whose body is @extern("sym") imports the C symbol sym. lower names it @sym (a name t27 source cannot declare), passes a str or slice argument as its bytes' address, and codegen emits each call's bl as a tagged data ref that only macho.t27 resolves: one ARM64_RELOC_BRANCH26 against an undefined _sym nlist, distinct names shared, sorted after the externals. Images without imports are byte-identical (reloc_slot is 2*i). The Rust glue nets negative by compacting the a64.rs disassembler tables. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…8946) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…l-11788-extern-boundary # Conflicts: # gen/rust/tri/t27b/macho.rs # specs/tri/t27b/macho.t27
…2_large with pointers, and the reject named nothing (Closes #11788) The chain that blocked make t27os on master: glass_window.t27 pulls specs/tri/t27b/libm.t27 through any libm builtin; lowering the spliced module reached t27b_libm_rem_pio2f, whose large branch still called t27b_libm_rem_pio2_large as (x*, y*, e0, nx, prec) after that routine had been ported to (x0, x1, x2, e0, nx) -> (n, y0, y1). A *[1]f64 reached the f64 coercion and the reject read 'type mismatch at line 1730 (expected f64, found a pointer)', naming nothing actionable. - libm.t27: rem_pio2f returns (n, y) by the file's own port convention; its large branch passes the x pieces as values (nx = 1 reads only x0); sinf and cosf destructure. The stray top-level 'return' after cosf (already on master, never lowered through the splice) is gone. New sweeps hold sinf and cosf to @sin/@cos of an f32 bit for bit across every branch, into rem_pio2f's large one: 32/32 under the reference and under t27b, 0 vacuous. - libm_plan.t27: @sin and @cos of an f32 wire to __t27b_libm_sinf and __t27b_libm_cosf -- the ports sat unwired behind REFUSE_NO_F32. has_f32, routine, the why text and the plan's tests follow; the plan sweep now walks all twenty-one builtins where it stopped at nineteen (18/18 both sides). - lower.rs (+10/-10): the pointer arms of coerce and cond_val name the pointee ('expected f64, found a pointer to *[1]f64'), so a refusal carries both conflicting types next to the line, and the redundant Val::P enum special cases fold into type_name -- the foreign debt does not grow (owner rule 2026-10-09). source.rs pins the new text. Evidence: make t27os builds T27OS.app and t27os-pty here (build 7215); the window launches. cargo test -p t27b 162/162. Seals re-stamped for both specs (check_seal_currency: the five stale rows left are the t27f ones master already carries; check_seal_coverage exits 0).
Contributor
Summary by t27-beesChecked by the t27 compilert27c parse, parse-complete and typecheck on each changed .t27 file at
Read by the modelThis PR fixes a Worth a look
Read: 11 files, +516 -282, the whole diff up to 60000 characters (the rest was not read). Head |
…dth, opaque black #000000 in the full window and full screen Owner, 2026-10-23, said directly: the matte glass was in the compact (docked) mode, and in the full one black #000000 -- isolate it and do not break it. Until today the decision lived in a full-screen bool threaded through backdrop_alpha, tint_alpha and the surface family, so the full 960x540 window rendered as glass. The rule now lives in exactly one pure fn, on_glass(sw): the docked panel is PANEL_W wide, the full window W_PT, so the scene width the frames already carry decides, and docking resizes the window -- the material follows the dock with no second state to keep in step (a module var is impossible here: the Mach-O object has no writable data section, VarDecl(module, object file)). The readers are backdrop_alpha, backdrop_clears, tint_alpha, surface_alpha_with and draw_ground's frost; every draw path that needed the width now carries sw (the card chain, the side rows, the lane tail, the plate and the page fns) -- t27b typechecks the whole file, no decision left outside on_glass. Tests pin the mapping both ways: glass at PANEL_W (backdrop 0.0, tint 0.55, surfaces their tints, frost on), opaque #000000 at W_PT (backdrop 1.0, tint 1.0, every surface 1.0, the same as the frame and the layer). 62/62 under t27b test. The reference test-report block on this file is master's (the master text blocks the same way at spec.zig:217) -- the window's gate is t27b, and make t27os builds and launches (build 7216).
… (owner, 2026-10-23) The avatar picture and the name come from the GitHub user after the sign-in that already exists (share.t27's device flow): the host fetches the signed-in user and writes both through the existing atomic profile.t27p / avatar.png contract. Nothing manual. The recorded decision lives in profile.t27's header with what stays as the fallback until the sign-in slice of #11560 is wired.
Contributor
PR DashboardGenerated at: 2026-10-23 13:27:16 UTC
Summary
Seal Status
|
…ncy, the reference boundary - libm_trigf.t27 (conformance, the libm_logf shape): @sin and @cos of an f32 held to the reference bit for bit -- 3000+2990-point sweeps over both bit patterns (every exponent, both signs), 300 points inside rem_pio2f's large branch, dense near one, the kernel bands, fixed reference bits extracted from this repository's own reference via the assert_eq prelude, signed zero on the bits, inf/NaN held to 'is a NaN' per side. 5/5 under t27c test-report and under t27b test, 0 vacuous. - seal currency: compiler/cli/git.t27 and the three t27f page specs re-sealed where master's merge train moved the specs after sealing -- check_seal_currency reports 0 STALE again; check_seal_coverage 0 with the new spec's seal (3046 seals). - glass_window.t27 header: the reference block on this file is by construction -- gen-zig renders @extern verbatim and Zig takes two arguments (master's text blocks the same way). The gates that hold it are t27b and make t27os; the extern-stub slice for gen-zig is filed with evidence on #5980.
…aster's copy # Conflicts: # .trinity/seals/cli_git_commands.json # .trinity/seals/git_commands.json # .trinity/seals/t27f_T27fWorkGraph.json # .trinity/seals/t27f_display_list.json # .trinity/seals/t27f_export_video.json
Contributor
PR DashboardGenerated at: 2026-10-23 13:44:47 UTC
Summary
Seal Status
|
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
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.
Closes #11788.
Root cause
glass_window.t27pullsspecs/tri/t27b/libm.t27through any libm builtin; lowering the spliced module reachedt27b_libm_rem_pio2f, whose large branch still calledt27b_libm_rem_pio2_largeas(x*, y*, e0, nx, prec)after that routine had been ported to(x0, x1, x2, e0, nx) -> (n, y0, y1). A*[1]f64reached the f64 coercion and the reject readtype mismatch at line 1730 (expected f64, found a pointer)-- a merged line that moved with nothing and named no construct. The line did not map to the source becauseself.linewas stale: the construct lived inside the spliced libm closure.What changed
rem_pio2freturns(n, y)by the file's own port convention; the large branch passes the x pieces as values (nx = 1reads onlyx0);sinf/cosfdestructure. The stray top-levelreturnaftercosf(already on master, never lowered through the splice) is deleted. New sweeps holdsinf/cosfto@sin/@cosof an f32 bit for bit across every branch, intorem_pio2f's large one.@sin/@cosof an f32 wire to__t27b_libm_sinf/__t27b_libm_cosf-- the ports existed and sat unwired behindREFUSE_NO_F32(the why-text also claimedtanhunported; it is wired). The plan sweep now walks all twenty-one builtins where it stopped at nineteen, soSIN/TANHare covered for the first time.coerceandcond_valname the pointee --expected f64, found a pointer to *[1]f64-- so a refusal carries both conflicting types next to the line (t27b: glass_window.t27 fails on master -- type mismatch (expected f64, found a pointer) blocks make t27os #11788's requirement); the redundantVal::Penum special cases fold intotype_name.source.rspins the new text. Replacement plan for this glue stays on port epic epic: t27b in t27 -- port the backend and its tooling from Rust/Python into .t27 #6198.Evidence
make t27osbuildstarget/t27os/T27OS.appandt27os-pty(build 7215, ad hoc signed); the window launches.libm.t27: 32/32 undert27c test-reportAND undert27b test, 0 vacuous.libm_plan.t27: 18/18 both sides.cargo test -p t27b: 162/162.check_seal_currencyleaves only the five t27f rows master already carries;check_seal_coverageexits 0.Follow-ups (not in this PR)
specs/tri/t27b/conformance/libm_trigf.t27, in thelibm_logf.t27shape) once the reference bits are measured on the lab.#000000-- to be isolated as the single material switch inglass_window.t27with tests pinning it.