test: re-cite retired-mechanism comments in surviving carriers

The migrate-and-retire audit left the surviving carriers as the only
files still describing the retired combined.ww amalgamator as live:
m2wwi/m3sep headers claimed .wwi was dead code, four byteid carriers
carried dead cleanup of never-written .combined.ww intermediates, and
sepbuild's enumeration mirror plus lib_byteid's completeness scan
still skipped the retired artifact form (the same silent
accommodation just removed from both drivers — dropping it here
surfaced and flushed 39 stale untracked amalgamator outputs across
lib/, which the gates now reject loudly). Remaining mentions are
past-tense history or quoted diagnostics.
This commit is contained in:
2026-08-08 01:34:05 +09:00
parent f6628b985e
commit 939d0938e2
19 changed files with 42 additions and 57 deletions

View File

@@ -7,10 +7,9 @@
* so w6c_ww must produce byte-identical output to wwdump_ww -c on
* every program — this test pins that.
*
* (We do not diff against C-side `w6c` here because the C compiler has
* features the ww cgen has not yet ported — float compare, fn-address
* @symbol, indirect call. Test 990 probe 5 covers the C-vs-ww diff on
* the subset that the ww cgen handles today.)
* (We do not diff against C-side `w6c` here: this carrier's oracle is
* the wwdump_ww -c TOOL OUTPUT, not the tool binary. The C-vs-ww diff
* on real programs is owned by the blanket byte-identity gates.)
*/
#include <stdio.h>
#include <stdlib.h>
@@ -189,9 +188,8 @@ main(void)
char cwd[1024];
if (getcwd(cwd, sizeof cwd) == NULL) return 1;
/* In-source corpus: programs the ww-side cgen handles today.
* Mirrors probe 5 in 990_selfhost without the parts that drift
* against C w6c. */
/* In-source corpus: programs the ww-side cgen handles today,
* kept free of the parts that drift against C w6c. */
struct { const char *label; const char *src; } progs[] = {
{ "ret42",
"fn main() i32 = { return 42; };" },
@@ -360,9 +358,9 @@ main(void)
* The load-bearing M4 oracle: build w6c by SEPARATE compilation under
* BOTH driver stages and prove they sep-produce byte-identical per-package
* .s/.o/.wwi (cs==ww on the real tool's sep artifacts, rule 10). The
* combined corpus + the sep-out==combined-out leg are gone — combined-mode
* is being retired (E3 flip), so combined.ww is no longer fed as compiler
* INPUT; the per-pkg cs==ww comparison is the surviving live oracle. Its
* combined corpus + the sep-out==combined-out leg are gone with the
* retired combined mode (E3 flip); the per-pkg cs==ww comparison is
* the surviving live oracle. Its
* cold pid-keyed scratch is wiped after the explicit bootstrap gate. */
{
char inc[4096], cmd[16384];