Cesarus85--mjolnir-vr/docs/object_matching_logs/libcmt_gmtime_obj.md

libs/libcmt/gmtime.obj exact-match ledger

Plain-English behavior

gmtime converts a non-negative Unix timestamp into UTC calendar fields. It uses the CRT thread's reusable struct tm buffer, allocating that buffer on first use; if allocation fails it uses the object's zero-initialized fallback buffer. Negative timestamps return null.

The arithmetic divides time into four-year cycles, identifies the year and whether it is the cycle's leap year, then uses the external _lpdays or _days cumulative-month table to derive the month and day. Finally it fills weekday, hour, minute, second, and sets daylight-saving time to false because the result is UTC.

January target inventory

• _gmtime: one 263-byte public .text COMDAT with five relocations.
• _bss_004e4150: one external 36-byte writable fallback struct tm.
• Calls: _getptd, malloc.
• Data dependencies: _lpdays, _days.
• No owned .data or .rdata.

Reconstruction method

The initial readable C follows the January control flow and old Microsoft CRT algorithm directly. It defines typed struct tm and thread-data fields rather than using raw pointer arithmetic. No historical source branch exists for this object; modern CRT source and old algorithm descriptions are corroboration, while January machine code remains authoritative.

Experiment ledger

Shape	Strict measurement	Result
One final return, assigning null to result on negative time	261/263 code bytes; 5/5 relocation identities, all post-guard relocations shifted by exactly -2; BSS exact	Rejected. VC7 reused the null value in ECX, emitted a six-byte near negative branch, and changed the epilogue ordering.
Evidence-backed early return for negative time	263/263 bytes; 5/5 relocations at exact addresses/destinations; normalized SHA-256 ebb4a58ef517ae1661effba4387b102df64296c5f53d2f0d753602de75d386d5; BSS exact	Accepted. The target has a dedicated xor EAX,EAX block and jump to the common restore-only epilogue, so this is the original-style control flow rather than a codegen trick.

Every experiment preserves the natural behavior above and is judged on padded size, normalized bytes, relocation identity/addends, BSS ownership, and the complete object—not merely an objdiff percentage.

The fallback BSS is 36/36 bytes with no relocations and normalized SHA-256 6db65fd59fd356f6729140571b5bcd6bb3b83492a16e1bf0a3884442fc3c8a0e.

Ownership and ABI proof

The final build uses the configured XDK 3911 compiler, CL 13.00.9254.1, with libcmt's unchanged /O1 /Gy /I"xbox/include" flags. The target and candidate both expose _gmtime as an external cdecl function in a no-duplicates .text COMDAT aligned to one byte. The five ordered relocations are exact:

offset	type	destination
+0x0C	REL32	__getptd
+0x26	REL32	_malloc
+0x31	DIR32	_bss_004e4150
+0x87	DIR32	__lpdays
+0x9B	DIR32	__days

Both objects expose _bss_004e4150 as external storage at offset zero of a 36-byte, four-byte-aligned, writable uninitialized section. There are no other target-owned runtime sections. Compiler-only directives and debug metadata in the rebuilt object do not represent linked runtime ownership.

House/Berth audit

The implementation uses ordinary typed C, vertical parameters, an explicit return type, typed struct tm and thread-data fields, and natural arithmetic constants. The early return is retained because the January bytes prove that control flow; converting it to one final return was measured and rejected. There is no assembly, volatile, forced inlining, undefined behavior, packed surrogate, byte patch, raw structure offset, alignment workaround, comparator exception, or compiler-flag override. No original behavioral bug was found, so no bug-preservation note is required.

Campaign gates

• The complete Halo build and complete libcmt build passed from the isolated current-tip lane.
• Regenerated progress credited gmtime.obj as one complete libcmt object, 263 code bytes, one function, and 36 BSS bytes. The lane reported libcmt at 22/212 objects, 23/476 functions, 1,773/55,015 code bytes, and 192/8,637 data bytes.
• The semantic audit evaluated 3,460 functions with zero unit errors and 3,376 accepted exact functions.
• The admission audit produced no candidates or revocations; its sole existing contradiction remains the unrelated known shell_xbox entry.
• The complete tool suite passed 179/179 tests.
• The fail-closed per-unit regression snapshot/check passed after committing the strict candidate: _gmtime is still_exact, with no failures, warnings, changed non-exact siblings, unexpected new matches, or ownership changes.
• git diff --check, final scope inspection, and the prohibited-construct source scan were clean.