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

libs/libcmt/strstr.obj provenance and park ledger

Plain-English behavior

strstr searches a byte string for the first occurrence of another byte string. The January implementation includes a special case for a one-byte needle: instead of running the general substring loop, it transfers directly into the middle of the separately assembled strchr routine.

That last detail is decisive. This is not C that merely happens to compile unusually. It is a hand-written Microsoft Macro Assembler object whose source file and assembler version survive in the shipped XDK object.

Status

Parked as original, vendored Microsoft CRT assembly. No production C file was retained and config/config.json remains MISSING. The object receives no matching credit under the project's readable-C-only policy.

January target inventory

• Runtime section: one 128-byte, 16-byte-aligned .text section
• Defined runtime entry point: _strstr at offset zero
• Runtime relocations: one IMAGE_REL_I386_REL32 relocation at +0x6F to ___from_strstr_to_strchr, addend zero
• Owned .data, .bss, or .rdata: none
• Extra symbol: ___from_strstr_to_strchr is undefined here and is defined by the separate strchr.obj at offset +0x16 inside its .text section

The raw target sequence around the transfer is:

33 C0 5E 5B 5F 8A C2 E9 00 00 00 00

It clears the result register, restores the saved registers, moves the first needle byte into AL, and tail-jumps through the relocation into the internal strchr entry. This is not a call to the public _strchr function boundary.

Authoritative XDK 3911 provenance

The locally installed XDK 3911 archive is:

C:\tmp\xdk3911_extract\XDK\xbox\lib\libcmt.lib
SHA-256: 60AA4C87C97EEB6662B31D9D37CF3CABA1AFABAEC72C47B090E7A0CE8B632F55

Its library index contains obj\i386\strstr.obj and obj\i386\strchr.obj. The extracted strstr.obj has timestamp 2001-08-11 11:19:36, and its COFF .file record is ..\i386\strstr.asm. Its CodeView data identifies:

Microsoft (R) Macro Assembler Version 6.15.8803

The companion member is independently identified as ..\i386\strchr.asm by the same assembler version and defines ___from_strstr_to_strchr at +0x16. This proves both the implementation language and the cross-object internal label; no web source or cross-build inference is needed.

The extracted XDK strstr.obj itself has SHA-256 635D8ED9D2ADCF994CC6BBA8F64AA72C449E2F023329B6B56CECEECA32567ADC. Comparing its _strstr runtime section to the January split target gives:

• size: 128/128
• relocations: 1/1
• normalized SHA-256, both sides: 9b149ba7733ac0566a87a6c361389c8fbfe61b51b0fc881f638975ac0b003797
• relocation address/type/destination/addend: identical
• section_infos_equal = true

The shipped member also contains debug records and a zero-length .data section. Those are non-runtime assembler bookkeeping; the January split object correctly owns only the 128 runtime text bytes.

Measured readable-C experiments

Both experiments used XDK 3911 CL 13.00.9254.1 with the libcmt object's unchanged /O1 /Gy flags. They were disposable fixtures, not production source.

ID	Readable-C source shape	Size / relocs	Normalized SHA / decisive result
C01	Natural nested candidate/needle comparison loop, including empty-needle handling	86 / 0	9bea69da7c64d4e8680c6718ce4fa19af4d53b0b613f6551c2f863b4a8dd863e; 42 bytes short and no relocation
C02	Same natural loop with a one-character fast path calling public strchr	107 / 1	5ad8820f5248eb515e6d307f471bf98f605beb57189183041d352d5a31395da0; relocation is at +0x1F to _strchr, not at +0x6F to the internal label

Declaring ___from_strstr_to_strchr as a C function would not recover source; it would invent a false function boundary and ABI for a label proven to be in the middle of strchr.asm. Inline assembly, an external assembler file, or a byte blob could reproduce the vendor object, but all three are outside the current readable-C exact-match standard. Further C spelling experiments would therefore optimize around the wrong source language and are not justified.

Validation of the docs-only decision

• Full halobetacache_build and libcmt_build: successful
• Progress and semantic reports regenerated without error
• Tooling unit tests: 179/179 passed
• Production/config diff: empty; only this ledger is committed
• Because a MISSING unit has no rebuilt base object, it cannot itself be a regression-gate baseline. A clean snapshot/check of the adjacent accepted libs/libcmt/wcscoll unit is used after this documentation commit to prove the lane has not disturbed generated build or ownership evidence.

Reopen criterion

Reopen only if the project explicitly adopts a reviewed policy for preserving original vendor assembly members (preferably by importing the authenticated XDK member or original assembly source with provenance), or if the standard is changed to permit such non-C runtime objects. Do not reopen for declaration, branch, or register-allocation sweeps.

House/Berth audit

The final repository change is documentation only. No assembly, volatile, forced inline, undefined behavior, byte patching, codegen cast, alignment coercion, compiler-flag change, comparator exception, invented alias, or raw object/tag access was added to production. The target behavior is ordinary CRT behavior rather than a confirmed engine bug, so no bug-preservation comment is appropriate.