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

libs/libcmt/fputc.obj provenance and extraction-blocker ledger

This atomic lane starts from campaign commit 7d664c91. At lane start the object was MISSING, absent from config/parked.json, and had no source, object ledger, topic branch, or active worktree. The January target contains two 88-byte routines, _fputc and _putc, plus one 12-byte SEH scope table owned by each function.

Plain-English behavior

Both public routines do the same thing. They lock a stream, decrement its remaining-buffer count, and either place one byte directly into the buffer or call _flsbuf when the buffer is full. A compiler-generated __finally region always unlocks the stream. The result is the written unsigned byte or the _flsbuf result.

Primary provenance

The XDK 3911 libcmt.lib archive has SHA-256 60aa4c87c97eeb6662b31d9d37cf3caba1afabaec72c47b090e7a0ce8b632f55. Its actual obj\i386\fputc.obj member identifies ordinary Microsoft C source d:\xbox-finalfre\private\sdktools\crt\crtw32\stdio\fputc.c, compiler identifier 0x001c2426, and the two public functions. This is compiled C, not vendor assembly.

The authentic archive member reproduces the January linked instructions and SEH records. It also exposes a strict-comparison boundary introduced when csplit reconstructed an object from the linked executable: csplit cannot recover whether two already-resolved same-section displacements originally had COFF relocations, so it synthesizes symbolic relocations that are absent from the authentic Microsoft object.

Exact measured obstruction

For both _fputc and _putc:

• code size is 88 bytes in January and the authentic XDK object;
• the ordinary external relocations occur at the same addresses and target __SEH_prolog, __lock_file, __flsbuf, __SEH_epilog, and __unlock_file;
• the authentic object has six relocations;
• the January csplit target has seven relocations because it adds a synthetic IMAGE_REL_I386_REL32 at function offset 0x40;
• that synthetic relocation spells the compiler-generated local-finally call as _fputc+0x50 or _putc+0x50; the authentic object encodes the already resolved call +0x0c with no relocation;
• each authentic 12-byte SEH table is ffffffff 00000000 00000000 with a relocation at +8 to the private local finally label; each csplit table is ffffffff 00000000 4d000000 with a relocation to the public function plus addend 0x4d. Both resolve to the same handler address.

The hardened comparator therefore reports, for each function, target 88 bytes / 7 relocs versus archive 88 bytes / 6 relocs. The target normalized SHA-256 is c077a79c43afec8ff22c594c9c8b1037987c4e5aed20e31f6213b393e17ab16b; the authentic-object normalized SHA-256 is d042946199d1efe48d27be97a6dd046dc48bb2e817a7fc7cd797f2afae9f3062. The difference is relocation representation, not a missing algorithm or a compiler code-generation tie.

Why readable C cannot close it

The extra csplit relocation does not correspond to a source-level call. It is the compiler's private call into the local __finally region generated for a locked CRT stream operation. Ordinary VC7 C emits that same-section call as a resolved displacement, exactly as the authenticated XDK object proves. C has no identifier for the compiler-generated finally label and cannot request a COFF relocation to the enclosing public function plus an interior addend.

Replacing the finally region with a separately named C helper would invent a false ABI, change the scope-table target and function inventory, and no longer be the authenticated implementation. Assembly, byte patches, hand-authored COFF, comparator waivers, and source constructs used only to forge relocation records are prohibited. No source-shape experiment can change a relocation that was created later by executable splitting rather than by CL.

Disposition and reopen criteria

This object is an honest extraction/provenance park. Production libs/libcmt/fputc.c was never created, config/config.json remains MISSING, and no matching credit or semantic exception is granted. The authentic readable source is not admitted because the current project standard requires exact symbolic relocation identity, not merely identical linked behavior.

Reopen only if one of these project-wide policy questions is resolved through independent review:

1. regenerate the January target object while preserving original same-section relocation provenance;
2. add a fail-closed, generally applicable rule for csplit-generated local SEH relocations, with ambiguity tests and no change to external relocation identity; or
3. obtain an original January COFF member that independently proves different relocation provenance from XDK 3911.

Do not retry statement ordering, local declarations, barriers, pragmas, or compiler flags: none can control this post-link csplit artifact.

House/Berth and integrity audit

No production C, configuration, matching ledger, compiler flag, comparator, or target artifact was changed. The investigation introduced no assembly, volatile, forced inline, undefined behavior, byte forcing, magic offsets, or invented declarations. The archive and target were read only. This ledger is the sole intended tracked result, and it deliberately grants zero object or function credit.