Add a reproducible MinGW PE EH-frame fixture

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Problem

The coordinated CLE and angr regressions need a real PE whose wrapper at 0x401006 already owns the bytes of an unnamed six-byte function at 0x40100a. Binaries master has no tests/x86/windows/eh-frame-occupied-start.exe, so that GNU .eh_frame boundary cannot be exercised.

Root cause

Ordinary PE fixtures do not combine an occupied CFG range, no symbol for the inner target, and a GNU FDE that is the only metadata identifying the target as a function.

Fix

Add a source-built i386 MinGW PE together with its complete assembly source, BSD-2-Clause license, and deterministic build recipe. The sole GNU FDE starts at 0x40100a; the committed PE is timestamp-free and stripped of COFF symbols.

Testing

Exact-head readback reports PE32/i386, image base 0x400000, entry RVA 0x1000, zero timestamp, zero COFF symbols, and SHA-256 1cd9ae0b00d1c4cde4cd95579cf9aae14104d9f99031aab77b0116e4e9aa7c8a. Two documented GCC 15.3.0/Binutils 2.46 builds were byte-identical. Validation: #196 (comment)

session: sharpen

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Validation record for head ffc011f3de013f23711ded6a05a48df5ea2f726f against baseline a87538bc248531d2a1d22434143e8b9752459bac.

• Exact PR shape: the own-merge-base/head scan verifies the added compiler-produced fixture and its load-bearing format/address shape: PE32 i386 executable; .eh_frame carries an FDE for 0x40100a of size 6.
• Provenance/reproduction: two fresh GCC 15.3.0 / GNU Binutils 2.46 builds with zero timestamp was rerun twice in separate fresh directories; both outputs byte-compare with each other and with the committed bytes.
• Input/native identity: committed SHA-256 1cd9ae0b00d1c4cde4cd95579cf9aae14104d9f99031aab77b0116e4e9aa7c8a; angr rustylib mapped from the isolated consumer worktrees on both arms, SHA-256 d02d517eceb3e81de4a96330678444f0faf2c3b065953f0cb8e1813583a0380f; CLE is Python-only here.
• Real consumer: PE: Separate GNU EH-frame hints from unwind entries cle#789 and CFGFast: Preserve authoritative function starts during reconstruction angr#6948 at baseline cle a4fb8003198229d33c84df6a82f749729232fd31; angr b0feae57a9e401292e03a6f9ffbd4b388d172bf5 and head cle 89ae8c133b4c9c2ba2921429d3504ac2a0624407; angr 1b99ccbe452ed556e79177c41e624093f26ad8ed loaded the committed fixture and produced the complete output linked in this PR's output record.
• Repeat/order control: A/B/B/A, ASLR disabled and PYTHONHASHSEED=0 for the generated angr text; the CLE hint record is a static structural capture; same-arm output is byte-identical and the cross-arm delta is nonempty.
• Local scope: this recapture validates fixture bytes, fixture shape, and the actual consumer path only. It does not claim a fresh full-workspace gate.
• Hosted status: fresh hosted dependency validation for this exact binaries head has not been observed in this recapture; it remains pending and no hosted success is claimed here.

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Exact fixture-shape and consumer-output record for #196.

• Fixture comparison: own merge base a87538bc248531d2a1d22434143e8b9752459bac versus exact head ffc011f3de013f23711ded6a05a48df5ea2f726f. The target path/shape is absent on the merge base and present on the head. PE32 i386 executable; .eh_frame carries an FDE for 0x40100a of size 6.
• Fixture identity: SHA-256 1cd9ae0b00d1c4cde4cd95579cf9aae14104d9f99031aab77b0116e4e9aa7c8a; source provenance: the adjacent complete assembly source at this exact binaries head (SHA-256 1b2246525aca4cddbb510b4836116cf4aa78d8d4722ff4cb58caf7e029dbd99b). The complete readelf/objdump scan is retained with the exact script. The recorded recipe (two fresh GCC 15.3.0 / GNU Binutils 2.46 builds with zero timestamp) was rerun twice in fresh directories; both runs are byte-identical to the committed fixture.
• Actual consumer: PE: Separate GNU EH-frame hints from unwind entries cle#789 and CFGFast: Preserve authoritative function starts during reconstruction angr#6948, exact baseline cle a4fb8003198229d33c84df6a82f749729232fd31; angr b0feae57a9e401292e03a6f9ffbd4b388d172bf5 versus exact head cle 89ae8c133b4c9c2ba2921429d3504ac2a0624407; angr 1b99ccbe452ed556e79177c41e624093f26ad8ed; target CLE function hints and angr function 0x40100a; configuration CLE load plus full / SAILR.
• Native identity: angr rustylib mapped from the isolated consumer worktrees on both arms, SHA-256 d02d517eceb3e81de4a96330678444f0faf2c3b065953f0cb8e1813583a0380f; CLE is Python-only here.
• Determinism: fresh A/B/B/A order, ASLR disabled and PYTHONHASHSEED=0 for the generated angr text; the CLE hint record is a static structural capture; each same-arm pair is byte-identical and every base/head comparison below is nonempty. Absolute checkout prefixes in tracebacks are normalized to <workspace> for publication; raw output is retained unchanged.

Before — cle

Exact consumer baseline

sections=['.edata', '.eh_frame', '.idata', '.rdata', '.text']
function_hints=[]

After — cle

Exact consumer head

sections=['.edata', '.eh_frame', '.idata', '.rdata', '.text']
function_hints=[(4198410, 6, '0')]

Before — angr

Exact consumer baseline

Traceback (most recent call last):
  File "<workspace>/scratch/apply-latest-skills-pr-audit-20260828/output-regeneration-medium/capture_decompile.py", line 125, in <module>
    function = cfg.functions[int(args.function_address, 0)]
               ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<workspace>/scratch/apply-latest-skills-pr-audit-20260828/fixture-output-recapture/consumers/worktrees/6948-base/angr/knowledge_plugins/functions/function_manager.py", line 1175, in __getitem__
    raise KeyError(k)
KeyError: 4198410

After — angr

Exact consumer head

unsigned int sub_40100a(void)
{
    return 42;
}

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Rebased onto binaries master; git range-diff reports the commit unchanged (=), so this is a base move only.

The reason: angr/cle#789 pins this pull request with a sync: line, and CI resolves a sibling at refs/pull/N/head rather than merged with master. At three commits behind, this tree was missing tests/i386/deep_sp_chain, which tests/analyses/decompiler/test_block_simplifier.py loads. That test is not yet in the angr tree those shards resolve, so the failure was dormant rather than live — it would have appeared the moment the resolved angr tree gained it, on a diff that has nothing to do with it.

session: sharpen