objfile: fallback to modern moduledata layouts when runtimeVersion is unknown

[mlhaigh]

Summary of Changes

When a Go binary has its build information stripped or obfuscated (e.g., using tools like garble that remove runtime.buildVersion), runtimeVersion is detected as "unknown". Previously in objfile/objfile.go, GoReSym fell back to using layoutVersion directly ("1.20" for all Go 1.20+ binaries).

While Go 1.20 through Go 1.25 share identical moduledata struct offsets, Go 1.26 introduced epclntab, shifting subsequent fields (Textsectmap, Typelinks, and ITablinks) by 8 bytes on 64-bit and 4 bytes on 32-bit (e.g., Textsectmap moved from offset 328 to 336 on 64-bit).

Because runtimeVersion was defaulted strictly to "1.20", GoReSym selected the 1.20 layout, causing candidate moduledata validation in validateAndConvertModuleData to fail on stripped Go 1.26 binaries, resulting in {"error": "Failed to parse file: no valid pclntab found"}.

Solution

In objfile/objfile.go, when runtimeVersion is unknown and layoutVersion is "1.20", ModuleDataTable now tests candidate modern layouts ("1.20" followed by "1.26") during candidate validation. Testing "1.20" first ensures backwards compatibility and avoids false-positive matches on Go 1.20–1.25 binaries, while cleanly falling back to "1.26" for stripped Go 1.26 binaries. In addition, candidate ignorelist handling was cleaned up to avoid duplicate entries during candidate layout evaluation.

Testing & Verification

1. Unit Tests: Full test suite passes cleanly:
   • go test -v -count=1 ./objfile: 20 test suites, 70 subtests (90 PASS assertions, 0 failures).
   • go test -v -count=1 ./...: 111 passed, 7 skipped, 0 failures.
2. Obfuscated / Stripped Corpus Validation:
   • Verified across a 40-binary test corpus of Go 1.26 binaries built under standard, stripped, and Garble obfuscation modes (garble, garble -literals, garble -tiny) across Linux amd64, Windows amd64, and Windows 386.
   • Prior to this change, Garble-obfuscated Go 1.26 binaries failed (0/15 passed) with Failed to parse file: no valid pclntab found.
   • With this change, all 40 binaries parse successfully (40/40 passed, 100%) without requiring manual -v flag overrides.

[stevemk14ebr]

thanks for the contribution, small changes requested

[mlhaigh]

Thanks for the review and suggestions @stevemk14ebr!

I've updated the PR with both changes:

• Renamed the candidate slice and loop variable to runtimeVersions / runtimeVersion so the naming stays consistent with the rest of the codebase.
• Preserved the ignorelist = newIgnorelist logic across all validation branches to maintain proper handling of invalid/decoy candidate tables.

All unit tests (./...) pass cleanly and verified. Ready for another look!