iframe-proxy | Sunbelt Computer Software

Repository navigation

# Line endings for the evidence corpus.

# The body hashes in research/raw/.fetches.jsonl were computed over the bytes the

# collector fetched, which end in LF. On a default Windows checkout

# (core.autocrlf=true) git's smudge filter rewrites every text file it checks out

# to CRLF, so every one of those hashes stops recomputing: verifyLedger reports

# all six captures as body-unmodified, handoff and preflight exit with blockers,

# and the two-machine workflow is unusable on a default Windows builder.

# Nothing was lost and nothing needs re-collecting - the files travelled fine and

# git rewrote them locally on the way out of the object store. Pinning them here

# makes the checkout write what the ledger hashed, on every platform, regardless

# of the reader's core.autocrlf. This file outranks that setting.

# See ADR-0020 and fetchproof/lib/handoff.mjs, which detects this exact case

# and says so instead of blaming a missing push.

research/raw/* text eol=lf

*.jsonl text eol=lf

research/*.md text eol=lf

# And the same reasoning, applied to the rest of the repository rather than only to the

# part whose hashes made the problem visible. Without this, every other text file still

# follows the reader's core.autocrlf: `git add` warns on ~200 files, and a checkout on a

# second machine differs byte-for-byte from this one for no reason anybody chose.

* text=auto eol=lf

# The commit gate is a POSIX sh script. CRLF in a shebang script is a syntax error

# on every POSIX host, and the failure is a hook that cannot run - which, per

# ADR-0001, git reports as a clean commit.

fetchproof/githooks/pre-commit text eol=lf