# Line endings, pinned per file type rather than left to each contributor's
# core.autocrlf setting.
#
# Nothing is broken today - the blobs in this repository are already LF, and a
# clone on Linux or macOS gets LF. But nothing enforced that: a contributor on
# Windows with core.autocrlf=false commits exactly what is in their editor, and
# a shell script committed with CRLF fails on the first line of real work with
#
#     set: pipefail: invalid option name
#
# which reads as a broken installer rather than a line-ending problem. This file
# makes the outcome independent of who committed it.

# Anything Git decides is text gets LF in the repository.
* text=auto

# Must be LF wherever they are checked out, because they run on Linux even when
# they were written on Windows.
*.sh          text eol=lf
*.command     text eol=lf
Dockerfile    text eol=lf
*.yml         text eol=lf
*.yaml        text eol=lf

# Windows entry points, which cmd.exe and PowerShell expect to find as CRLF.
*.bat         text eol=crlf
*.cmd         text eol=crlf
*.ps1         text eol=crlf

# Never touched: conversion corrupts these.
*.png         binary
*.gif         binary
*.webp        binary
*.jpg         binary
*.ico         binary
*.woff        binary
*.woff2       binary
*.ttf         binary
*.msi         binary
*.exe         binary
*.pdf         binary

# The front end is built into a committed bundle that CI rebuilds on Linux and compares byte for
# byte. Its sources must be LF on every machine: Vite copies index.html's line endings, so a
# Windows checkout with CRLF produced a bundle a few bytes different from CI's.
web-ui/index.html  text eol=lf
web-ui/src/**      text eol=lf

# The built bundle is generated: never convert its line endings, so a Windows checkout, a Linux
# checkout and CI's rebuild are the same bytes.
creature_lab/web/static/**  -text
