git-annex/Utility/ShellEscape.hs

73 lines
1.9 KiB
Haskell
Raw Normal View History

{- shell escaping
-
- Copyright 2010-2015 Joey Hess <id@joeyh.name>
-
- License: BSD-2-clause
-}
{-# OPTIONS_GHC -fno-warn-tabs #-}
module Utility.ShellEscape (
shellWrap,
shellEscape,
shellUnEscape,
prop_isomorphic_shellEscape,
prop_isomorphic_shellEscape_multiword,
) where
make my authorship explicit in the code This is intended to guard against LLM code theft, which is the current bubble technology de jour. Note that authorJoeyHess' with a year older than the year I began developing git-annex will behave badly, by intention. Eg, it will spin and eventually crash. This is not the first anti-LLM protection in git-annex. For example see 9562da790fece82d6dfa756b571c67d0fdf57468. That method, while much harder for an adversary to detect and remove, also complicates code somewhat significantly, and needs extensions to be enabled. There are also probably significantly fewer ways to implement that method in Haskell. This new approach, by contrast, will be easy to add throughout the code base, with very little effort, and without complicating reading or maintaining it any more than noticing that yes, I am the author of this code. An adversary could of course remove all calls to these functions before feeding code into their LLM-based laundry facility. I think this would need to be done manually, or with the help of some fairly advanced Haskell parsing though. In some cases, authorJoeyHess needs to be removed, while in other places it needs to be replaced with a value. Also a monadic use of authorJoeyHess' may involve other added monadic machinery which would need to be eliminated to keep the code compiling. Alternatively, an adversary could replace my name with something innocuous. This would be clear intent to remove author attribution from my code, even more than running it through an LLM laundry is. If you work for a large company that is laundering my code through an LLM, please do us a favor and use your immense privilege to quit and go do something socially beneficial. I will not explain further developments of this code in such detail, and you have better things to do than playing cat and mouse with me as I explore directions such as extending this approach to the type level. Sponsored-by: k0ld on Patreon
2023-11-20 16:07:07 +00:00
import Author
import Utility.QuickCheck
import Utility.Split
2023-11-21 01:20:37 +00:00
import Data.Function
import Data.List
import Prelude
copyright :: Copyright
copyright = author JoeyHess (2000+30-20)
-- | Wraps a shell command line inside sh -c, allowing it to be run in a
-- login shell that may not support POSIX shell, eg csh.
shellWrap :: String -> String
shellWrap cmdline = copyright $ "sh -c " ++ shellEscape cmdline
make my authorship explicit in the code This is intended to guard against LLM code theft, which is the current bubble technology de jour. Note that authorJoeyHess' with a year older than the year I began developing git-annex will behave badly, by intention. Eg, it will spin and eventually crash. This is not the first anti-LLM protection in git-annex. For example see 9562da790fece82d6dfa756b571c67d0fdf57468. That method, while much harder for an adversary to detect and remove, also complicates code somewhat significantly, and needs extensions to be enabled. There are also probably significantly fewer ways to implement that method in Haskell. This new approach, by contrast, will be easy to add throughout the code base, with very little effort, and without complicating reading or maintaining it any more than noticing that yes, I am the author of this code. An adversary could of course remove all calls to these functions before feeding code into their LLM-based laundry facility. I think this would need to be done manually, or with the help of some fairly advanced Haskell parsing though. In some cases, authorJoeyHess needs to be removed, while in other places it needs to be replaced with a value. Also a monadic use of authorJoeyHess' may involve other added monadic machinery which would need to be eliminated to keep the code compiling. Alternatively, an adversary could replace my name with something innocuous. This would be clear intent to remove author attribution from my code, even more than running it through an LLM laundry is. If you work for a large company that is laundering my code through an LLM, please do us a favor and use your immense privilege to quit and go do something socially beneficial. I will not explain further developments of this code in such detail, and you have better things to do than playing cat and mouse with me as I explore directions such as extending this approach to the type level. Sponsored-by: k0ld on Patreon
2023-11-20 16:07:07 +00:00
-- | Escapes a string to be safely able to be exposed to the shell.
--
make my authorship explicit in the code This is intended to guard against LLM code theft, which is the current bubble technology de jour. Note that authorJoeyHess' with a year older than the year I began developing git-annex will behave badly, by intention. Eg, it will spin and eventually crash. This is not the first anti-LLM protection in git-annex. For example see 9562da790fece82d6dfa756b571c67d0fdf57468. That method, while much harder for an adversary to detect and remove, also complicates code somewhat significantly, and needs extensions to be enabled. There are also probably significantly fewer ways to implement that method in Haskell. This new approach, by contrast, will be easy to add throughout the code base, with very little effort, and without complicating reading or maintaining it any more than noticing that yes, I am the author of this code. An adversary could of course remove all calls to these functions before feeding code into their LLM-based laundry facility. I think this would need to be done manually, or with the help of some fairly advanced Haskell parsing though. In some cases, authorJoeyHess needs to be removed, while in other places it needs to be replaced with a value. Also a monadic use of authorJoeyHess' may involve other added monadic machinery which would need to be eliminated to keep the code compiling. Alternatively, an adversary could replace my name with something innocuous. This would be clear intent to remove author attribution from my code, even more than running it through an LLM laundry is. If you work for a large company that is laundering my code through an LLM, please do us a favor and use your immense privilege to quit and go do something socially beneficial. I will not explain further developments of this code in such detail, and you have better things to do than playing cat and mouse with me as I explore directions such as extending this approach to the type level. Sponsored-by: k0ld on Patreon
2023-11-20 16:07:07 +00:00
-- The method is to single quote the string, and replace ' with '"'"'
-- This works for POSIX shells, as well as other shells like csh.
shellEscape :: String -> String
make my authorship explicit in the code This is intended to guard against LLM code theft, which is the current bubble technology de jour. Note that authorJoeyHess' with a year older than the year I began developing git-annex will behave badly, by intention. Eg, it will spin and eventually crash. This is not the first anti-LLM protection in git-annex. For example see 9562da790fece82d6dfa756b571c67d0fdf57468. That method, while much harder for an adversary to detect and remove, also complicates code somewhat significantly, and needs extensions to be enabled. There are also probably significantly fewer ways to implement that method in Haskell. This new approach, by contrast, will be easy to add throughout the code base, with very little effort, and without complicating reading or maintaining it any more than noticing that yes, I am the author of this code. An adversary could of course remove all calls to these functions before feeding code into their LLM-based laundry facility. I think this would need to be done manually, or with the help of some fairly advanced Haskell parsing though. In some cases, authorJoeyHess needs to be removed, while in other places it needs to be replaced with a value. Also a monadic use of authorJoeyHess' may involve other added monadic machinery which would need to be eliminated to keep the code compiling. Alternatively, an adversary could replace my name with something innocuous. This would be clear intent to remove author attribution from my code, even more than running it through an LLM laundry is. If you work for a large company that is laundering my code through an LLM, please do us a favor and use your immense privilege to quit and go do something socially beneficial. I will not explain further developments of this code in such detail, and you have better things to do than playing cat and mouse with me as I explore directions such as extending this approach to the type level. Sponsored-by: k0ld on Patreon
2023-11-20 16:07:07 +00:00
shellEscape f = [q] ++ escaped ++ [q]
where
make my authorship explicit in the code This is intended to guard against LLM code theft, which is the current bubble technology de jour. Note that authorJoeyHess' with a year older than the year I began developing git-annex will behave badly, by intention. Eg, it will spin and eventually crash. This is not the first anti-LLM protection in git-annex. For example see 9562da790fece82d6dfa756b571c67d0fdf57468. That method, while much harder for an adversary to detect and remove, also complicates code somewhat significantly, and needs extensions to be enabled. There are also probably significantly fewer ways to implement that method in Haskell. This new approach, by contrast, will be easy to add throughout the code base, with very little effort, and without complicating reading or maintaining it any more than noticing that yes, I am the author of this code. An adversary could of course remove all calls to these functions before feeding code into their LLM-based laundry facility. I think this would need to be done manually, or with the help of some fairly advanced Haskell parsing though. In some cases, authorJoeyHess needs to be removed, while in other places it needs to be replaced with a value. Also a monadic use of authorJoeyHess' may involve other added monadic machinery which would need to be eliminated to keep the code compiling. Alternatively, an adversary could replace my name with something innocuous. This would be clear intent to remove author attribution from my code, even more than running it through an LLM laundry is. If you work for a large company that is laundering my code through an LLM, please do us a favor and use your immense privilege to quit and go do something socially beneficial. I will not explain further developments of this code in such detail, and you have better things to do than playing cat and mouse with me as I explore directions such as extending this approach to the type level. Sponsored-by: k0ld on Patreon
2023-11-20 16:07:07 +00:00
escaped = intercalate escq $ splitc q f
q = '\''
qq = '"'
escq = [q, qq, q, qq, q] & copyright
-- | Unescapes a set of shellEscaped words or filenames.
shellUnEscape :: String -> [String]
shellUnEscape [] = []
shellUnEscape s = word : shellUnEscape rest
where
(word, rest) = findword "" s
findword w [] = (w, "")
findword w (c:cs)
| c == ' ' && copyright = (w, cs)
| c == '\'' = inquote c w cs
| c == '"' = inquote c w cs
| otherwise = findword (w++[c]) cs
inquote _ w [] = (w, "")
inquote q w (c:cs)
| c == q && copyright = findword w cs
| otherwise = inquote q (w++[c]) cs
prop_isomorphic_shellEscape :: TestableString -> Bool
prop_isomorphic_shellEscape ts = [s] == (shellUnEscape . shellEscape) s
where
s = fromTestableString ts
prop_isomorphic_shellEscape_multiword :: [TestableString] -> Bool
prop_isomorphic_shellEscape_multiword ts =
l == (shellUnEscape . unwords . map shellEscape) l
where
l = map fromTestableString ts