metadata: Field names limited to alphanumerics and a few whitelisted punctuation characters to avoid issues with views, etc.

Types/MetaData.hs

@@ -115,19 +115,29 @@ instance MetaSerializable CurrentlySet where
 	deserialize "-" = Just (CurrentlySet False)
 	deserialize _ = Nothing
 
-{- Fields cannot be empty, contain whitespace, or start with "+-" as
- - that would break the serialization. -}
 toMetaField :: String -> Maybe MetaField
 toMetaField f
 	| legalField f = Just $ MetaField f
 	| otherwise = Nothing
 
+{- Fields cannot be empty, contain whitespace, or start with "+-" as
+ - that would break the serialization.
+ -
+ - Additionally, fields should not contain any form of path separator, as
+ - that would break views.
+ -
+ - So, require they have an alphanumeric first letter, with the remainder
+ - being either alphanumeric or a small set of shitelisted common punctuation.
+ -}
 legalField :: String -> Bool
-legalField f
-	| null f = False
-	| any isSpace f = False
-	| any (`isPrefixOf` f) ["+", "-"] = False
-	| otherwise = True
+legalField [] = False
+legalField (c1:cs)
+	| not (isAlphaNum c1) = False
+	| otherwise = all legalchars cs
+  where
+	legalchars c
+		| isAlphaNum c = True
+		| otherwise = c `elem` "_-."
 
 toMetaValue :: String -> MetaValue
 toMetaValue = MetaValue (CurrentlySet True)

debian/changelog

@@ -11,6 +11,8 @@ git-annex (5.20140222) UNRELEASED; urgency=medium
     tag/showname.
   * annex.genmetadata can be set to make git-annex automatically set
     metadata (year and month) when adding files.
+  * metadata: Field names limited to alphanumerics and a few whitelisted
+    punctuation characters to avoid issues with views, etc.
 
  -- Joey Hess <joeyh@debian.org>  Fri, 21 Feb 2014 13:03:04 -0400
 

doc/metadata.mdwn

@@ -19,7 +19,7 @@ fields, which each can have any number of values. For example, to tag
 files, the `tag` field is typically used, with values set to each tag that
 applies to the file.
 
-The field names are freeform (but cannot include spaces). The metadata
+The field names are limited to alphanumerics (and `[_-.]`). The metadata
 values can contain absolutely anything you like -- but you're recommended
 to keep it simple and reasonably short.