fix Arbitrary AssociatedFile to not crash when LANG=C

Even letting through things that Data.Char.generalCategory said
wereUppercaseLetter caused the crash. Apparently what's going on is
that, in LANG=C, it does not expect to find unicode chars in a String,
except presumably ones that are surrogates.

But ascii is good enough to test the things we need to test about
associated files.

Key.hs

@@ -1,6 +1,6 @@
 {- git-annex Keys
  -
- - Copyright 2011-2019 Joey Hess <id@joeyh.name>
+ - Copyright 2011-2020 Joey Hess <id@joeyh.name>
  -
  - Licensed under the GNU AGPL version 3 or higher.
  -}
@@ -28,6 +28,7 @@ module Key (
 	prop_isomorphic_key_encode
 ) where
 
+import Data.Char
 import qualified Data.Text as T
 import qualified Data.ByteString as S
 import qualified Data.Attoparsec.ByteString as A
@@ -79,11 +80,15 @@ instance Arbitrary KeyData where
 		<*> ((succ . abs <$>) <$> arbitrary) -- chunknum cannot be 0 or negative
 
 -- AssociatedFile cannot be empty, and cannot contain a NUL
--- (but can be Nothing)
+-- (but can be Nothing). 
 instance Arbitrary AssociatedFile where
-	arbitrary = (AssociatedFile . fmap toRawFilePath <$> arbitrary)
+  	arbitrary = (AssociatedFile . fmap conv <$> arbitrary)
 		`suchThat` (/= AssociatedFile (Just S.empty))
 		`suchThat` (\(AssociatedFile f) -> maybe True (S.notElem 0) f)
+	  where
+		-- Generating arbitrary unicode leads to encoding errors
+		-- when LANG=C, so limit to ascii.
+		conv = toRawFilePath . filter isAscii
 
 instance Arbitrary Key where
 	arbitrary = mkKey . const <$> arbitrary

doc/bugs/build_of_7.20191230+git152-gefb981388_fails_the_prop__95__read__95__write__95__transferinfo_test.mdwn

@@ -18,3 +18,5 @@ Full build logs are at http://neuro.debian.net/_files/_buildlogs/git-annex/7.201
 
 [[!meta author=yoh]]
 [[!tag projects/datalad]]
+
+> [[fixed|done]] --[[Joey]]

doc/bugs/build_of_7.20191230+git152-gefb981388_fails_the_prop__95__read__95__write__95__transferinfo_test/comment_1_e9db58f71eedc99ccfd7a7a446843316._comment

@@ -0,0 +1,31 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 1"""
+ date="2020-02-02T19:41:34Z"
+ content="""
+Minimal reproducer:
+
+	bash$ LANG=C ghci Utility/FileSystemEncoding.hs
+	ghci> useFileSystemEncoding
+	ghci> toRawFilePath "\611584"
+	"*** Exception: recoverEncode: invalid argument (invalid character)
+
+No such problem in a unicode locale.
+
+The problem does not, though, affect actually using git-annex in LANG=C
+with a filename with that in its name.
+
+Odd because the filesystem encoding is supposed to round-tip well,
+anything, but here encoding a string with it is failing internally.
+Maybe the thing is, it's not really round-tripping? QuickCheck arbitrary
+magics up a FilePath that contains that, so it's starting in the middle and
+trying to convert it out.
+
+[[!commit 70395659db9f662e61009d984fc9b0b2f24fdece]] introduced this while
+fixing another intermittent encoding test case failure.
+
+	ghci> Data.Char.generalCategory  '\611584'
+	NotAssigned
+
+I think it would make sense to filter out NotAssigned and PrivateUse.
+"""]]