allow glob matches of the form dir/**/FileNoExtension (#8005)

* allow glob matches of the form dir/**/FileNoExtension

* allow glob matches of the form dir/**/FileNoExtension

* tests, docs, changelong

Co-authored-by: Gershom Bazerman <[email protected]>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

Cabal-tests/tests/CheckTests.hs

@@ -36,6 +36,8 @@ checkTests = testGroup "regressions"
     , checkTest "pre-1.6-glob.cabal"
     , checkTest "pre-2.4-globstar.cabal"
     , checkTest "bad-glob-syntax.cabal"
+    , checkTest "globstar-literal.cabal"
+    , checkTest "pre-3.8-globstar-literal.cabal"
     , checkTest "cc-options-with-optimization.cabal"
     , checkTest "cxx-options-with-optimization.cabal"
     , checkTest "ghc-option-j.cabal"

Cabal-tests/tests/ParserTests/regressions/globstar-literal.cabal

@@ -0,0 +1,16 @@
+cabal-version: 3.8
+name: globstar-literal
+version: 0
+extra-source-files:
+  foo/**/bar
+
+license: BSD-3-Clause
+synopsis: no
+description: none
+category: Test
+maintainer: none
+
+library
+  default-language: Haskell2010
+  exposed-modules:
+    Foo

Cabal-tests/tests/ParserTests/regressions/pre-3.8-globstar-literal.cabal

@@ -0,0 +1,16 @@
+cabal-version: 3.6
+name: globstar-literal
+version: 0
+extra-source-files:
+  foo/**/bar
+
+license: BSD-3-Clause
+synopsis: no
+description: none
+category: Test
+maintainer: none
+
+library
+  default-language: Haskell2010
+  exposed-modules:
+    Foo

Cabal-tests/tests/ParserTests/regressions/pre-3.8-globstar-literal.check

@@ -0,0 +1 @@
+In the 'extra-source-files' field: invalid file glob 'foo/**/bar'. Prior to 'cabal-version: 3.8' if a wildcard '**' is used as a parent directory, the file's base name must be a wildcard '*'.

Cabal/src/Distribution/Simple/Glob.hs

@@ -96,7 +96,8 @@ explainGlobSyntaxError filepath NoExtensionOnStar =
   ++ "'. If a wildcard '*' is used it must be with an file extension."
 explainGlobSyntaxError filepath LiteralFileNameGlobStar =
      "invalid file glob '" ++ filepath
-  ++ "'. If a wildcard '**' is used as a parent directory, the"
+  ++ "'. Prior to 'cabal-version: 3.8'"
+  ++ " if a wildcard '**' is used as a parent directory, the"
   ++ " file's base name must be a wildcard '*'."
 explainGlobSyntaxError _ EmptyGlob =
      "invalid file glob. A glob cannot be the empty string."
@@ -111,7 +112,7 @@ explainGlobSyntaxError filepath VersionDoesNotSupportGlob =
   ++ "Alternatively if you require compatibility with earlier Cabal "
   ++ "versions then list all the files explicitly."
 
-data IsRecursive = Recursive | NonRecursive
+data IsRecursive = Recursive | NonRecursive deriving Eq
 
 data MultiDot = MultiDotDisabled | MultiDotEnabled
 
@@ -125,7 +126,7 @@ data GlobFinal
     -- ^ First argument: Is this a @**/*.ext@ pattern?
     --   Second argument: should we match against the exact extensions, or accept a suffix?
     --   Third argument: the extensions to accept.
-  | FinalLit FilePath
+  | FinalLit IsRecursive FilePath
     -- ^ Literal file name.
 
 reconstructGlob :: Glob -> FilePath
@@ -134,7 +135,8 @@ reconstructGlob (GlobStem dir glob) =
 reconstructGlob (GlobFinal final) = case final of
   FinalMatch Recursive _ exts -> "**" </> "*" <.> exts
   FinalMatch NonRecursive _ exts -> "*" <.> exts
-  FinalLit path -> path
+  FinalLit Recursive path -> "**" </> path
+  FinalLit NonRecursive path -> path
 
 -- | Returns 'Nothing' if the glob didn't match at all, or 'Just' the
 --   result if the glob matched (or would have matched with a higher
@@ -159,8 +161,8 @@ fileGlobMatchesSegments pat (seg : segs) = case pat of
       let (candidateBase, candidateExts) = splitExtensions seg
       guard (null segs && not (null candidateBase))
       checkExt multidot ext candidateExts
-    FinalLit filename -> do
-      guard (null segs && filename == seg)
+    FinalLit isRecursive filename -> do
+      guard ((isRecursive == Recursive || null segs) && filename == seg)
       return (GlobMatch ())
 
 checkExt
@@ -181,12 +183,14 @@ parseFileGlob version filepath = case reverse (splitDirectories filepath) of
         Left EmptyGlob
   (filename : "**" : segments)
     | allowGlobStar -> do
-        ext <- case splitExtensions filename of
+        finalSegment <- case splitExtensions filename of
           ("*", ext) | '*' `elem` ext -> Left StarInExtension
                      | null ext       -> Left NoExtensionOnStar
-                     | otherwise      -> Right ext
-          _                           -> Left LiteralFileNameGlobStar
-        foldM addStem (GlobFinal $ FinalMatch Recursive multidot ext) segments
+                     | otherwise      -> Right (FinalMatch Recursive multidot ext)
+          _                           -> if allowLiteralFilenameGlobStar
+                                           then Right (FinalLit Recursive filename)
+                                           else Left LiteralFileNameGlobStar
+        foldM addStem (GlobFinal finalSegment) segments
     | otherwise -> Left VersionDoesNotSupportGlobStar
   (filename : segments) -> do
         pat <- case splitExtensions filename of
@@ -196,7 +200,7 @@ parseFileGlob version filepath = case reverse (splitDirectories filepath) of
                      | otherwise           -> Right (FinalMatch NonRecursive multidot ext)
           (_, ext)   | '*' `elem` ext      -> Left StarInExtension
                      | '*' `elem` filename -> Left StarInFileName
-                     | otherwise           -> Right (FinalLit filename)
+                     | otherwise           -> Right (FinalLit NonRecursive filename)
         foldM addStem (GlobFinal pat) segments
   where
     allowGlob     = version >= CabalSpecV1_6
@@ -207,6 +211,7 @@ parseFileGlob version filepath = case reverse (splitDirectories filepath) of
     multidot
       | version >= CabalSpecV2_4 = MultiDotEnabled
       | otherwise                = MultiDotDisabled
+    allowLiteralFilenameGlobStar = version >= CabalSpecV3_8
 
 -- | This will 'die'' when the glob matches no files, or if the glob
 -- refers to a missing directory, or if the glob fails to parse.
@@ -300,7 +305,20 @@ runDirFileGlob verbosity rawDir pat = do
           return $ mapMaybe checkName candidates
         else
           return [ GlobMissingDirectory joinedPrefix ]
-    FinalLit fn -> do
+    FinalLit Recursive fn -> do
+      let prefix = dir </> joinedPrefix
+      directoryExists <- doesDirectoryExist prefix
+      if directoryExists
+        then do
+          candidates <- getDirectoryContentsRecursive prefix
+          let checkName candidate
+                  | takeFileName candidate == fn = Just $ GlobMatch (joinedPrefix </> candidate)
+                  | otherwise = Nothing
+          return $ mapMaybe checkName candidates
+        else
+          return [ GlobMissingDirectory joinedPrefix ]
+
+    FinalLit NonRecursive fn -> do
       exists <- doesFileExist (dir </> joinedPrefix </> fn)
       return [ GlobMatch (joinedPrefix </> fn) | exists ]
 

changelog.d/pr-8005

@@ -0,0 +1,9 @@
+synopsis: Allow glob-star matches with literal filenames (no extensions)
+packages: Cabal
+prs: #8005
+issues: #5883
+description: {
+
+- Cabal file glob syntax extended to allow matches of the form dir/**/FileNoExtension
+
+}

doc/cabal-package.rst

@@ -742,10 +742,13 @@ describe the package as a whole:
 
     - ``**`` wildcards can only appear as the final path component
       before the file name (e.g., ``data/**/images/*.jpg`` is not
-      allowed). If a ``**`` wildcard is used, then the file name must
-      include a ``*`` wildcard (e.g., ``data/**/README.rst`` is not
       allowed).
 
+    - Prior to Cabal 3.8, if a ``**`` wildcard is used, then
+      the file name must include a ``*`` wildcard (e.g.,
+      ``data/**/README.rst`` was not allowed). As of ``cabal-version:
+      3.8`` or greater, this restriction is lifted.
+
     - A wildcard that does not match any files is an error.
 
     The reason for providing only a very limited form of wildcard is to
@@ -2319,7 +2322,7 @@ system-dependent values for these fields.
        Cabal files with :pkg-field:`cabal-version` < 3.0 suffer from an
        infelicity in how the entries of :pkg-field:`mixins` are parsed: an
        entry will fail to parse if the provided renaming clause has whitespace
-       after the opening parenthesis. 
+       after the opening parenthesis.
 
        See issues :issue:`5150`, :issue:`4864`, and :issue:`5293`.
 

doc/file-format-changelog.rst

@@ -19,7 +19,7 @@ relative to the respective preceding *published* version.
     versions of the ``Cabal`` library denote unreleased development
     branches which have no stability guarantee.
 
-``cabal-version: 3.x``
+``cabal-version: 3.8``
 ----------------------
 
 * Added field ``code-generators`` to :pkg-section:`test-suite` stanzas. This
@@ -45,6 +45,11 @@ relative to the respective preceding *published* version.
   When :pkg-field:`extra-lib-dirs-static` is not given, it defaults to
   :pkg-field:`extra-lib-dirs`.
 
+* Wildcard matching has been slightly expanded. Matches are now
+  allowed of the form ``foo/**/literalFile``. Prior, double-star
+  wildcards required the trailing filename itself be a wildcard.
+
+
 ``cabal-version: 3.6``
 ----------------------