diff --git a/shell-completion/hledger-completion.bash.m4 b/shell-completion/hledger-completion.bash.m4
index 7a261489b..c77b88cda 100644
--- a/shell-completion/hledger-completion.bash.m4
+++ b/shell-completion/hledger-completion.bash.m4
@@ -9,6 +9,14 @@ set -o pipefail
 # contains regex special chars. But it might be no problem because of
 # COMP_WORDBREAKS.
 
+# TODO compgen and compopt is pretty complicated. Piping to
+# grep "^$wordToComplete"
+# seems like a hack - I'd rather use
+# compgen ... -- "$wordToComplete"
+# But what options to use? I don't want to use -W because it may exceed the
+# maximum command line length. -C "cat file" is not working either. It would be
+# best if compgen can read from stdin but it does not.
+
 # Working with bash arrays is nasty compared to editing a text file. Consider
 # for example grepping an array or mapping a substitution on it.
 # Therefore, we create temp files in RAM for completion suggestions (see below).
@@ -49,7 +57,9 @@ _hledger_completion_function() {
 	    # This does not work because assignment to 'files' in the "pipe
 	    # subshell" has no effect!
 	    #compgen -df | grep "^$filenameSoFar" | readarray -t files
-	    readarray -t files < <(compgen -df | grep "^$filenameSoFar")
+
+	    compopt -o filenames -o dirnames
+	    readarray -t files < <(compgen -f -- "$filenameSoFar")
 	    COMPREPLY=( "${files[@]}" )
 
 	else