From b911d6547444b5f8d17b224bafa5ee1b5aafaff5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ludovic=20Court=C3=A8s?= <ludo@gnu.org>
Date: Tue, 15 Sep 2020 14:24:05 +0200
Subject: [PATCH] authenticate: Encode strings as ISO-8859-1.

Fixes <https://bugs.gnu.org/43421>.

* guix/scripts/authenticate.scm (read-command): Decode strings as
ISO-8859-1, not UTF-8.
(guix-authenticate)[send-reply]: Encode strings as ISO-8859-1, not
UTF-8.
* tests/guix-authenticate.sh: Add test.
---
 guix/scripts/authenticate.scm | 8 +++++---
 tests/guix-authenticate.sh    | 9 +++++++++
 2 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/guix/scripts/authenticate.scm b/guix/scripts/authenticate.scm
index 0bac13edee..45f62f6ebc 100644
--- a/guix/scripts/authenticate.scm
+++ b/guix/scripts/authenticate.scm
@@ -31,6 +31,7 @@
   #:use-module (ice-9 rdelim)
   #:use-module (ice-9 match)
   #:use-module (ice-9 vlist)
+  #:use-module (ice-9 iconv)
   #:export (guix-authenticate))
 
 ;;; Commentary:
@@ -122,8 +123,9 @@ by colon, followed by the given number of characters."
                 (reverse result))
                (else
                 (let* ((len (string->number (read-delimited ":" port)))
-                       (str (utf8->string
-                             (get-bytevector-n port len))))
+                       (str (bytevector->string
+                             (get-bytevector-n port len)
+                             "ISO-8859-1" 'error)))
                   (loop (cons str result))))))))))
 
 (define-syntax define-enumerate-type              ;TODO: factorize
@@ -150,7 +152,7 @@ by colon, followed by the given number of characters."
 
   (define (send-reply code str)
     ;; Send CODE and STR as a reply to our client.
-    (let ((bv (string->utf8 str)))
+    (let ((bv (string->bytevector str "ISO-8859-1" 'error)))
       (format #t "~a ~a:" code (bytevector-length bv))
       (put-bytevector (current-output-port) bv)
       (force-output (current-output-port))))
diff --git a/tests/guix-authenticate.sh b/tests/guix-authenticate.sh
index f3b36ee41d..3a05b232c1 100644
--- a/tests/guix-authenticate.sh
+++ b/tests/guix-authenticate.sh
@@ -61,6 +61,15 @@ sed -i "$sig"											\
 code="$(echo "verify $(cat $sig)" | guix authenticate | cut -f1 -d ' ')"
 test "$code" -ne 0
 
+# Make sure byte strings are correctly encoded.  The hash string below is
+# "café" repeated 8 times.  Libgcrypt would normally choose to write it as a
+# string rather than a hex sequence.  We want that string to be Latin-1
+# encoded independently of the current locale: <https://bugs.gnu.org/43421>.
+hash="636166e9636166e9636166e9636166e9636166e9636166e9636166e9636166e9"
+latin1_cafe="caf$(printf '\351')"
+echo "sign 21:tests/signing-key.sec 64:$hash" | guix authenticate \
+    | LC_ALL=C grep "hash sha256 \"$latin1_cafe"
+
 # Test for <http://bugs.gnu.org/17312>: make sure 'guix authenticate' produces
 # valid signatures when run in the C locale.
 hash="5eff0b55c9c5f5e87b4e34cd60a2d5654ca1eb78c7b3c67c3179fed1cff07b4c"