bug#60202: tests/cpio failure

["Ludovic Courtès" <ludo@gnu.org>]

[-- Attachment #1: Type: text/plain, Size: 2539 bytes --]

Hi,

Christopher Baines <mail@cbaines.net> skribis:

> This test seems to fail, maybe because of high inode numbers, maybe
> something to with btrfs.

Uh.

> scheme@(guix-user)> (file->cpio-header "guix/guix.scm")
> $1 = #<<cpio-header> magic: 460545 ino: 5031515288 mode: 33188 uid: 1003 gid: 998 nlink: 1 mtime: 1671460627 file-size: 1452 dev-maj: 0 dev-min: 24 rdev-maj: 0 rdev-min: 0 name-size: 14 checksum: 0>

[...]

>         (let ((port (open-bytevector-input-port (get-bv))))
>           (equal? (peek "A" header)
>                   (peek "B" (read-cpio-header port))))))
>
> ;;; ("A" #<<cpio-header> magic: 460545 ino: 5031515288 mode: 33188 uid: 1003 gid: 998 nlink: 1 mtime: 1671460627 file-size: 1452 dev-maj: 0 dev-min: 24 rdev-maj: 0 rdev-min: 0 name-size: 14 checksum: 0>)
>
> ;;; ("B" #<<cpio-header> magic: 460545 ino: 736547992 mode: 33188 uid: 1003 gid: 998 nlink: 1 mtime: 1671460627 file-size: 1452 dev-maj: 0 dev-min: 24 rdev-maj: 0 rdev-min: 0 name-size: 14 checksum: 0>)
> $3 = #f

(guix cpio) defines cpio headers like this:

--8<---------------cut here---------------start------------->8---
(define-pack <cpio-header>
  %make-cpio-header cpio-header?
  write-cpio-header read-cpio-header
  (magic     6  cpio-header-magic)
  (ino       8  cpio-header-inode)
  (mode      8  cpio-header-mode)
  (uid       8  cpio-header-uid)
  (gid       8  cpio-header-gid)
  (nlink     8  cpio-header-nlink)
  (mtime     8  cpio-header-mtime)
  (file-size 8  cpio-header-file-size)
  (dev-maj   8  cpio-header-device-major)
  (dev-min   8  cpio-header-device-minor)
  (rdev-maj  8  cpio-header-rdevice-major)
  (rdev-min  8  cpio-header-rdevice-minor)
  (name-size 8  cpio-header-name-size)
  (checksum  8  cpio-header-checksum))            ;0 for "newc" format
--8<---------------cut here---------------end--------------->8---

The ‘ino’ value is stored as an 8-digit hexadecimal number—i.e., on 32
bits.  And, guess what:

--8<---------------cut here---------------start------------->8---
scheme@(guile-user)> 5031515288  ;the inode number above
$15 = 5031515288
scheme@(guile-user)> (expt 2 32)
$16 = 4294967296
scheme@(guile-user)> (- $16 $15)
$17 = -736547992
--8<---------------cut here---------------end--------------->8---

Noways, libc + kernel typically use a 64-bit ‘ino_t’, so fundamentally
the cpio format is “wrong”.  But what can we do?

At least we can skip this test when that is the case (patch below).
WDYT?

Thanks,
Ludo’.


[-- Attachment #2: Type: text/x-patch, Size: 1056 bytes --]

diff --git a/tests/cpio.scm b/tests/cpio.scm
index 516de0655b..832101d1bb 100644
--- a/tests/cpio.scm
+++ b/tests/cpio.scm
@@ -1,5 +1,5 @@
 ;;; GNU Guix --- Functional package management for GNU
-;;; Copyright © 2015 Ludovic Courtès <ludo@gnu.org>
+;;; Copyright © 2015, 2022 Ludovic Courtès <ludo@gnu.org>
 ;;;
 ;;; This file is part of GNU Guix.
 ;;;
@@ -31,12 +31,18 @@ (define-module (test-cpio)
 (define %cpio-program
   (which "cpio"))
 
+(define %test-file
+  (search-path %load-path "guix.scm"))
+
 \f
 (test-begin "cpio")
 
+;; The cpio format expects 'ino' to fit in 32 bits.  If we have a bigger inode
+;; number, skip this test.
+(test-skip
+ (if (>= (stat:ino (lstat %test-file)) (expt 2 32)) 1 0))
 (test-assert "file->cpio-header + write-cpio-header + read-cpio-header"
-  (let* ((file   (search-path %load-path "guix.scm"))
-         (header (file->cpio-header file)))
+  (let* ((header (file->cpio-header %test-file)))
     (call-with-values
         (lambda ()
           (open-bytevector-output-port))