bug#44741: ‘copy-recursively’ #:keep-mtime? #t broken for directories

bug#44741: ‘copy-recursively’ #:keep-mtime? #t broken for directories

[Ludovic Courtès]

When passing #:keep-mtime? #t to ‘copy-recursively’, the mtime of
regular files is preserved, but the mtime of directories (and symlinks)
is not.

This is because ‘set-file-time’ is called in the ‘down’ procedure of
‘file-system-fold’, before the directory is populated.

‘populate-store’ is affected by this bug, meaning that it creates stores
where directories have “wrong” timestamps.  In practice this is probably
less likely to have undesirable side effects than regular files with
wrong timestamps (think ‘make’, .go files, etc.)

Ludo’.

bug#44741: ‘copy-recursively’ #:keep-mtime? #t broken for directories

[Ludovic Courtès]

Ludovic Courtès <ludo@gnu.org> skribis:

> When passing #:keep-mtime? #t to ‘copy-recursively’, the mtime of
> regular files is preserved, but the mtime of directories (and symlinks)
> is not.
>
> This is because ‘set-file-time’ is called in the ‘down’ procedure of
> ‘file-system-fold’, before the directory is populated.

Fixed by 201cbcac3e9c23cc28352fa9ece3e677e22cc12c (core-updates).

> ‘populate-store’ is affected by this bug, meaning that it creates stores
> where directories have “wrong” timestamps.  In practice this is probably
> less likely to have undesirable side effects than regular files with
> wrong timestamps (think ‘make’, .go files, etc.)

In some cases, such as the initrd, the problem is harmless because
timestamps are reset anyway in the archive (for instance
‘write-cpio-archive’ does that.)

There might be problems in other cases, such as (gnu build image).

Ludo’.