Re: [PATCH v3] dpkg-base: resolve DL_DIR in do_adjust_git

[Jan Kiszka <jan.kiszka@siemens.com>]

On 04.02.22 11:54, Cedric Hombourger wrote:
> From: Cedric Hombourger <cedric.hombourger@siemens.com>
> 
> git_link is resolved using os.path.realpath() but git_dl is not.
> If DL_DIR points to a symbolic link, the comparison will always
> fail and do_adjust_git() will attempt to re-create the symbolic
> link. Resolve DL_DIR for a comparison between resolved paths.
> In the event where paths do differ, the symbolic link needs to
> be deleted first.
> 
> Signed-off-by: Cedric Hombourger <cedric.hombourger@siemens.com>
> ---
>  meta/classes/dpkg-base.bbclass | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/meta/classes/dpkg-base.bbclass b/meta/classes/dpkg-base.bbclass
> index 2add0b2..258e040 100644
> --- a/meta/classes/dpkg-base.bbclass
> +++ b/meta/classes/dpkg-base.bbclass
> @@ -20,9 +20,13 @@ python do_adjust_git() {
>      rootdir = d.getVar('WORKDIR', True)
>  
>      git_link = os.path.join(d.getVar('GIT_DL_LINK_DIR'), '.git-downloads')
> -    git_dl = os.path.join(d.getVar("DL_DIR"), "git")
> +    dl_dir = os.path.realpath(d.getVar("DL_DIR"))
> +    git_dl = os.path.join(dl_dir, "git")
>  
> -    if not os.path.exists(git_link) or os.path.realpath(git_link) != git_dl:
> +    if os.path.exists(git_link) and os.path.realpath(git_link) != git_dl:
> +        os.unlink(git_link)
> +
> +    if not os.path.exists(git_link):
>          os.symlink(git_dl, git_link)
>  
>      for src_uri in (d.getVar("SRC_URI", True) or "").split():
> @@ -34,7 +38,7 @@ python do_adjust_git() {
>  
>              if os.path.islink(ud.localpath):
>                  realpath = os.path.realpath(ud.localpath)
> -                filter_out = os.path.join(d.getVar("DL_DIR"), "git") + "/"
> +                filter_out = git_dl + "/"
>                  if realpath.startswith(filter_out):
>                      # make the link relative
>                      link = realpath.replace(filter_out, '', 1)

Looks good and was surely an untested case so far. It's actually two
fixes in one (os.symlink != ln -sf...)

Does this explain your "race condition" from before?

Jan

-- 
Siemens AG, Technology
Competence Center Embedded Linux