[pacman-dev] [PATCH] [WIP] Extend database upgrade script to handle alpm db version 9

[Andrew Gregory]

On 08/10/14 at 01:16pm, Allan McRae wrote:
> On 03/08/14 15:31, Andrew Gregory wrote:
> > Original-work-by: Allan McRae <allan at archlinux.org>
> > Signed-off-by: Andrew Gregory <andrew.gregory.8 at gmail.com>
> > ---
> > 
> 
> Big thanks for doing this!
> 
> > Changes:
> > * set alternate root with PACROOT environment variable
> 
> As you have pointed out, that needs to be a -r/--root flag and not an
> environmental variable
> 
> > * filter results through grep and sort to remove empty lines and duplicates
> > * add parent directories (duplicates filtered out by sort)
> > * verify that the resolved directory is inside our root
> > * replace longer symlinks first
> > * handle conflict between resolved directory and existing filelist entry
> > 
> >  scripts/pacman-db-upgrade.sh.in | 92 ++++++++++++++++++++++++++++++++++++-----
> >  1 file changed, 82 insertions(+), 10 deletions(-)
> > 
> > diff --git a/scripts/pacman-db-upgrade.sh.in b/scripts/pacman-db-upgrade.sh.in
> > index b0b0ac8..7ea4077 100644
> > --- a/scripts/pacman-db-upgrade.sh.in
> > +++ b/scripts/pacman-db-upgrade.sh.in

<snip>

> > +			# construct the parents of the new directory
> > +			parents=""
> > +			parent="$(dirname "$newdir")"
> 
> use bash:   parent=${newdir%/*}
> 
> > +			while [[ $parent != "." ]]; do
> 
> Should that be  != '/' ?
> 
> $ dirname /usr
> /
> 
> Anyway, using the bash above this becomes
> while [[ -n "$parent" ]]; do

This won't work.  $newdir has a trailing slash that would need to be
removed and it's relative to $pacroot, so ${parent%/*} would
eventually result in the base directory, not an empty string (which is
why dirname eventually gives '.').  If the symlink points to $pacroot,
$newdir could even be empty.  My attempts to account for all of that
in pure bash have been overly convoluted, but I'm open to suggestions.

> > +				parents+="$parent/\n"
> > +				parent="$(dirname "$parent")"
> > +			done
> > +
> 
> My awk is suboptimal...  So I'll comment my understanding as I go...
> 
> > +			for f in "$dbroot"/local/*/files; do
> > +				awk -v "olddir=$olddir" -v "newdir=$newdir" -v "parents=$parents" '
> 
> "parents=${parents[@]}"

$parents is actually a string rather than an array because I don't
think there's a way to pass an array to awk without mangling something.

> > +					BEGIN {
> > +						i = length(olddir) + 1
> > +						file = substr(newdir, 0, length(newdir) - 1)
> > +					}
> > +					{
> > +						if ($0 == olddir) {
> > +							printf("%s", parents)
> > +							printf("%s\n", newdir)
> 
> OK (replacing old dir with newdir and its parents)
> 
> > +						} else if ($0 == file) {
> > +							# skip
> 
> I do not understand what this bit is achieving!

If the newdir already exists in the file list as a file (i.e. somebody
replaced a symlink with a directory) this will prevent a duplicate
entry.  I'll expand that comment so it's clearer.

> > +						} else if (index($0, olddir) == 1) {
> > +							printf("%s%s\n", newdir, substr($0, i))
> 
> OK ("moving" file in olddir to "newdir")
> 
> > +						} else {
> > +							print
> 
> OK (not touching files outside of olddir)
> 
> > +						}
> > +					}' "$f" | grep . | LC_ALL=C sort -u > "$f.tmp"
> 
> Why the grep?

Removes blank lines in case $newdir has no parents or equals $pacroot.

> > +				mv "$f.tmp" "$f"
> > +			done
> > +		done
> > +	fi
> > +fi
> > +
> > +echo "9" > "$dbroot"/local/.alpm_db_version
> > +
> 
> OK
> 
> >  # remove the lock file
> >  rm -f "$lockfile"