doinst.sh improvements suggestion

opty · 03-26-2021, 05:56 PM

doinst.sh improvements suggestion:

Code:

config() {
  NEW="$1"
  OLD="${NEW%.new}"
  # If there's no config file by that name, mv it over:
  if [ ! -r "$OLD" ]; then
    mv "$NEW" "$OLD"
  elif [ "$(md5sum < "$OLD")" = "$(md5sum < "$NEW")" ]; then
    # toss the redundant copy
    rm "$NEW"
  fi
  # Otherwise, we leave the .new copy for the admin to consider...
}

"for infile in $1; do [...] done" variant seems pointless to me but I may be missing something. "md5sum < file" already used but not everywhere.

Code:

preserve_perms() {
  NEW="$1"
  OLD="${NEW%.new}"
  if [ -e "$OLD" ]; then
    for cmd in ch{own,mod}; do
      $cmd --reference="$OLD" "$NEW"
    done
  fi
  config "$NEW"
}

Maybe it should do more than just chown and chmod? (Due to "cp -a ${OLD} ${NEW}.incoming" and I saw even "touch -r ${NEW} ${NEW}.incoming".)

Some ideas by lopid from ##slackware-help at Freenode.

crts · 03-27-2021, 10:15 AM

Quote:

Originally Posted by opty

Code:

config() {
  NEW="$1"
  OLD="${NEW%.new}"
  # If there's no config file by that name, mv it over:
  if [ ! -r "$OLD" ]; then
    mv "$NEW" "$OLD"
  elif [ "$(md5sum < "$OLD")" = "$(md5sum < "$NEW")" ]; then
    # toss the redundant copy
    rm "$NEW"
  fi
  # Otherwise, we leave the .new copy for the admin to consider...
}

... "md5sum < file" already used but not everywhere.

Hi,

I do not see the point of using md5sum here. For once, there is the chance of hash collisions. Even if the chances are low, why risk it? A diff -q would not only eliminate that risk, it is also faster - not that performance matters much with files this size, you only notice the difference with files >100MB. But still, why introduce the computational overhead of a checksum comparison to get a worse result?

Clearly I, too, am missing something here. So if you know of a reason why md5sum is preferred over diff then please let me know.

opty · 03-27-2021, 10:30 AM

I took md5sum as an axiom but sure, why not to use diff?

Wait, not so sure, md5sum comes with coreutils and you need diffutils for diff. Also, later we could need that MD5 stored and use for improved *.new handling.

GazL · 03-27-2021, 11:55 AM

diff is not the correct utility. If you must change it from md5sum, then I'd suggest cmp.

crts · 03-27-2021, 12:32 PM

Quote:

Originally Posted by opty

Wait, not so sure, md5sum comes with coreutils and you need diffutils for diff.

Good point, I assume coreutils is installed before diffutils, so it makes sense to have it this way during system installation.

Quote:

Originally Posted by opty

Also, later we could need that MD5 stored and use for improved *.new handling.

I have a hunch that storing MD5 sums for later comparisons would introduce too big a complication. Just think about the possibility of stale files. As I mentioned earlier, with the file sizes we are dealing here speed is not much of an issue.

Quote:

Originally Posted by GazL

diff is not the correct utility. If you must change it from md5sum, then I'd suggest cmp.

Why? Cmp is in the diffutils package, just as diff.

opty · 03-27-2021, 01:38 PM

Ad MD5: e.g. slack-install by tadgy

tadgy · 03-27-2021, 06:27 PM

It should be noted that my slack-install doinst.sh will no longer work with -current - Pat now uses a cookie in the extracted /install/ path in order to allow multiple package installations at the same time (presumably, I can't think of other reasons to make that path random).

Also, the slack-install stuff was never meant for wide package use - it is used for my own custom packages because I can do the necessary md5suming on a small number Vs a huge number in the distro... it would prove to be far too much effort to maintain. opty and I have already discussed an alternative that would work automatically and still provide the same feature (ie, not leaving .new files around if the admin has no edited the file) - I just need to code it up sometime.

chrisretusn · 03-27-2021, 08:48 PM

Quote:

Originally Posted by crts

Why? Cmp is in the diffutils package, just as diff.

Perhaps this:

Quote:

GNU diff - compare files line by line

Quote:

GNU cmp - compare two files byte by byte

opty · 03-28-2021, 08:25 AM

Maybe due to config() used for binaries too? (sudo in the middle warning!):

Code:

sed -n '/\.new$/{ s/\(.\+\)\.new$/\/\1/; p; }' /var/lib/pkgtools/packages/* |& sudo file -f - |& egrep -v "text|empty"

Most probably the reason for using md5sum.

allend · 03-28-2021, 08:27 AM

A feature of Slackware is that the install media can also be used as rescue media, so you also need to consider compatibility with busybox.
If you use md5sum, then the output is a string variable that is the same in GNU and busybox.
If you use 'cmp -s', then the exit values are the same in GNU and busybox, although the outputs differ if -s is not specified and files are different.

opty · 03-28-2021, 08:50 AM

BTW, maybe the migration from md5sum to e.g. b2sum (BLAKE2) or even b3sum (BLAKE3) still doesn't yield the cost-benefit ratio worth doing it.