On Fri, Aug 21, 2020 at 10:01:50PM +0200, Manuel Reimer wrote:

There have to be edge cases where the replacement does not happen even if I never touched a file myself. But I don't know how to debug this.

No, you are mistaken. There are two issues, which is quite apparent if you do read the bugreports. 1) People modifying /etc/pam.d files and promptly forgetting it. Breaking their system as the new file is inserted as `/etc/pam.d/system-login.pacnew`. 2) Some change introduced into the new /etc/pam.d/system-login where auxillary files such as ~/.pam_environment can sometimes break your system. I have no clue how the latter works.

There are more ".pacnew" in my /etc. All for files that I never touched myself.

That is concerning, but it is simply something touching those files you are unaware of. It's however you system and it's expected that you are aware of what does this.

I guess there is something wrong with the way how pacman decides when to move ".pacnew" into place directly.

There isn't. Peopel with `.pacnew` has modified their files, wittingly or not. -- Morten Linderud PGP: 9C02FF419FECBE16

On 8/19/20 2:13 PM, Giancarlo Razzolini via arch-general wrote:

Em agosto 19, 2020 16:02 Manuel Reimer escreveu:

...

Hello,

Some minutes ago I did a regular system update and after that decided to reboot. After reboot I was unable to log into my system. After fiddling a bit I rebooted to an Arch boot stick to find the following message in pacman.log:

[2020-08-19T20:42:55+0200] [ALPM] warning: /etc/pam.d/system-login installed as /etc/pam.d/system-login.pacnew

The .pacnew should've been handled *before* rebooting.

...

As this seemed to be a candidate that may cause login problems, I deleted "system-login" and moved the ".pacnew" into place.

After reboot I'm now able to log in again...

IMHO something like this should not happen...

Maybe it's worth a note on the Arch homepage that it is important to move this pacnew into place before reboot?

This only affected you and whomever else changed system-login. It's not news material. Also, if you're messing with PAM, you should be responsible for applying the new stuff, otherwise it'll break, like it did for you.

Regards, Giancarlo Razzolini

I've always questioned the wisdom of dropping a .pacnew just when the file is different from the default. There's really no reason for it considering any changes you made were deliberate and presumably thought out. The end result is pacman cluttering /etc with a default configuration file whose only reason for existing is to, if it's used, clear settings. Why? What pacman SHOULD do is compare /etc files between package versions and see if there's a change BETWEEN DEFAULTS. *Then* there's an actual reason to need a new default config file for the user to examine because then there's an actual indicator some meaningful change in default configuration or how the package handles configs happened. All most pacnew files wind up doing is sitting there for thirty seconds before being deleted without anyone even opening them because they're literally just what the file was before the user ALREADY changed it before... because it's utterly useless to get a default config file when you've intentionally changed it and there's nothing in the new version of the package that calls for an examination of the defaults. Yaro

On 8/19/20 2:48 PM, Giancarlo Razzolini via arch-general wrote:

Em agosto 19, 2020 16:37 Yaro Kasear escreveu:

...

I've always questioned the wisdom of dropping a .pacnew just when the file is different from the default. There's really no reason for it considering any changes you made were deliberate and presumably thought out. The end result is pacman cluttering /etc with a default configuration file whose only reason for existing is to, if it's used, clear settings. Why?

The .pacnew is there to indicate that something new exists, or that you changed something. Most of the time you can remove .pacnew files, but not always. Also, it's only "cluttering" /etc (and /boot, btw), if you don't handle them.

...

What pacman SHOULD do is compare /etc files between package versions and see if there's a change BETWEEN DEFAULTS. *Then* there's an actual reason to need a new default config file for the user to examine because then there's an actual indicator some meaningful change in default configuration or how the package handles configs happened.

That's way beyond the scope of a package manager, and also, there's no way to tell what "DEFAULTS" (why caps?) should be. Caps for emphasis is all.

...

All most pacnew files wind up doing is sitting there for thirty seconds before being deleted without anyone even opening them because they're literally just what the file was before the user ALREADY changed it before... because it's utterly useless to get a default config file when you've intentionally changed it and there's nothing in the new version of the package that calls for an examination of the defaults.

I don't know why you said that .pacnew sits for thirty seconds before being deleted. Are you using a hook that does this? Because nothing handles them automatically, that's the user's job. There are tools to aid in doing that, but in the end the user should know what to apply, and what to discard. I wasn't being literal about thirty seconds. Exaggerating.

Regards, Giancarlo Razzolini

Yaro

Em agosto 19, 2020 17:04 Yaro Kasear escreveu:

Yes there is. The defaults are literally what's in the config file in the archive and not on the filesystem. How would that not be a way to determine default settings?

I'm not suggesting the package manager would have to understand the settings, but it would be able to tell if the contents of that file are different from another version. (Which it obviously does already, otherwise it wouldn't know to make a pacnew file.)

I can't imagine it'd be that difficult for pacman to compare checksums between files in /etc or /boot between versions of a package (If a previous version is available.) and what's on /etc and determine if it really needs to bother putting a pacnew file on the filesystem that doesn't need to be there. It's already doing some sort of check between what's in the package and what's on the filesystem already.

How is everything you just said, different than what pacman already does? How would it determine not to create a .pacnew? If you can answer both these questions, I'd encourage you to send patches to pacman. Because I couldn't understand how what you said is any different than the current pacnew logic. Regards, Giancarlo Razzolini

Le 20/08/2020 à 10:08, Kusoneko a écrit :

On August 19, 2020 8:16:08 PM UTC, Giancarlo Razzolini via arch-general wrote:

...

Em agosto 19, 2020 17:04 Yaro Kasear escreveu:

...

Yes there is. The defaults are literally what's in the config file in the archive and not on the filesystem. How would that not be a way to determine default settings?

I'm not suggesting the package manager would have to understand the settings, but it would be able to tell if the contents of that file are different from another version. (Which it obviously does already, otherwise it wouldn't know to make a pacnew file.)

I can't imagine it'd be that difficult for pacman to compare checksums between files in /etc or /boot between versions of a package (If a previous version is available.) and what's on /etc and determine if it really needs to bother putting a pacnew file on the filesystem that doesn't need to be there. It's already doing some sort of check between what's in the package and what's on the filesystem already.

How is everything you just said, different than what pacman already does? How would it determine not to create a .pacnew? If you can answer both these questions, I'd encourage you to send patches to pacman. Because I couldn't understand how what you said is any different than the current pacnew logic.

Regards, Giancarlo Razzolini I think he's trying to imply that pacman stores a copy of the archive containing the previous version somewhere and that pacman should extract the config files from both and see if something changed before providing a .pacnew. Only thing is, that would cost much more storage space than if one lazily ignored and let the pacnew files be wherever they're placed. In other words, not really a good idea unless you have tons of storage space. I guess you could theoretically patch pacman to do so for yourself if you really wanted to, but for most people it wouldn't be worth it.

Please read the remaining of the thread you’re replying to before posting. ;) This has already been addressed, and pacman actually does this (using hashes as to not waste storage space). Bruno/Archange

I don't see what all the fuss is about

If you're using Arch, then you should:

1. Check Arch news before running update 2. Update regularly 3. Watch output from pacman for warnings/advice 4. Run pacdiff after update and before reboot

After step 4 there are no longer any pacnew files

That would have shown that the new file doesn't have pam_tally2, it wont say you wouldn't be able to login after reboot. actually, even assuming you would know to fix the issue, but you didn't fix the problem immediately and you went to make a coffee - your screen saver was activated, you are locked out. Now I don't understand all the defensiveness - let's all work together to improve things. This is not a non-issue. -- damjan

Le 20/08/2020 à 00:04, Yaro Kasear a écrit :

On 8/19/20 2:56 PM, Yaro Kasear wrote:

...

On 8/19/20 2:48 PM, Giancarlo Razzolini via arch-general wrote:

...

Em agosto 19, 2020 16:37 Yaro Kasear escreveu:

...

I've always questioned the wisdom of dropping a .pacnew just when the file is different from the default. There's really no reason for it considering any changes you made were deliberate and presumably thought out. The end result is pacman cluttering /etc with a default configuration file whose only reason for existing is to, if it's used, clear settings. Why?

The .pacnew is there to indicate that something new exists, or that you changed something. Most of the time you can remove .pacnew files, but not always. Also, it's only "cluttering" /etc (and /boot, btw), if you don't handle them.

...

What pacman SHOULD do is compare /etc files between package versions and see if there's a change BETWEEN DEFAULTS. *Then* there's an actual reason to need a new default config file for the user to examine because then there's an actual indicator some meaningful change in default configuration or how the package handles configs happened.

That's way beyond the scope of a package manager, and also, there's no way to tell what "DEFAULTS" (why caps?) should be. Caps for emphasis is all.

...

All most pacnew files wind up doing is sitting there for thirty seconds before being deleted without anyone even opening them because they're literally just what the file was before the user ALREADY changed it before... because it's utterly useless to get a default config file when you've intentionally changed it and there's nothing in the new version of the package that calls for an examination of the defaults.

I don't know why you said that .pacnew sits for thirty seconds before being deleted. Are you using a hook that does this? Because nothing handles them automatically, that's the user's job. There are tools to aid in doing that, but in the end the user should know what to apply, and what to discard. I wasn't being literal about thirty seconds. Exaggerating. Regards, Giancarlo Razzolini Yaro

Oh, also:

"That's way beyond the scope of a package manager, and also, there's no way to tell what "DEFAULTS" (why caps?) should be."

Yes there is. The defaults are literally what's in the config file in the archive and not on the filesystem. How would that not be a way to determine default settings?

I'm not suggesting the package manager would have to understand the settings, but it would be able to tell if the contents of that file are different from another version. (Which it obviously does already, otherwise it wouldn't know to make a pacnew file.)

I can't imagine it'd be that difficult for pacman to compare checksums between files in /etc or /boot between versions of a package (If a previous version is available.) and what's on /etc and determine if it really needs to bother putting a pacnew file on the filesystem that doesn't need to be there. It's already doing some sort of check between what's in the package and what's on the filesystem already.

Yaro

pacman does this: if the *packaged file* changed between the installed version and the new one, and the *installed file* is different from the *packaged file*, then drop a .pacnew. I’m not sure what you want more… Bruno/Archange

On Wed, Aug 19, 2020 at 3:32 PM Yaro Kasear wrote:

On Wed, Aug 19, 2020 at 3:17 PM Archange wrote:

...

...

On 8/19/20 2:56 PM, Yaro Kasear wrote:

...

On 8/19/20 2:48 PM, Giancarlo Razzolini via arch-general wrote:

...

Em agosto 19, 2020 16:37 Yaro Kasear escreveu:

...

I've always questioned the wisdom of dropping a .pacnew just when the file is different from the default. There's really no reason for it considering any changes you made were deliberate and presumably

...

...

...

...

out. The end result is pacman cluttering /etc with a default configuration file whose only reason for existing is to, if it's used, clear settings. Why?

The .pacnew is there to indicate that something new exists, or that you changed something. Most of the time you can remove .pacnew files, but not always. Also, it's only "cluttering" /etc (and /boot, btw), if you don't handle

Le 20/08/2020 à 00:04, Yaro Kasear a écrit : thought them.

...

...

...

...

What pacman SHOULD do is compare /etc files between package versions

and

...

...

see if there's a change BETWEEN DEFAULTS. *Then* there's an actual reason to need a new default config file for the user to examine because then there's an actual indicator some meaningful change in default configuration or how the package handles configs happened.

That's way beyond the scope of a package manager, and also, there's no way to tell what "DEFAULTS" (why caps?) should be. Caps for emphasis is all.

...

All most pacnew files wind up doing is sitting there for thirty seconds before being deleted without anyone even opening them because they're literally just what the file was before the user ALREADY changed it before... because it's utterly useless to get a default config file when you've intentionally changed it and there's nothing in the new version of the package that calls for an examination of the defaults.

I don't know why you said that .pacnew sits for thirty seconds before being deleted. Are you using a hook that does this? Because nothing handles them automatically, that's the user's job. There are tools to aid in doing that, but in the end the user should know what to apply, and what to discard. I wasn't being literal about thirty seconds. Exaggerating. Regards, Giancarlo Razzolini Yaro

Oh, also:

"That's way beyond the scope of a package manager, and also, there's no way to tell what "DEFAULTS" (why caps?) should be."

Yes there is. The defaults are literally what's in the config file in the archive and not on the filesystem. How would that not be a way to determine default settings?

I'm not suggesting the package manager would have to understand the settings, but it would be able to tell if the contents of that file are different from another version. (Which it obviously does already, otherwise it wouldn't know to make a pacnew file.)

I can't imagine it'd be that difficult for pacman to compare checksums between files in /etc or /boot between versions of a package (If a previous version is available.) and what's on /etc and determine if it really needs to bother putting a pacnew file on the filesystem that doesn't need to be there. It's already doing some sort of check between what's in the package and what's on the filesystem already.

Yaro

pacman does this: if the *packaged file* changed between the installed version and the new one, and the *installed file* is different from the *packaged file*, then drop a .pacnew.

I’m not sure what you want more…

Bruno/Archange

But that is not what I am talking about.

I'm discussing what is essentially three configuration files here:

- The config in old-package. - The config in new-package. - The config on the filesystem.

I already know pacman compares new-package and filesystem config. It does NOT do any check between old-package and new-package.

What I'm saying is a pacnew seems unnecessary if the file between old-package and new-package are the same, because it means there's absolutely nothing of use to the user there, as either the user's made changes themselves and thus that file from the package is just going to be settings they don't want, or it's never been changed, in which case extracting the file's entirely redundant as it's literally still the same file. Pacman wouldn't do a pacnew in that case as it's implemented so that's beside the point.

I'm not talking about the existing new-package to filesystem version checks, I'm talking about a case where if a package upgrade doesn't bring a new configuration change, the pacnew's just a waste of time and space (Albeit tiny amounts of space.) for the user.

I'm suggesting pacman check not just between installed and new-package versions of the config but, if the old-package is still in the cache, take a moment to see if the old-package and new-package version of the config file are different, to stop pacman from dropping a file that will literally just need to be deleted.

A pacnew is only really useful when the config file changes between versions *because* a change in the file between old-package and new-package could be because of some upstream change that the user might need to address. But when it's just the same default configuration they already changed from, a pacnew's pointless.

Unless I missed something, pacman drops a pacnew regardless if there's a difference between new-package and what's on the filesystem and doesn't care if the pacnew is really needed.

Yaro

Correcting myself on a badly worded sentence. "Unless I missed something, pacman drops a pacnew regardless if there's a difference between new-package and what's on the filesystem and doesn't care if the pacnew is really needed." What I meant to say was, "It will drop a pacnew regardless of differences between packages if it sees a difference between what's in the new package and what's installed regardless of if the file is needed." Sorry about that. Yaro

On 8/19/20 3:41 PM, Morten Linderud via arch-general wrote:

https://www.archlinux.org/pacman/pacman.8.html#_handling_config_files_a_id_h...

Can you please read this section a few times before writing more emails?

I see. I was sure pacman made no comparisons at all to previous config versions, just what is on the filesystem now and what's new. Sorry about that, I was wrong. Yaro