xbps-install -Sy foo fails, performs unexpected action, exits zero

[roman-neuhauser]

when i run xbps-install -Sy foo, i expect xbps to either install foo or fail. instead, it does something else (updates libxbps/xbps) and claims success (exits zero) even though it clearly failed to do what was requested, not to mention it did something i didn't ask for instead.

the man page lacks any mention of this POLA violation, as well as recommendations regarding reliable unattended use.

[Duncaen]

The point of updating libxbps and xbps before doing any other updates is to allow to hotfix issues in case it is required.

This issue #142 describes the same problem which we are facing in xbps-src after every xbps update.

I think we shouldn't remove the mechanism, in case something goes very wrong we can fix issues before users will be affected by bugs and the last time we had to make use of this feature was less than a month ago.
https://voidlinux.org/news/2019/08/xbps-bug-results-in-missing-bin-sh.html

One of the last release tried to make the mechanism more bullet proof by also updating xbps itself alone when installing or updating another specific package.

One solution would be to reexecute xbps-install with the same arguments when it updated itself first, but there are some edge cases which are not easy to fix.

Another solution would be to just not allow any update and exit with ENOTSUP or any other error code until the user updates xbps, which would require unattended installations to check the exit code and then try to update xbps and then the original action. The problem with this is that the bad design forced people to do things like xbps-install -Su && xbps-install -Su in their scripts and in those cases we would break it. I'm not sure if this is a good point against it because in the end if we have a better designed solution everyone would benefit from getting rid of this issue.

In both solutions and the current implementation if xbps-install is used with a separate root directory the hotfix mechanism is useless because it never updates the actual xbps installation that does the installation.

I'm really not sure what the best way to go forward would be and how to tackle the different rootdir issues.

[Chocimier]

I'm really not sure (…) how to tackle the different rootdir issues.

Do update-in-memory analogous to memory-sync? More serious, I agree it would be better when `xbps-install bar` return error that update xbps and call it a day.

[roman-neuhauser]

After reading this thread I came to the conclusion that it is better to return an appropiate error code and a message explaining that xbps needs to be updated

What do you think?

i'm all for it! please make sure to use a dedicated exit code to make this particular condition easily detectable in scripts.

the remaining question is, what should happen on xbps-install -u. i haven't thought it through but it looks like the same reaction would work.