Bug 342045 – sys-apps/portage: emerge checks too many files for package collisions 'nodoc/noinfo' FEATURE is enabled

• Status: CONFIRMED
• Product: Portage Development

[Description Marc Burkhardt 2010-10-21 16:54:18 UTC]

- emerge an ebuild using emerge with FEATURES += nodoc enabled
- emerge checks file collisions
- not N but only M files should have been checked due to the files will not be installed anyways

Example (iftop as just one case of actually any package installing something into /usr/share/doc) - the below example has 2 files to be installed, not 5 that have been checked

 * checking 5 files for package collisions
 * Removing /usr/share/info
 * Removing /usr/share/doc
--- /usr/
--- /usr/share/
--- /usr/share/man/
--- /usr/share/man/man8/
>>> /usr/share/man/man8/iftop.8.bz2
--- /usr/sbin/
>>> /usr/sbin/iftop

[Comment 1 Marc Burkhardt 2010-10-21 16:57:16 UTC]

The same actually aplies for the 'noinfo' FEATURE

[Comment 2 Zac Medico 2010-10-27 18:45:36 UTC]

The collision check is before pkg_preinst, and the no{doc,info,man} code runs after pkg_preinst. So, the count displayed in the message is accurate at the time that it's run. If that's not good enough then I suppose we could just remove the count entirely since it's non-essential.

[Comment 3 Marc Burkhardt 2010-10-27 18:49:48 UTC]

Yes, that's correct. What I meant to say is:

why do we check the 5 files when only 2 of the files will be installed? It would be enough to check the remaining 2 files _after_ the nodoc/noinfo stuff is run. So we could move the collision check after the nodoc/noinfo stuff and anything is great...

[Comment 4 Zac Medico 2010-10-27 18:59:45 UTC]

The collision check has to go before pkg_preinst, since pkg_preinst is not supposed to execute when we bail out due to collisions.

I suppose the no{doc,info,man} code could run before pkg_preinst too.

[Comment 5 Zac Medico 2010-10-27 19:18:49 UTC]

(In reply to comment #4)
> I suppose the no{doc,info,man} code could run before pkg_preinst too.

However, this makes it easier for ebuild developers to commit ebuilds which have file collisions that are hidden by local features settings. So, I think it's better to simply omit the file count from the collision checking message.

[Comment 6 Marc Burkhardt 2010-10-27 19:38:22 UTC]

(In reply to comment #4)
> I suppose the no{doc,info,man} code could run before pkg_preinst too.
> 

That's great! :)

What we have then is:

 * Removing /usr/share/info
 * Removing /usr/share/doc
 * Checking 2 files for package collisions

or even cooler:

 * Removing /usr/share/info
 * Removing /usr/share/doc
 * Checking 2 of initially 5 files for package collisions

[Comment 7 Zac Medico 2010-10-27 19:53:01 UTC]

You missed comment #5.

[Comment 8 Marc Burkhardt 2010-10-27 19:55:26 UTC]

(In reply to comment #7)
> You missed comment #5.
> 


I didn't get the point there ... what use-case do you mean?

[Comment 9 Zac Medico 2010-10-27 20:30:08 UTC]

If an ebuild developer happens to have FEATURES=no{doc,info,man}, it's useful for file collisions to be detected with these files anyway. However, I guess this is hopeless since if they have these features enabled then there won't be any installed files to trigger the collisions.

[Comment 10 Marc Burkhardt 2010-10-27 20:36:56 UTC]

(In reply to comment #9)
> If an ebuild developer happens to have FEATURES=no{doc,info,man}, it's useful
> for file collisions to be detected with these files anyway. However, I guess
> this is hopeless since if they have these features enabled then there won't be
> any installed files to trigger the collisions.
> 

Not only that. If an ebuild developer has these FEATUREs set, it is more than bad, isn't it?

So ... I guess one could implement it as in comment #6 (I like the 'even cooler' approach best!)