Bug 487474 – [Future EAPI] Make it possible for pkg_setup to create build-writable files

• Status: CONFIRMED
• Product: Gentoo Hosted Projects

[Description Michał Górny 2013-10-09 20:41:01 UTC]

While not strictly covered by PMS, most of the PMs have some kind of userpriv. As a result, pkg_* phases are done with root privileges and src_* phases with some kind of pre-defined build user. However, this user is inconsistent between PMs and there's no good way of obtaining it.

As a result, there's no clear way for pkg_setup() to create files that will be clearly readable & writable by the remaining phases.

Please either:

1. guarantee that any files created during pkg_setup will be owned by the user used for build (or have equivalent permissions),

2. provide us a way to obtain the user that will be used for build, so we could chown them ourselves.

[Comment 1 Arfrever Frehtes Taifersar Arahesis 2013-10-09 20:48:54 UTC]

(In reply to Michał Górny from comment #0)
> 2. provide us a way to obtain the user that will be used for build, so we
> could chown them ourselves.

http://git.overlays.gentoo.org/gitweb/?p=proj/portage.git;a=commitdiff;h=4b22701aa7733e53778fc4d456819fc010e198b0

[Comment 2 Ciaran McCreesh 2013-10-09 20:49:55 UTC]

It feels like you're trying to do something icky here. What's the use case, and how does it interact with binary packages?

[Comment 3 Michał Górny 2013-10-09 20:53:59 UTC]

(In reply to Ciaran McCreesh from comment #2)
> It feels like you're trying to do something icky here. What's the use case,
> and how does it interact with binary packages?

The use case is preparing Python wrappers. It may be done in pkg_setup(), and then the same wrapper can be reused for remaining phases. Since pkg_setup is called both for binary & source installs, the wrappers are created in both cases the same way.

[Comment 4 Jouni Kosonen 2013-10-09 20:59:33 UTC]

(In reply to Arfrever Frehtes Taifersar Arahesis from comment #1)
> http://git.overlays.gentoo.org/gitweb/?p=proj/portage.git;a=commitdiff;
> h=4b22701aa7733e53778fc4d456819fc010e198b0

Interesting. Apropos and OT, surely 

@@ -1488,0 +1489,1 @@ def spawn(mystring, mysettings, debug=0, free=0, droppriv=0, sesandbox=0, fakero
+                               "umask": 0o02

is a typo?

[Comment 5 Arfrever Frehtes Taifersar Arahesis 2013-10-09 21:03:40 UTC]

(In reply to Jouni Kosonen from comment #4)

Where do you see a typo?

[Comment 6 Jouni Kosonen 2013-10-09 21:27:08 UTC]

(In reply to Arfrever Frehtes Taifersar Arahesis from comment #5)
> Where do you see a typo?

Sorry, braino. Of course the group write bit is on when the group is set.

[Comment 7 Ulrich Müller 2013-10-09 23:14:13 UTC]

(In reply to Michał Górny from comment #0)
> 1. guarantee that any files created during pkg_setup will be owned by the
> user used for build (or have equivalent permissions),

I still don't understand the use case. _Where_ would pkg_setup possibly create such files? WORKDIR and D are not valid in pkg_setup, and I assume that you don't want to write to the live system. That would leave only HOME and T.

[Comment 8 Michał Górny 2013-10-10 06:03:35 UTC]

(In reply to Ulrich Müller from comment #7)
> (In reply to Michał Górny from comment #0)
> > 1. guarantee that any files created during pkg_setup will be owned by the
> > user used for build (or have equivalent permissions),
> 
> I still don't understand the use case. _Where_ would pkg_setup possibly
> create such files? WORKDIR and D are not valid in pkg_setup, and I assume
> that you don't want to write to the live system. That would leave only HOME
> and T.

${T}/_${EPYTHON}, to be more precise.

[Comment 9 Ulrich Müller 2013-10-10 07:50:25 UTC]

Would a couple of variables like BUILD_USER and BUILD_GROUP be enough? Or a single variable BUILD_USER containing "user:group"?

[Comment 10 Michał Górny 2013-10-10 08:00:16 UTC]

(In reply to Ulrich Müller from comment #9)
> Would a couple of variables like BUILD_USER and BUILD_GROUP be enough? Or a
> single variable BUILD_USER containing "user:group"?

Either will work for me.