Revert "Revert "Only build bundled fts if system has a bad version th…

[ignatenkobrain]

…at doesn't handle LFS""

Test failure is bug in fakechroot.

This reverts commit 7e65bec.

[ignatenkobrain]

I've tried to build with Fedora's fakechroot version: https://koji.fedoraproject.org/koji/taskinfo?taskID=21748556

Seems to be green!

[pmatilai]

Yeah it fails because fakechroot doesn't have LFS support - and looking at upstream still doesn't. Fedora packages dont count.

Revisit when fakechroot upstream has a release that supports LFS, it's not like there's any actual urgency in getting this change in.

[ignatenkobrain]

@pmatilai, upstream is looking on it btw.

[mhatle]

Instead of 'fakechroot', have you considered using 'pseudo'? It does support large filesystems and other APIs.

http://git.yoctoproject.org/cgit/cgit.cgi/pseudo/

Both inside and outside of the OpenEmbedded work, we exclusively use pseudo to replace 'fakeroot' and 'fakechroot' capabilities.

[ldv-alt]

@pmatilai, how often do you sync your copy of fts.c with glibc?
I bet your fts.c doesn't have any fixes made in glibc since the last sync.

[ldv-alt]

@n3npq I wish to get rid of fts.c completely, as well as all other stuff that belongs to glibc.
I wish you haven't added that stuff in the first place! :)

[ldv-alt]

@n3npq Why should rpm ever want to walk a remote URI?

Anyway, rpm.org doesn't need its copy of fts.c when linked with glibc >= 2.23, and the bug is that

• rpm.org's maintainer is not ready to admit this fact and act accordingly;
• his attitude to contributors discourages further contributions.

[Conan-Kudo]

Why should rpm ever want to walk a remote URI?

There are at least a couple of occasions where it does. RPM is capable of installing packages from remote locations. It'll fetch the RPM and go through the correct process to install the package as described by the user. I use this all the time for installing source packages, and it can be done with binary ones too.

[Conan-Kudo]

@mhatle Would you consider proposing a patch to switch from fakechroot to pseudo? It looks like we have pseudo in Fedora, and if RPM's testsuite depends on it, I can introduce packages for Mageia and openSUSE so that the test suite can be run there.

[mhatle]

@Conan-Kudo patch for what? Generally speaking switching from fakeroot/fakechroot to pseudo should be seamless... (simple search/replace)

[Conan-Kudo]

@mhatle Aside from that, @n3npq mentioned this:

Almost everything Just Works ... the few tests failing can likely be fixed rapidly.

If you could provide a PR with the change and fixed the failing tests, I think it'd be quite appreciated.

[ignatenkobrain]

fakechroot patch has been merged in upstream btw.

[pmatilai]

What a tempest in a teapot. Folks, time for a reality check.

I dislike bundled copies at least as much as the next guy, but:

• Replacing X with Y broke Z, yes Y gets blamed first
• Sometimes life's more complicated than that
• Short-term, a working test-suite is FAR MORE VALUABLE to me than ideological "bundling is not nice"

We're talking about something bundled 20 years ago and on my watch, I don't recall seeing a single bug beyond portability tweaks reported on the bundled FTS copy. And I sure as hell dont have time to listen up this OH MY GOD HOUSE IS ON FIRE yack-yack on what is from rpm functional point a total non-issue.

I'll revisit the issue when its time.

[ldv-alt]

@pmatilai, you are a perfect marvel! Thank you for being so grateful to us for getting rid of bundled X and fixing your buggy Z.
Thanks everybody for this enlightening discussion, I suppose no more comments are needed here.

[Conan-Kudo]

@ldv-alt That's just rude. 👎

[pmatilai]

@ldv-alt - the thing is, a "use foo on glibc" kind of change doesn't really do anything to get rid of the bundled code because it'll still be needed for other platforms. Instead these things end up adding more burden by introducing another path requiring validation. So when such a change breaks something (even if the something is broken by itself) without actually bringing in any concrete bug fixes or enhancements, pardon me for not jumping up and down in joy in here.

You really want to get rid of the bundled FTS, rewrite it using something that's actually standardized, such as nftw().

[pmatilai]

Half a year later, there's still no fakechroot upstream release with the relevant fix in it. Not exactly a huge surprise since the latest release is from 2016.

We can reconsider when they finally cut that release. In the meanwhile, see my earlier comment about nftw()...