Axel Beckert:
Hi Niels,
Niels Thykier wrote:
[...]
It was indeed an oversight.
Yep, looked like that: Was the only occurence far away from the
remaining calls.
Also wondered if that call better should have been a simple
install_file() instead.
The $mode parameter gets in the way. However, no one is using that
function outside of debhelper, so I can just simplify it so we can use
install_file instead - done!
Thanks for the suggestion.
I will probably not change anything for now. Having slept on it and given
additional evidence, I think the underlying issue was a bug in fakeroot
(#1023286 is most my most likely candidate) and not a bug in debhelper
I see. Given that comment and the fact that #1023286 only affects 32
bit architectures, I think that #1024261 (in debhelper) is related and
might have the same cause as only 32-bit files seem to be affected.
(Or was #1024261 the whole trigger for that 4-argument conversion? Saw
no comment from you there so far.)
I got #1024261 filed against debhelper, which made me assume this was a
bug in debhelper. While I had seen #1023286, I originally thought
debhelper was not doing the job properly as well. This prompted me to do
the 4-argument conversion to resolve it. To be honest, it was a
precursor for a change to `install_dir`, which I thought we would be the
full fix - but `install_dir` had "complicated" usage outside debhelper,
so I got rid of the easy parts first.
However, post upload, I finally understood that the real problem is not
related to debhelper creating the directory with the wrong ownership (if
it did, *all* directories in the dbgsym would be wrong). Instead, it is
most likely a "cp" or a "mv" call that copies a file into the directory
that causes fakeroot to drop its ownership information.
I.e., the 4-argument conversion changes will not change anything as the
bug is 100% fakeroot. At least hindsight is still 20/20.
(meaning the change to 4 parameters should probably be undone).
Ack, sounds as if this is probably the way to go once it's clear that
#1023286 was really the cause. I also wonder if some versioned Breaks
or Conflicts would be helpful, too.
Regards, Axel
I am not sure version Breaks/Conflicts is necessary to be honest. My
13.11 upload included a fail-safe that works around this particular bug
in fakeroot in dbgsyms - one could argue that I "fixed" #1024261 with
that. At the time, I felt it was a work around, so I kept it open.
However, I think once the dusts settles, I will probably close #1024261
as "not a bug in debhelper" and undo the 4-argument conversion.
But for now - I will just let the dust settle; after all the fakeroot
bug is still present in mipsel due to a FTBFS (#1024544)...
Thanks,
~Niels
