On 9/20/16 10:37 PM, Robert Yang wrote:
> 
> 
> On 09/21/2016 04:33 AM, Mark Hatle wrote:
>> On 9/20/16 10:00 AM, Burton, Ross wrote:
>>>
>>> On 20 September 2016 at 09:15, Hongxu Jia <hongxu....@windriver.com
>>> <mailto:hongxu....@windriver.com>> wrote:
>>>
>>>     -Upstream-Status: Submitted [Sent email to rpm-de...@rpm5.org
>>>     <mailto:rpm-de...@rpm5.org>]
>>>     +Upstream-Status: Rejected [Sent email to rpm-de...@rpm5.org
>>>     <mailto:rpm-de...@rpm5.org>]
>>>     +http://rpm5.org/community/rpm-devel/5655.html
>>>     <http://rpm5.org/community/rpm-devel/5655.html>
>>>
>>>
>>> Considering upstream has explicitly rejected this patch, why should we 
>>> accept it?
>>>
>>> Ross
>>>
>>>
>>
>> I'm confused by what the patch is doing looking at it.
>>
>> It sounds like from the description there is a bug that without the change,
>> packages with (intentionally) bad checksums and such are allowed to be 
>> installed.
>>
>> The bug is caused by a previous patch that enabled nosignature, etc -- 
>> because
>> the comparisons turned out to be backwards.
>>
>> So really nosignature, etc is already in place -- it's just not working 
>> properly?
>>
>> What was rejected upstream is the use of nosignature in any context.  RPM5
>> maintainer believes it is unwise and unsafe to permit uses to install 
>> packages
>> that have failed basic validation.  (I tend to agree.)  Similarly, even being
>> able to run queries and other operations against them may be dangerous as 
>> well.
> 
> If command like "rpm5 -qp --nosignature --nodigest <file.rpm>" queries 
> database,
> then it would cause rpm5 hang when more than one "rpm5 -qp" is running, so I
> fixed the *query* problem, and the *install* problem is not caused by the fix.
> Btw., *rpm4* doesn't query database when "rpm -qp file.rpm", if we are going
> to use dnf in next release, maybe we can consider using rpm4 as Fedora does.
> I did a rough statistics for oe-core's local patches, the winner is rpm, it
> has 77 local patches, that's not good for maintaining, and you can imagine 
> that

30 patches are simply configuration changes.

42 patches are bug fixes that have been submitted and will likely go away in the
next uprev.

> how many times it surprised us. rpm4 should be more stable than rpm5, I don't
> know how many distros use rpm5, I can't access rpm5.org atm, it seems that the
> web is down (It was OK yesterday), but widely used distros like Redhat and 
> Suse

The website is working for me today.

> uses rpm4, so maybe rpm4 will make our life much more easier, and also for
> yocto's user. I think that why we need rpm5 before was because we need use it

There is a major problem with rpm4.  It doesn't support cross compiling at all.
We know in the past there have been significant problems with it and BerkleyDB
with endian issues.  Also it was not possible to do cross-architecture installs
(in the past, this one might be resolved.)

The general functionality between RPM4 and RPM5 is nearly identical.  RPM 4 has
a plug-in interface (which I personally don't believe should be used due to
security issues.)  RPM 5 has a focus around specific security enhancements.
Some are reasonable, some are a bit too experimental to be used.

The other main difference is software license.  RPM 4 is 'GPLv2', while RPM5 is
'LGPLv2'.  This doesn't sound like a big deal at first -- but RPM 4 can't be
linked to OpenSSL for it's crypto library.  RPM 4 is linked ONLY to NSS, while
RPM5 can use beecrypt, nss, openssl and others.  This makes for a significantly
more flexible embedded system.

> to compute the package dependencies, but now we have smartpm, so we don't rely
> on that any more.
> 
> Here is the recipe list which has more than 10 local patches, and we have 1762
> patches atm:
>       77 rpm

I'd think it's a better comparison to say '30', the configuration items.  So
it's similar to perl or openssl in complexity.  I'd agree that is a fair 
comparison.

>       59 gcc-5.4
>       49 gcc-6.2
>       36 ltp
>       31 python3
>       30 perl
>       29 openssl
>       28 man
>       27 glibc
>       24 python
>       23 tcp-wrappers-7.6
>       23 systemd
>       18 busybox
>       14 syslinux
>       14 python-smartpm
>       14 elfutils-0.166
>       14 apt
>       13 qemu
>       13 libtool
>       13 gstreamer1.0-plugins-bad
>       13 glib-2.0
>       13 binutils
>       13 bind
>       12 coreutils-6.9
>       11 valgrind
>       11 gdb
>       11 dhcp
>       11 autoconf
>       10 unzip
>       10 dpkg
>       10 console-tools-0.3.2
> 
>>
>> If fixing the problem is as simple as reverting the other change -- and that
>> doesn't cause other problems elsewhere...  I'd rather see that.
> 
> We can't revert the patch, it would break packagefeed-stability.bbclass,
> we will see the hang

Something is wrong with the implementation then.

You should not be doing the equivalent of:

list = '<all packages>'
for each in list; do rpm -qp $each ; done

This is HORRIBLY inefficient, even with rpm4.

The proper way to query a number of packages is in a single transaction.  This
permits RPM to load the data and validation structures once.

This is equivalent to:
  rpm -qp `cat list`

or

  echo list > /tmp/tmpfile
  rpm -qp /tmp/tmpfile

(The later solving the problem of 'too many arguments')

This may seem counter intuitive, as the first way looks to be better on a
multi-core machine -- but in my experience the overhead of starting rpm, setting
up the data structures is very expensive, much more then the
load/validate/display of the package information.

--Mark

> 
>>
>> --Mark
>>

-- 
_______________________________________________
Openembedded-core mailing list
Openembedded-core@lists.openembedded.org
http://lists.openembedded.org/mailman/listinfo/openembedded-core

Reply via email to