[Clfs-dev] Cross-Compiled Linux From Scratch - Version GIT-20140611-x86_64-Pure64

[Martin Ward]

On 15/06/14 01:37, William Harrington wrote:
>
> On Jun 14, 2014, at 1:59 PM, Martin Ward wrote:
>
>> Hi Everyone
>>
>> Just finished building the above  and all in all , its in good shape, 
>> the changes to the boot method have worked great with one or two 
>> exceptions
>>
>> 1. I though we were in a package freeze, found two updates to packages ?
>
> Security issues, which the package freeze mail I hope to have made 
> clear that upgrades or patches would be considered for security issues.
well yes  but there is no link as per lfs/blfs trac saying why  we did it
>
>> 2 . There does seem a penchant for alphabetising the order of the 
>> packages for no particular reason other than abc, which undoubtedly 
>> works fine if you are scripting but if you are trying to read the 
>> book, doesn't work for me. I much prefer a functional approach, but 
>> do appreciate its a moot point.
>
> Alphabetical is okay. Some are trivial, but others need to be in a 
> certain order for testsuite coverage.
>
>> However I do have what I think would be some improvements, mostly 
>> text explanations
>>
>> Section 5 Cross Compile Tools
>> Some of the wording in why we are doing what we are doing, i think 
>> needs to be expanded a little
>> ie we install these particular versions of the programs as we need 
>> these versions, rather than making do with ones that are already on 
>> the host system which could be unsuitable for a variety of reasons.
>> example
>> 5.4.1
>> "Here we will install an *m4* binary that can run on the host system, 
>> as this program is required by GMP"
>>
>> Perhaps we should be more explict, in saying we install this 
>> particular version of m4 as it provides the required features needed 
>> by GMP in this chapter and in the basic tools chapter,as the host may 
>> not have m4 or one that is compatible with the particular version of  
>> GMP . ditto file 5.3.1
>>
>
> We'll look into those a bit closer. Thanks for the input.
>
>> Chapter 6  constructing a temp system loooks good
>>
>> Chapter 7
>> if your going to boot
>>
>> 7.1 No idea why Bc was moved here , to me it fits better in chapeter 
>> 5 as its a "host" program but moot point
>
> BC isn't a host system requirement to build CLFS with unless the boot 
> method is being used. In this case, BC needs to be available for 
> building the kernel with the cross compiler. It isn't required when 
> chrooting and building the kernel.
appreciated
>
>> 7.3 Bootscripts , really should go either before or after compiling 
>> the kernel, flows much better and then is in line with chapter 10
>
> Well, it is before compiling the kernel at the beginning of chapter 7.
to  me if would  it flows better if it went after udev and before the 
kernel compile
>
>> These seem to work great apart from the setclock script, either i did 
>> something wrong, or after the reboot, my clock was always set to UTC 
>> , whatever value I set the environment variable to.
>> I suspect a hard coded path in util-linux some where.
>>
>
> We'll need to look into this.
>
>> 7.14
>>
>> both  nogroup and nobody have the same GID
>> suggest change nobody to 65534
> I'm missing something in what you are stating.
>
> nogroup gid in /etc/group is 65533
> nobody uid is 65534 and gid is 65533
>
>     groups nobody:
>     nobody : nogroup
>
> We don't have a nobody group with any gid in /etc/group.
>
> We have a nobody user and a nogroup group. One uid, one gid.
>
its in the additional groups bit
>> 10.4.1 Temporary Perl , this seems a better fit in section 9 test 
>> suite tools, but moot point again
>>
>
> Have you attempted cross compiling perl? It's in ch10 cause of the issue.
> I suppose if you want to give this patch a whirl and try to cross 
> compiler perl for a minimal perl to use in ch10 then there is this:
>
> http://mirrors.ustc.edu.cn/lfs/clfs/conglomeration/perl/perl-5.10.0-cross_compile-1.patch 
>
>
> May be a bit rough with Perl 5.20.0.  I haven't attempted cross 
> compiling perl, and I suspect it is still problematic.
>
i did once or twice many moons ago, but didn't get anywhere, as you say 
problematic

>> 10.26.1 Adjusting the hwclock location, doesn't seem necessary, as 
>> systemd uses /etc/adjtime to hold the data
>>
>
> CLFS and LFS devs had a long discussion about the whole issue. We left 
> it alone.
>
> Two specific issues which arise:
>
> anything in /etc should be static and if it is changing is a security 
> issue.
> Although /var may not be mounted when needed for hwclock if /var is a 
> separate filesystem.
>
> Then there is the FHS.
>
>> 10.67 Man-db should come after 10.54 Libpipeline, flows better
>>
>
> It would be best moved after Llibpipeline.
>
>> Otherwise good job
>> No particular errors that looked serious in the gcc, glibc testsuites
>>
>> cheers
>>
>> Martin
>
> Thank you for your input.
>
> Sincerely,
>
> William Harrington
>
thanks for the explanations

martin