Le 22/04/2012 20:23, DJ Lucas wrote:
> On 04/22/2012 12:19 PM, Jeremy Huntwork wrote:
>> On 4/22/12 11:33 AM, LFS Trac wrote:
>>> #3066: Chapter 5 ncurses fails with (old?) gpm on host
>>> -------------------------------------+--------------------------------------
>>>     Reporter:  dj@…                     |       Owner:  lfs-book@…
>>>         Type:  task                     |      Status:  new
>>>     Priority:  normal                   |   Milestone:  7.2
>>> Component:  Book                     |     Version:  SVN
>>>     Severity:  normal                   |    Keywords:
>>> -------------------------------------+--------------------------------------
>>>     Host system is again Gentoo Live DVD 2011. I had to add --without-gpm to
>>>     the configure flags to get around it.
>> (I tried updating the ticket in trac, but that required me resetting my
>> password and now the site puts me in an infinite redirect loop... trac
>> is being wonky)
>>
>> Anyway, unless ncurses is doing something non-standard to detect gpm, I
>> don't think this should be happening. The point of the chapter 5
>> toolchain is to remove /usr or anything like it from the search paths.
>> DJ can you dig in the logs to find out what ncurses is doing to detect
>> gpm? In the meantime, I've downloaded the gentoo live dvd and am going
>> to play a bit.
>>
>> JH
> As it turns out, it was a problem with the host. /usr/lib64/libgpm.so is
> not a symlink, but rather a linker script that points to another linker
> script that points to an invalid destination (ie: no 64bit libgpm). I
> was going to close as invalid, but then I wondered if we want our
> chapter 5 ncurses linked to something that does not exist in the chroot
> environment?
>
> -- DJ Lucas
Hi,

There is still a big flaw in LFS SVN, as I said in 
http://linuxfromscratch.org/pipermail/lfs-dev/2012-March/066119.html

Usually, when testing the presence of a library, configure tries first 
to include the corresponding header file into a short program (using 
gcc), then to link to the library. If /usr/include is in the search path 
of gcc, it may find a header on the host, and then bomb out because it 
does not find the library, since the search path for libraries does not 
include {,/usr}/lib.

This happened to me twice : first on fedora, which has selinux by 
default: coreutils could not build because it wanted selinux libraries. 
Second on Debian with gpm, where I saw the same error as the bug which 
started this thread.

Solution:
add the switch --with-native-system-header-dir=/tools/include to 
gcc-pass2 configure command.
I've been building ten times on various (virtual) hosts with this switch 
without a problem.

Regards
Pierre
-- 
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page

Reply via email to