Glibc 2.16.0 Suggestions

[Armin]

Hi there. I just built Glibc 2.16.0 on my currently running system that has GCC 4.7.1, Binutils 2.22, Glibc 2.14.1, running Linux Kernel 3.4.5 with other tools more or less as in SVN version of the book.

That means I didn't build temporary toolchain and use it to build glibc, but instead I've built it only for testing.

I've noticed few things there - in Chapter 6 Glibc:

"As in Chapter 5, add the needed compiler flags to CFLAGS for x86 machines. Here, the optimization of the library is also set for the gcc compiler to enhance compilation speed (-pipe) and package performance (-O3)."

I didn't use that command to create configparms and I was running 32bit operating system. It didn't fail with 2.16.0, but I do remember it failed with 2.13 when I missed that step.

"--libexecdir=/usr/lib/glibc"

The libexecdir now contains gconv or getconf dir (not sure), not only pt_chown program. I suggest you change it to just /usr/lib as other distros do.

"cp -v ../glibc-2.16.0/iconvdata/gconv-modules iconvdata"

I didn't run this one, but iconvdata/gconv-modules exists in my build dir. I guess it isn't necesary anymore.

"If you have mounted the LFS partition with the noatime option, the atime test will fail. As mentioned in Section 2.4, “Mounting the New Partition”, do not use the noatime option while building LFS."

Strange enough, all my ext4 partitions are mounted with noatime including the partition where glibc was built and tested. No tests failed regarding that. So I guess that warning doesn't apply anymore.

and at last

grep Error glibc-check-log
make[2]: [/home/armin/build/posix/annexc.out] Error 1 (ignored)
make[2]: [/home/armin/build/conform/run-conformtest.out] Error 1 (ignored)

The run-conformtest.out are some strange tests which I've never seen untill today. They take good amount of time and their results are like gcc's or binutils' results.

[ken@…]

For the libexecdir, the directory is getconf/ with three executables. I don't actually care what we call it - BLFS has started to allow /usr/libexec.

I too get the ignored Error 1 for run-conformtest.out, guess we ought to mention it.

(no comment on ext4, my /mnt/lfs is mounted with the defaults)

However, I'm suspicious about iconvdata/gconv-modules : when I built and installed 2.16.0, I copied that. After the install and a full localedata/install-locales I have a /usr/lib/gconv directory containing a lot of things. I tried without the copy for an install_root (DESTDIR) install : certainly that file is in the glibc-build tree, but I *don't* have the gconv directory in the install_root.

The only difference between two versions of iconvdata/gconv-modules is that the one created in the build tree is mode 644 instead of 664.

[Armin]

Replying to ken@…:

For the libexecdir, the directory is getconf/ with three executables. I don't actually care what we call it - BLFS has started to allow /usr/libexec.

I just want to say that I personaly won't make use of /usr/libexec anywhere where I make the changes untill it is official.

I was even thinking about reverting that for Postfix and vte-0.28 untill it's official.

[bdubbs@…]

I made several changes in my sandbox.

• remove --libexecdir. The default is /usr/libexec with one file, pt_chown, and a directory getconf/ with the three files Ken mentions. I don't have any problem with leaving /usr/libexec.

• omit cp ...gconv-modules iconvdata. There were no issues

• remove CFLAGS in both Chapter 5 and Chapter 6. It doesn't make any difference in x86_64, but I'll test is x86 tomorrow

• mount /mnt/lfs noatime. So far no issues, but it still has a couple of hours to go. The gcc and automake tests take a long time, so I'll let that run over night.

[bdubbs@…]

Omitting cp ...gconv-modules iconvdata did not cause any problems on either a x86 or x86_64 system.

Mounting the build partition with noatime did not cause any problems on either a x86 or x86_64 system.

Removing the CFLAGS in Chapter 5 or 6 did not cause any problems on either a x86 or x86_64 system.

For now, lets keep the --libexecdir as it is.

[bdubbs@…]

Fixed at revision 9956.