Re: [PATCH] use gzip compression with --no-name to create identicals steps backups

["Yann E. MORIN" <yann dot morin dot 1998 at anciens dot enib dot fr>]

Bruno,
All,

On Tuesday 23 March 2010 11:44:50 Bruno Tarquini wrote:
> > That's absolutely not the purpose of those tarballs. What you really want
> > to package is the resulting toolchain. The bckup tarballs are only for
> > debug purposes.
> It wasn't to make a package to distribute but to make faked packages for
> the target system:
> basically, sys-kernel/linux-headers and sys-libs/glibc (gentoo
> embedded). The entire toolchain shouldn't be installed on the target
> (and will not run). And, yeah, I should have rebuilt
> libc but importing faked package from the toolchain was easier and less
> time consuming.

I still don't understand. What do you want those packages for? Do you
mean you want to copy something on the target?

If you mean that your root file system is missing the libraries from
the toolchain, then look at the populate script (that is installed
with your toolchain):

# arm-unknown-linux-uclibc-populate -h
NAME
    arm-unknown-linux-uclibc-populate - populate the target root file system

SYNOPSIS
    arm-unknown-linux-uclibc-populate OPTIONS -s source_root -d destination_root

DESCRIPTION
    arm-unknown-linux-uclibc-populate will 'populate' your target root file
    system ('src_dir') with libraries from the toolchain (eg. libc.so...),
    storing the result into 'dst_dir'.

OPTIONS
    -s src_dir
        use 'src_dir' as the un-populated (source) root directory

    -d dst_dir
        use 'dst_dir' as the place to put the populated root directory

    -l name1[:name2[...]]
        Always add the specified shared library/ies name1, name2... from the
        toolchain (in the sys-root). Actual library names are searched as
        follows (where 'name' is replaced with the given name) in the
        sys-root directory:
          - libname.so
          - name.so
          - name
        If the file is found, then the SONAME of the library is used, and the
        library is copied with that name. If the library was not found, this
        yields an error (unless -f was given).

    -L file
        Read 'file' for a list of shared libraries to always add from the
        toolchain. The file should contain one library name per line; text
        after a # is ignored until the end of the line; spaces are ignored;
        empty lines are ignored. Libraries are searched for as with -l.

    -f  force execution: if destination directory already exists, it will be
        removed first; if a specified library (above) was not found, continue.

    -v  Be verbose. By default, populate is absolutely silent.


Basically, you use it like that:

1) Build all your packages (adapt as needed!):
# cd my-package
# ./configure --prefix=/usr                             \
              --build="build-tuple"                     \
              --host="host-tuple"                       \
              CFLAGS="-I/some/where/staging/usr/include \
              LDFLAGS="-L/some/where/staging/usr/lib
# make
# make DESTDIR=/some/where/staging install
And so on for all your packages...

2) And finally:
# host-tuple-populate -s /some/where/staging \
                      -d /some/where/staging-full

At this point, staging-full/ contains all that your package(s) did install
*and* the required libraries from the toolchain (eg. libc.so, ld-linux.so...)
without touching the staging/ area, thus leaving in staging/ *only* the files
that your packages did install.

> > Wanna come with a patch to this effect, please?
> For correcting what ? changing -3 to -6 in the documentation ? or using
> -3 everywhere ?
> which is what my previous patch did (with the add of the --no-name
> option to which is pretty harmful, IMHO).

Forget it, I'll do the gzip -3. :-)

Just for fun, I re-did the measurements that I used to do a long time ago,
to back my claims that gzip -3 is just enough. I leave up to the reader
the leasure to interpret the folowing numbers! :-)

File:   linux-2.6.33.tar    Size of file:   395100160 (377MiB)
Measured with:  time gzip -1 -c linux-2.6.33.tar |wc -c
(replace 'gzip -1' with any appropriate combo below!).

compression      size  ratio    time       gain  gain/time  (1-ratio)/time
gzip -1     105849688  0.268   6.527  289250472   16217204        0.112164
gzip -3      97533167  0.247   8.213  297566993   11875462        0.091701
gzip -6      85317011  0.216  16.093  309783149    5301498        0.048721
gzip -9      84532819  0.214  31.270  310567341    2703320        0.025137
bzip2 -1     79515133  0.201  44.798  315585027    1774971        0.017830
bzip2 -3     71257550  0.180  44.733  323842610    1592953        0.018323
bzip2 -6     67779650  0.172  46.977  327320510    1442826        0.017635
bzip2 -9     66266488  0.168  49.214  328833672    1346497        0.016911

The 'time' columns is the 'real' part returned by the time command, in
seconds. The file was entirely cached in memory prior to the tests, and
the machine entirely idle, except for the compression.

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'



--
For unsubscribe information see http://sourceware.org/lists.html#faq