Error in multilib build when doing canadian cross

Error in multilib build when doing canadian cross

[Per Arnold Blaasmo]

In gcc.sh you find the following:
    if [ "${CT_MULTILIB}" = "y" ]; then
        multilibs=( $( "${core_prefix_dir}/bin/${CT_TARGET}-gcc"
-print-multi-lib   \
                       |tail -n +2 ) )

This fails when trying to do a candian cross where target is win32.
The error is:
..../scripts/build/cc/gcc.sh: line 413:
.../.build/arm-none-eabi/build/gcc-core-static/bin/arm-none-eabi-gcc: No
such file or directory

The arm-none-eabi-gcc.exe is there, but when doing a canadian cross that
fails on my linux machine. If the build had been run on a windows it
would have accepted the filename without '.exe', but not if the build is
run on a linux machine.

So if it is a canadian build it should rather run the build servers
cross toolchain.

Sugested fix:

    if [ "${CT_MULTILIB}" = "y" ]; then
        if [ "{CT_CANADIAN}" = "y" ]; then
            multilibs=( $( "${CT_TARGET_PREFIX}-gcc" -print-multi-lib   \
                       |tail -n +2 ) )
        else
            multilibs=( $( "${core_prefix_dir}/bin/${CT_TARGET}-gcc"
-print-multi-lib   \
                   |tail -n +2 ) )
       fi


-- 
Per Arnold Blåsmo
Senior Design Engineer, Atmel Norway


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

Re: Error in multilib build when doing canadian cross

[Yann E. MORIN]

On Tuesday 06 March 2012 11:38:19 Per Arnold Blaasmo wrote:
> In gcc.sh you find the following:
>     if [ "${CT_MULTILIB}" = "y" ]; then
>         multilibs=( $( "${core_prefix_dir}/bin/${CT_TARGET}-gcc"
> -print-multi-lib   \
>                        |tail -n +2 ) )
> 
> This fails when trying to do a candian cross where target is win32.
> The error is:
> ..../scripts/build/cc/gcc.sh: line 413:
> .../.build/arm-none-eabi/build/gcc-core-static/bin/arm-none-eabi-gcc: No
> such file or directory

This has already been fixed in the repository:
    http://crosstool-ng.org/hg/crosstool-ng/rev/4fcedd2c14b2

But since you bring up the problem, there still is an issue elsewhere in
that same build script. On line 736, we have a similar construct (for the
final gcc build):

780   if [ "${CT_MULTILIB}" = "y" ]; then
781       multilibs=( $( "${CT_PREFIX_DIR}/bin/${CT_TARGET}-gcc" -print-multi-lib \
782                      |tail -n +2 ) )
783       if [ ${#multilibs[@]} -ne 0 ]; then
784           CT_DoLog EXTRA "gcc configured with these multilibs (besides the default):"
785           for i in "${multilibs[@]}"; do
786               dir="${i%%;*}"
787               flags="${i#*;}"
788               CT_DoLog EXTRA "   ${flags//@/ -}  -->  ${dir}/"
789           done
790       else
791           CT_DoLog WARN "gcc configured for multilib, but none available"
792       fi
793   fi

Looking at line 781, this would indeed not work for canadian crosses.
So, we can not use the newly-built compiler to get the list of available
multilibs... And because we are doping a canadian cross, we do *not* have
the core gcc either. And the compiler that was used to build the C library
(a step that you did not reach yet) might not be multilib aware either...

So, I am afraid that multilib is currently a no-no for canadian crosses.
Yes, that *is* a hard limitation.

Fortunately, I already have some patches localy, that do revamp the canadian
case code-path. In this rework-in-progess, the core compilers will actually
be built, so there will no longer be a requirement for a target cross-compiler
to pre-exist, so we'll be able to query the core compiler to get the multilib
listing, even (and especially) ion the canadian case.

Quoting again that part of the message:
> .../.build/arm-none-eabi/build/gcc-core-static/bin/arm-none-eabi-gcc

So, you are building for a bare-metal target?
That's probably why you have an issue in the core compiler code, because
the core compiler code path is used to build the bare-metal final gcc.

So, again, for canadian cross, we would not be able to call the newly-built
core compiler (which is in fact a final compiler)...

When I thought I had most bases covered, and a quiet evening in sight!...
Sigh... ;-)

Thanks for pointing the issue! :-)

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

Re: Error in multilib build when doing canadian cross

[Per Arnold Blaasmo]

Hi,

On 06. mars 2012 20:00, Yann E. MORIN wrote:
> But since you bring up the problem, there still is an issue elsewhere in
> that same build script. On line 736, we have a similar construct (for the
> final gcc build):
> 
> 780   if [ "${CT_MULTILIB}" = "y" ]; then
> 781       multilibs=( $( "${CT_PREFIX_DIR}/bin/${CT_TARGET}-gcc" -print-multi-lib \
> 782                      |tail -n +2 ) )
> 783       if [ ${#multilibs[@]} -ne 0 ]; then
> 784           CT_DoLog EXTRA "gcc configured with these multilibs (besides the default):"
> 785           for i in "${multilibs[@]}"; do
> 786               dir="${i%%;*}"
> 787               flags="${i#*;}"
> 788               CT_DoLog EXTRA "   ${flags//@/ -}  -->  ${dir}/"
> 789           done
> 790       else
> 791           CT_DoLog WARN "gcc configured for multilib, but none available"
> 792       fi
> 793   fi
> 
> Looking at line 781, this would indeed not work for canadian crosses.
> So, we can not use the newly-built compiler to get the list of available
> multilibs... And because we are doping a canadian cross, we do *not* have
> the core gcc either. And the compiler that was used to build the C library
> (a step that you did not reach yet) might not be multilib aware either...
> 
> So, I am afraid that multilib is currently a no-no for canadian crosses.
> Yes, that *is* a hard limitation.
Hmmm... that's to bad :-(

> 
> Fortunately, I already have some patches localy, that do revamp the canadian
> case code-path. In this rework-in-progess, the core compilers will actually
> be built, so there will no longer be a requirement for a target cross-compiler
> to pre-exist, so we'll be able to query the core compiler to get the multilib
> listing, even (and especially) ion the canadian case.
> 
> Quoting again that part of the message:
>> .../.build/arm-none-eabi/build/gcc-core-static/bin/arm-none-eabi-gcc
> 
> So, you are building for a bare-metal target?
Yes, arm-none-eabi target.

I have set up a Jnekins build job using an ant script to first build a
native x86_64 compiler for arm-none-eabi.

Then I do canadian cross for i686 and mingw32.

Currently I do not make it multilib, but to align it with what others do
when build arm compiler, I would bvery much like it to be multilib.

> That's probably why you have an issue in the core compiler code, because
> the core compiler code path is used to build the bare-metal final gcc.
> 
> So, again, for canadian cross, we would not be able to call the newly-built
> core compiler (which is in fact a final compiler)...
But when doing a canadian cross to mingw32 it is difficult to run the
core compiler since it is built for windows. And running windows exe on
linux demands that you have wine installed. And that complicates the
build very much.

So instead I though it was more correct to use the native compiler to
ask for multilib configuration. Of course that requires that the native
compiler is compile with the same parameters that the candian cross is
using.

> 
> When I thought I had most bases covered, and a quiet evening in sight!...
> Sigh... ;-)
> 
Sorry too open a can of worms ;-)

> Thanks for pointing the issue! :-)
>
My pleasure :-)

Regards
Per A.



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