[Buildroot] [PATCH v2] toolchain: add a harmless link when sysroot detection is not accurate

[Vicente Olivert Riera <Vincent.Riera@imgtec.com>]

Hello Romain,

On 02/02/16 15:21, Romain Naour wrote:
> Hi Vicente,
>
> Le 02/02/2016 11:57, Vicente Olivert Riera a ?crit :
>> ping
>>
>> Currently Codescape MIPS toolchains are broken for big-endian. This patch fixes
>> the problem.
>>
>> On 22/01/16 16:05, Vicente Olivert Riera wrote:
>>> Sometimes is not possible to detect if the sysroots are nested or side
>>> by side. For instance this happens for MIPS big endian, where the
>>> sysroot and the arch-sysroot directories are the same. Examples:
>
> By sometime, you mean toolchains generated by Crosstool-ng and Buildroot which
> doesn't support multi-lib.

Only toolchains generated by Crosstool-ng and Buildroot? Are you sure 
about that? I think the Codescape toolchains are not generated by 
Crosstool-ng and they do support multi-lib.

>>>
>>> === Mentor Sourcery CodeBench toolchain
>>>
>>> $ readlink -f $(LANG=C mips-linux-gnu-gcc -print-file-name=libc.a) | sed
>>> -r -e 's:(usr/)?lib(32|64)?([^/]*)?/([^/]*/)?libc\.a::'
>>> /toolchains/mips-2015.11/mips-linux-gnu/libc/
>>>
>>> $ readlink -f $(LANG=C mips-linux-gnu-gcc -EB -print-file-name=libc.a) |
>>> sed -r -e 's:(usr/)?lib(32|64)?([^/]*)?/([^/]*/)?libc\.a::'
>>> /toolchains/mips-2015.11/mips-linux-gnu/libc/
>>>
>>> === Codescape toolchain
>>>
>>> $ readlink -f $(LANG=C mips-img-linux-gnu-gcc -print-file-name=libc.a) |
>>> sed -r -e 's:(usr/)?lib(32|64)?([^/]*)?/([^/]*/)?libc\.a::'
>>> /toolchains/mips-img-linux-gnu/2015.06-05/sysroot/mips-r6-hard/
>>>
>>> $ readlink -f $(LANG=C mips-img-linux-gnu-gcc -EB
>>> -print-file-name=libc.a) | sed -r -e
>>> 's:(usr/)?lib(32|64)?([^/]*)?/([^/]*/)?libc\.a::'
>>> /toolchains/mips-img-linux-gnu/2015.06-05/sysroot/mips-r6-hard/
>>>
>>> For those cases where the sysroot detection is not accurate we add a
>>> harmless symlink necessary for certain toolchains such as Codescape for
>>> MIPS. The other toolchains in this situation (unknown sysroot layout)
>>> don't need this symlink to work, but as stated before, this symlink is
>>> harmless.
>
> Since Codescape are supported in Buildroot as external toolchain and the issue
> appear only with it for BE, you can probability use a
> TOOLCHAIN_EXTERNAL_POST_INSTALL_STAGING_HOOKS like for some Linaro toolchains
> instead ?
>
> Thoughts ?

Well, the whole point for this patch [1] was to get rid of that hook 
(which was the first approach when I sent the patches to add the 
Codescape toolchains) and add that functionality in the toolchain infra 
just in case more toolchains with this problem were added in the future.

1: 
https://git.busybox.net/buildroot/commit/?id=9a1e9efe2618ebbf12ed2567bcd6e8cca5619547

Maintainers, thoughts?

>>>
>>> Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
>>> ---
>>> Changes v1 -> v2:
>>>    - Update the comment above the copy_toolchain_sysroot function.
>>>    - Explain in the commit log which toolchains are affected and provide
>>>      some outputs of -print-file-name to explain the problem.
>>>      (Both changes requested by Thomas Petazzoni)
>>>
>>>    toolchain/helpers.mk | 16 ++++++++++++++++
>>>    1 file changed, 16 insertions(+)
>>>
>>> diff --git a/toolchain/helpers.mk b/toolchain/helpers.mk
>>> index 70695ee..c8245f7 100644
>>> --- a/toolchain/helpers.mk
>>> +++ b/toolchain/helpers.mk
>>> @@ -130,6 +130,12 @@ copy_toolchain_lib_root = \
>>>    #
>>>    # * Create a symbolic link
>>>    #
>>> +# If we are in the situation where we don't know if the sysroot layout
>>> +# is nested of side by side, we:
>>> +#
>>> +# * Create a symbolic link because it's harmless and is necessary for
>>> +#   certain toolchains like Codescape for MIPS.
>>> +#
>>>    # Finally, some toolchains (i.e Linaro binary toolchains) store
>>>    # support libraries (libstdc++, libgcc_s) outside of the sysroot, so
>>>    # we simply copy all the libraries from the "support lib directory"
>>> @@ -160,9 +166,12 @@ copy_toolchain_sysroot = \
>>>        done ; \
>>>        SYSROOT_DIR_CANON=`readlink -f $${SYSROOT_DIR}` ; \
>>>        ARCH_SYSROOT_DIR_CANON=`readlink -f $${ARCH_SYSROOT_DIR}` ; \
>>> +    echo "SYSROOT_DIR_CANON: $${ARCH_SYSROOT_DIR_CANON}" ; \
>
> should be SYSROOT_DIR_CANON

Yes, that's right.

Regards,

Vincent.

> Best regards,
> Romain
>
>>> +    echo "ARCH_SYSROOT_DIR_CANON: $${ARCH_SYSROOT_DIR_CANON}" ; \
>>>        if [ $${SYSROOT_DIR_CANON} != $${ARCH_SYSROOT_DIR_CANON} ] ; then \
>>>            relpath="./" ; \
>>>            if [ $${ARCH_SYSROOT_DIR_CANON:0:$${\#SYSROOT_DIR_CANON}} ==
>>> $${SYSROOT_DIR_CANON} ] ; then \
>>> +            echo "Nested sysroots detected." ; \
>>>                if [ ! -d $${ARCH_SYSROOT_DIR}/usr/include ] ; then \
>>>                    cp -a $${SYSROOT_DIR}/usr/include $(STAGING_DIR)/usr ; \
>>>                fi ; \
>>> @@ -174,9 +183,16 @@ copy_toolchain_sysroot = \
>>>                ln -s $${relpath} $(STAGING_DIR)/$${ARCH_SUBDIR} ; \
>>>                echo "Symlinking $(STAGING_DIR)/$${ARCH_SUBDIR} -> $${relpath}" ; \
>>>            elif [ `dirname $${ARCH_SYSROOT_DIR_CANON}` == `dirname
>>> $${SYSROOT_DIR_CANON}` ] ; then \
>>> +            echo "Side by side sysroots detected." ; \
>>>                ln -snf $${relpath} $(STAGING_DIR)/`basename
>>> $${ARCH_SYSROOT_DIR_CANON}` ; \
>>>                echo "Symlinking $(STAGING_DIR)/`basename
>>> $${ARCH_SYSROOT_DIR_CANON}` -> $${relpath}" ; \
>>>            fi ; \
>>> +    else \
>>> +        relpath="./" ; \
>>> +        echo "Unable to detect if sysroots are nested or side by side." ; \
>>> +        echo "Creating a harmless symlink necessary for certain toolchains." ; \
>>> +        ln -snf $${relpath} $(STAGING_DIR)/`basename
>>> $${ARCH_SYSROOT_DIR_CANON}` ; \
>>> +        echo "Symlinking $(STAGING_DIR)/`basename $${ARCH_SYSROOT_DIR_CANON}`
>>> -> $${relpath}" ; \
>>>        fi ; \
>>>        if test -n "$${SUPPORT_LIB_DIR}" ; then \
>>>            cp -a $${SUPPORT_LIB_DIR}/* $(STAGING_DIR)/lib/ ; \
>>>
>> _______________________________________________
>> buildroot mailing list
>> buildroot at busybox.net
>> http://lists.busybox.net/mailman/listinfo/buildroot
>