[meta-ti] checksum and path issues building gstreamer-ti

[Peter Bigot]

On Sun, Nov 13, 2011 at 10:59 PM, Denys Dmytriyenko <denis at denix.org> wrote:
> On Sun, Nov 13, 2011 at 09:37:47PM -0600, Peter Bigot wrote:
>> I didn't get an answer to this when I asked it as an aside on the
>> BeagleBoard google group, but just verified that the same problems are
>> present in oe-classic as is in oe-core.
>>
>> The recipe checksums don't match the current downloads:
>>
>> vmserver[52]$ cd Downloads/
>> vmserver[53]$ md5sum ti_cgt_c6000_6.1.17_setup_linux_x86.bin
>> 5ee5c8e573ab0a1ba1249511d4a06c27 ti_cgt_c6000_6.1.17_setup_linux_x86.bin
>> vmserver[54]$ sha256sum ti_cgt_c6000_6.1.17_setup_linux_x86.bin
>> 0cb99e755f5d06a74db22d7c814e4dfd36aa5fcb35eeab01ddb000aef99c08c1
>> ti_cgt_c6000_6.1.17_setup_linux_x86.bin
>>
>> When that's corrected still doesn't work because do_unpack puts the
>> toolchain in  /opt/TI/TI_CGT_C6000_6.1.17/ instead of under
>> ${installdir} where ti-paths.inc expects it.
>>
>> NOTE: package ti-cgt6x-1_6_1_17-r4: task do_install: Started
>> ERROR: Logfile of failure stored in:
>> /prj/oe/oe-dev/build/tmp/work/beagleboard-angstrom-linux-gnueabi/ti-cgt6x-1_6_1_17-r4/temp/log.do_install.20024
>> Log data follows:
>> | + do_install
>> | + install -d /prj/oe/oe-dev/build/tmp/work/beagleboard-angstrom-linux-gnueabi/ti-cgt6x-1_6_1_17-r4/image/usr/share/ti/ti-cgt6x-tree
>> | + cp -pPrf '/prj/oe/oe-dev/build/tmp/work/beagleboard-angstrom-linux-gnueabi/ti-cgt6x-1_6_1_17-r4/cgt6x_6_1_17/*'
>> /prj/oe/oe-dev/build/tmp/work/beagleboard-angstrom-linux-gnueabi/ti-cgt6x-1_6_1_17-r4/image/usr/share/ti/ti-cgt6x-tree
>> | cp: cannot stat
>> `/prj/oe/oe-dev/build/tmp/work/beagleboard-angstrom-linux-gnueabi/ti-cgt6x-1_6_1_17-r4/cgt6x_6_1_17/*':
>> No such file or directory
>> NOTE: package ti-cgt6x-1_6_1_17-r4: task do_install: Failed
>>
>> For the version in meta-ti, the LIC_FILES_CHKSUM value was reverted to
>> a wrong checksum when it got moved to the recipe file in commit
>> 7feec8987.
>>
>> Or maybe somebody at TI updated
>> ti_cgt_c6000_6.1.17_setup_linux_x86.bin and the recipe is now broken
>> for anybody who doesn't have a cached copy of a previous release.  I'm
>> going to scrap this and move on, but it'd probably be worth verifying
>> that current Angstrom (classic or core) works with current downloads
>> of the codegen tools.
>
> Yes, the binary download of CGT 6.1.17 was updated, unfortunately:
>
> https://lists.yoctoproject.org/pipermail/meta-ti/2011-October/000016.html

Thanks.  Has anybody successfully built with the new binary?  My
experiments suggest correcting the MD5 isn't enough; the installation
process also needs to be modified in non-obvious ways.

Peter

> And we did follow up with CGT team. As one of the options to handle this
> situation more gracefully, I pointed out at the way FSF handled it for when
> they needed to update all previous releases of binutils - they made available
> the updated binaries with a suffix.
>
> Anyway, we need to ping them again and see if this can be done similarly with
> CGT updates...
>
> Roger/Jason,
>
> Would you mind?
>
> --
> Denys
>