search

OE4T / meta-tegra

BSP layer for NVIDIA Jetson platforms, based on L4T
MIT License
413 stars 227 forks source link

NVIDIA_DEVNET_MIRROR requirements for core-image-base builds #292

Closed dwalkes closed 4 years ago

dwalkes commented 4 years ago

Hi @madisongh

This relates to discussion in #284

I think it would be great if people can get through mender setup with core-image-base without needing to mess with nvidia download tools, but if they can't I suppose should update the documentation to provide a link to the instructions about this. I was preparing the zeus meta-mender-community branch and noticed zeus-l4t-r32.3.1 build fails without NVIDIA_DEVNET_MIRROR defined.

ERROR: Nothing PROVIDES 'tegra-mmapi-glheaders' (but /home/dan/mender-tegra/build/../sources/meta-tegra/recipes-graphics/libglvnd/libglvnd_1.2.0.bb DEPENDS on or otherwise requires it)
tegra-mmapi-glheaders was skipped: Recipe requires NVIDIA_DEVNET_MIRROR setup

ERROR: Required build target 'core-image-base' has no buildable providers.
Missing or unbuildable dependency chain was: ['core-image-base', 'packagegroup-base-extended', 'ofono', 'bluez5', 'python3-pygobject', 'cairo', 'virtual/libgles2', 'tegra-mmapi-glheaders']

However, cherry-picking https://github.com/madisongh/meta-tegra/commit/fb4f2fb3346b7f26e1a4296dd7c48d1333d88ee3 onto the zeus-l4t-r32.3.1 branch of meta-tegra resolves the issue and I can build core-image-base without NVIDIA_DEVNET_MIRROR on both jetson-tx2 and jetson-nano-qspi-sd which are the two officially supported mender platforms.

Do you think it makes sense to plan to support core-image-base + mender without NVIDIA_DEVNET_MIRROR on warrior and later branches and, if so, would it be possible to get https://github.com/madisongh/meta-tegra/commit/fb4f2fb3346b7f26e1a4296dd7c48d1333d88ee3 cherry-picked into the zeus-l4t-r32.3.1 branch or would you like a new PR for this?

Thanks!

madisongh commented 4 years ago

Yeah, I held off on back-porting fb4f2fb for a couple of reasons - I wanted to be sure that builds checked out OK on master, and I didn't want to trigger massive rebuilds because of the change.

Builds on master seem to be checking out OK, and since this would generally be a desirable improvement, that probably outweighs the rebuild issue. I'll go ahead and back-port the change.