On 12/12/2017 11:01 PM, Jonas Gorski wrote: > On 12 December 2017 at 21:03, Jo-Philipp Wich <j...@mein.io> wrote: >> Hi Piotr, >> >> my rough idea was to somehow tie the manifest generation to the "define >> Device/*" macros present in the image building code because there you >> have all required information in a central place: >> >> - unique board identifier >> - image name / build steps >> - default package selection > > I also think this is the easiest to achieve way for creating a way to > lookup board_name => image_name. > > We already define BOARD_NAME or SUPPORTED_DEVICES for many devices, we > just need to set these consistently for all targets. Then we can > easily create a manifest based on that. > > That way we also have no restrictions on how we name the images; what > is part of the images etc. > > > Regards > Jonas > > P.S: We should also deprecate one of these for the other, we don't > need two different variables for the same purpose.
Note that board_name -> image name does not always work on ar71xx, where multiple different "equivalent" models share the same board_name, but have their own image files with different magic numbers in the header (especially for TP-Link devices). Enforcing that incompatible images use different board_name would be a way to fix this, but it would increase redundancy in /etc/board.d/*. In Gluon (LEDE-based mesh framework) we have resorted to basing the image selection of our autoupdater on the "model" rather than the "board_name" (except for x86, where "model" is set arbitrarily, and we always select the same image). At the moment this is realized by a deterministic sanitization of the model name, which is then used as image name (we also have a manifest, so it is actually not a hard requirement of our solution that image name and model name match, just how we do it at the moment.) Regards, Matthias
signature.asc
Description: OpenPGP digital signature
_______________________________________________ Lede-dev mailing list Lede-dev@lists.infradead.org http://lists.infradead.org/mailman/listinfo/lede-dev
