[U-Boot] [PATCH 00/19] fdt: Move to the new upstream pylibfdt library

[Simon Glass]

Hi Tom,

On 18 April 2017 at 10:33, Tom Rini <trini at konsulko.com> wrote:
> On Tue, Apr 18, 2017 at 10:27:37AM -0600, Simon Glass wrote:
>> Hi Tom,
>>
>> On 18 April 2017 at 10:25, Tom Rini <trini at konsulko.com> wrote:
>> >
>> > On Sun, Apr 16, 2017 at 08:22:14PM -0600, Simon Glass wrote:
>> >
>> > > Python libfdt bindings have recently been accepted upstream. While the
>> > > internals have changed a fair bit most of the API remains the same. Still,
>> > > a few functions are different from how they are used in U-Boot so changes
>> > > are needed to make this work.
>> > >
>> > > At present in U-Boot there are two libraries for accessing a device tree
>> > > file:
>> > >
>> > > - FdtNormal which uses U-Boot's own Python bindings
>> > > - FdtFallback which uses the fdtget command-line utility
>> > >
>> > > The latter is not a great solution: it is fairly slow since the DT is
>> > > re-read for every access and it cannot provide DT offsets or packing of
>> > > the DT.
>> > >
>> > > In addition, U-Boot now builds the libfdt module if swig is available,
>> > > meaning that the fallback module is not used in that case.
>> > >
>> > > Finally, at some point in the future distributions may start packaging the
>> > > libfdt Python module and it will be available without U-Boot needing to
>> > > build it itself.
>> > >
>> > > Therefore it seems like a good idea to take this opportunity to drop the
>> > > fallback module and just require that the Python libfdt bindings be
>> > > present (at least if need by the build).
>> > >
>> > > The bindings are needed in two situations:
>> > > - When dtoc is used to convert a device tree into C code. This is enabled
>> > >     by CONFIG_SPL_OF_PLATDATA
>> > > - When binman is used to produce a firmware image. This is used on all x86
>> > >     and sunxi boards at present
>> > >
>> > > This series:
>> > > - Plumbs in building the Python libfdt module to the U-Boot build system
>> > > - Ensures that the module is always built if needed, print an error if
>> > >     swig is not available (and thus the module cannot be built)
>> > > - Allows use of a libfdt.py module already installed on the machine
>> > > - Drops the FdtFallback support
>> > > - Moves fdt.h and libfdt.h into lib/libfdt to aid with syncing with
>> > >     upstream, building the Python bindings and to keep the code together
>> > > - Merges Fdt and FdtNormal to simplify the code
>> > > - Adjusts the Fdt library to work with the new libfdt module
>> > > - Adds a few more tests to check access to properties in the DT
>> > > - Adjusts binman and dtoc to work with the new approach
>> > >
>> > > It should be possible to easily sync libfdt's Python bindings with U-Boot
>> > > in the future, as development there proceeds.
>> >
>> > While this came in late, my gut feeling is that it would be best to have
>> > this change in the next release (so that various upstreams can get used
>> > to the idea of basically always needing python installed to build).
>>
>> I'm a little worried that it might cause problems. I made it so that
>> it only *requires* python if is it actually needed, though it always
>> builds libfdt.py if it can. Or perhaps that is what you are saying,
>> that the only way to flush out upstream issues is to apply it?
>
> What I'm saying is that distros (Fedora, openSUSE, OpenEmbedded) are
> going to make a blanket change (if they haven't already) to say that
> building U-Boot requires python to be installed (since they aren't
> likely to try and optimize it on a board-by-board basis when it's "easy"
> to just say it's an always dependency).  If you think there's risk with
> the change itself, we can wait a release, that's fine too.  Thanks!

I see. I'm not sure what is best. I checked for build errors and added
some more tests so I'm fairly confident. But it is late (sorry, it was
more work than I expected). I'll leave this up to you.

>
> --
> Tom

Regards,
Simon