[PATCH] binman: Avoid requiring a home directory on startup

[Simon Glass]

Hi Quentin,

On Fri, 17 Feb 2023 at 05:21, Quentin Schulz
<quentin.schulz at theobroma-systems.com> wrote:
>
> Hi all,
>
> On 2/17/23 03:55, Simon Glass wrote:
> > Hi Tom,
> >
> > On Thu, 16 Feb 2023 at 17:19, Tom Rini <trini at konsulko.com> wrote:
> >>
> >> On Thu, Feb 16, 2023 at 05:12:33PM -0700, Simon Glass wrote:
> >>> Hi Tom,
> >>>
> >>> On Tue, 14 Feb 2023 at 13:27, Tom Rini <trini at konsulko.com> wrote:
> >>>>
> >>>> On Tue, Feb 14, 2023 at 03:12:46PM -0500, Mike Frysinger wrote:
> >>>>> On Tue, Feb 14, 2023 at 3:08 PM Tom Rini <trini at konsulko.com> wrote:
> >>>>>> Downloading things from the internet and putting them in to the default
> >>>>>> PATH always and forever is also kinda not great?
> >>>>>
> >>>>> you just described a standard distribution.  this is like literally
> >>>>> how all of them work.  not to mention every other language-specific
> >>>>> distro tool out there (e.g. Python pip, Perl cpan, Go, etc...).
> >>>>>
> >>>>> maybe you'd like more guarantees on top (e.g. signature verification)
> >>>>> which is reasonable.
> >>>>>
> >>>>> but to be clear, this script is already merged & in the tree, so your
> >>>>> feedback doesn't block this patch.
> >>>>
> >>>> Yes, exactly. This is a fix on top of what we do today, so it should go
> >>>> in. But modern distributions only install signed packages, and
> >>>> language-specific tools tend to be a hive of bad examples. Looking over
> >>>> binman right now, I see that we're either using apt (and oh, there's
> >>>> "aot" typo in one spot) or downloading from a known Google drive, for
> >>>> only a few less common tools.
> >>>>
> >>>> So yes, I would like to see some ideas on how to improve things in the
> >>>> future so we aren't putting the binaries somewhere that's not a default
> >>>> (or frequently common) PATH location.
> >>>
> >>> Are you thinking they should go in ~/.binman-tools or something like
> >>> that? Then we would need to tell people to add it to their path. But
> >>> we could make binman look there automatically.
> >>
> >> We should document that it's where we're putting stuff, not so much
> >> "tell" them, unless you mean as a note when downloading.  But yes,
> >> ~/.binman-tools sounds reasonable.  Maybe a flag to point elsewhere?
> >
> > OK I will take a look.
> >
>
> I think this should be directly put into the output/build directory used
> by U-Boot, because what happens when you have two U-Boot git repos with
> different version requirements for those host tools? Then you need to
> make sure you're not building both at the same time, that you update
> them properly before each build, etc.

My advice: *Don't do that*

So far as binman is concerned, a tool is a tool. Tools should be
backwards compatible so updating to the new one should fix all the
problems.

The problem with using the output dir is we then have to download them
for each build, or cache them somewhere. To my mind, the 'binman tool'
feature is a convenience to reduce the pain involved in obtaining
tools needed to build. It is a not a panacea for strange situations.

>
> We also have an issue with buildman creating (image) temporary files in
> the current git repo which makes people unhappy because they then get a
> dirty tree because we tend to forget to update the .gitignore. What
> about building out-of-tree by default like Buildroot does? e.g. with
> make O=$PWD/output/ ?

I don't tend to use in-tree builds, but, yes, this is a problem.

I believe the best solution is, I think, to make binman put its
intermediate files in a subdir. At present there is no distinction
between output files and files produced along the way for debugging
purposes. This has been on my mind for a while and has been discussed
at least once before, too.

We do want the output files to be in the build directory. It would be
annoying to put u-boot.itb in a subdir.

Binman should all track all the files it produced (both final output
and intermediate) so that info is available.

For now, the solution is to use 'make O=xxx' as you say. With that,
binman should not write anything to the current dir. If it does, that
is a bug.

Regards,
Simon