Re: [PATCH v9 1/7] phy: can-transceiver: rename temporary helper function to avoid conflict

[Vladimir Oltean <olteanv@gmail.com>]

Hi Josua,

On Mon, Feb 16, 2026 at 08:19:27AM +0000, Josua Mayer wrote:
> >> In the future, when you have a series with cross-tree dependencies,
> >> please try to think of it as individual mini-series for each tree's
> >> 'next' branch, and specify clearly that you need stable tags (to be
> >> pulled into other trees).
> 
> I don't really understand how I could split my series up to avoid this 
> issue.
> 
> Due to the fact that one (and now two) drivers implemented local
> mux helpers, to undo that an atomic change must be made tree-wide.
> 
> Meanwhile it must be avoided that while the mux core helpers are being
> tested / reviewed, that any tree adds another driver-local mux helper
> like appears to have happened here.
> 
> Note that my patch-set did go to linux-phy@lists.infradead.org list, too.
> 
> The second challenge for this series was that mux framework is being
> enabled only by drivers Kconfig "select" - and not possible by menuconfig.
> This is e.g. responsible for being unable to test =m build with arm64
> defconfig - and lead to it only being detected through kernel robot
> x86_64 allmodconfig.

To avoid this, a combination of developer due diligence + maintainer due
diligence is probably required.

From linux-phy perspective, there will be some automated build testing
(which did not exist at the time of your submission). This would have
caught the 'hidden' devm_mux_state_get_optional() call present only in
linux-phy/next, when testing patch 2/7.

But, to work, the build automation needs to be able to apply the entire
patch set on linux-phy/next. So expect some pushback if it doesn't
(hence the recommendation to send a mini-series to linux-phy first, and
request a stable tag).

These are the tools we have, we need to find a way to make them work somehow.

Then there is the fact that local definitions of devm_mux_state_get_optional()
keep popping up, possibly in unrelated trees (not the case here). This seems
to be a bad practice which should be discouraged during review if caught.
Otherwise, some 'retries' will be required from the developer until all
occurrences are removed.

Note that the upcoming linux-phy automated build testing does have an
x86_64 allmodconfig test too.