Re: [PATCH v2] i2c-tools: Allow passing device file paths

["Brigham Campbell" <me@brighamcampbell.com>]

Hi Gero,

On Tue Apr 28, 2026 at 5:43 AM MDT, Gero Schwäricke wrote:
> I see that you're struggling to find a good way to pass the device
> information between lookup_i2c_bus() and open_i2c_dev(). How about we
> lookup and open the device in one go? Exit paths will have to close the
> file, but the rest should stay the same. It's worth to take a look at
> how libgpio does the same, see gpiod_chip_open_lookup().

I had worried that doing the device path lookup and opening in one go
would be overly invasive. Of course, if I can make that work, then it
would address the concerns that I brought up in my previous email. It
may be worth the extra effort to avoid the ugly variant struct I
introduced in this v2.

I'll take a look at libgpio. Thanks.

> Also, to keep things backward compatible, we should presumably do the
> path lookup only after bus number parsing and bus name lookup failed.

This makes sense. In the strange (but totally plausible) case of
conflicts between bus number, bus name, and bus file/symlink path, the
order of precedence should be as follows to maintain backwards
compatibility:

bus number -> bus name -> bus device path

I'll fix this in the next revision.

> Symlink support would be welcome as well.

As a matter of fact, this patch already supports symlinks. Passing
symlinks generated by udev to i2c{get,set} is the use case I'd like this
patch to support.

Thanks,
Brigham