[RFC PATCH v2 00/17] Tidy command option parsing and use it a bit

[Simon Glass]

Hi Tom,

On Tue, 19 May 2026 at 19:06, Tom Rini <trini at konsulko.com> wrote:
>
> On Tue, May 19, 2026 at 05:31:26PM -0600, Simon Glass wrote:
>
> > We have had getopt() for over five years but it is not much used. It is
> > much easier to understand arg-parsing using getopt() and it avoids
> > common errors. As the named maintainer I decided to look at how to make
> > more use of it.
> >
> > So this series explores the impact of converting a few commands to use
> > getopt() instead of ad-hoc parsing. It updates getopt() to handle flags
> > anywhere in the cmdline and provides a few helpers to reduce
> > boilerplate.
> >
> > The chosen commands are:
> >
> > - echo: very simple with one flag and fixed ordering (worst case?)
> > - hash: fairly simple with just one flag
> > - env grep/export/import - fuller examples
> >
> > The series also adds a recommendation to use getopt() for new commands.
> >
> > To try to reduce the code-size increase, a lower-case function is added
> > and called from a few places. The difference is fairly marginal.
> >
> > v2 includes further attempts to reduce code size, by making reordering
> > a Kconfig option, dropping some printing and some intensive optimisation
> > of the env commands.
>
> Let us return to the comments from v1 again:
> https://lore.kernel.org/u-boot/20260515203311.2555651-1-sjg@chromium.org/
>
> Sean was saying that nvedit should be a best case for the conversion. In
> v1, grep was no change and "export and import" was 164 bytes on "qcom",
> which I chose because it already enabled GETOPT. Let us now look at v2.
> Here, grep, export and import are three commits:
> 10: cmd: nvedit: Use getopt() in env grep
> 11: cmd: nvedit: Use getopt() in env export
>    aarch64:  w+   qcom
> +(qcom)   636 |         char    buf[32];
> +(qcom)       |                 ^~~
> w+(qcom) cmd/nvedit.c: In function 'do_env_export':
> w+(qcom) cmd/nvedit.c:636:17: warning: unused variable 'buf' [-Wunused-variable]
>    aarch64: (for 1/1 boards) all +4.0 data +16.0 rodata -12.0
>             qcom           : all +4 data +16 rodata -12
>                u-boot: add: 0/0, grow: 0/-1 bytes: 0/-32 (-32)
>                  function                                   old     new   delta
>                  do_env_export                              608     576     -32
> 12: cmd: nvedit: Use getopt() in env import
>    aarch64: (for 1/1 boards) all +1.0 data +56.0 rodata -55.0
>             qcom           : all +1 data +56 rodata -55
>                u-boot: add: 0/0, grow: 1/0 bytes: 100/0 (100)
>                  function                                   old     new   delta
>                  do_env_import                              692     792    +100
>
> So grep is unchanged, export is a small reduction but import is growing,
> but less.
>
> My feedback is this. If you are going to bother with RFC v3, please just
> limit it to converting nvedit and showcasing that it's an improvement to
> convert. And if it's not, as Sean said in v1 we might need to revisit
> the interface.

It is an improvement in terms of the interface, but not really in code
size. I've fiddled with it quite a bit - I'm not saying it can't be
done but if you look at the code you can see that it is similar before
and after.

Any thoughts on this point?:

"One option would be to make the getopt_init_state() in command.c and
pass struct getopt_state to the commands. This would reduce code size
at each size, but obviously requires a different command signature,
unless a global is used."

We could do it piecemeal, i.e. cmd by cmd. Removing the init call and
passing gs instead of argc/v would improve things.

Regards,
Simon