Re: [PATCH v1 3/4] advice: remove uses of global `advice_` variables

[Ben Boeckel <mathstuf@gmail.com>]

On Tue, Aug 03, 2021 at 00:06:48 +0200, Johannes Schindelin wrote:
> On Fri, 30 Jul 2021, Ben Boeckel wrote:
> > There are now function APIs to access this information, so the global
> > variables are no longer needed to communicate their values.
> 
> This commit message implies that the reader remembers that `hw/advise-ng`
> introduced a new advice API with the express intent to eventually replace
> those global `advice_*` variables.
> 
> However, it is not the responsibility of the reader to remember that. It
> is the responsibility of the commit message to describe this (so that the
> reader can either remember it, or learn about it in the first place).

That makes sense, thanks. I guess I'm used to projects where I'm at
least peripherally aware of most of what's going on, but that's because
I work on cross-cutting concerns on them for the most part (build
systems, CI, software process).

> Now, as for the diff, I can guess just how tedious all of the
> semi-repetitive `advice_*` -> `advice_enabled(ADVICE_*)` changes were from
> trying to verify that they are all correct.

One of the times a case-insensitive word diff rendering would be handy.
(Then letting the compiler verify that the new side actually works.)

> Big thank you for that!

You can thank this Vim macro which made the tedious bits trivial:

    iadvice_enabled(<Esc>wgUwea)<Esc>]q

> However, even reading through the diff the second time, I managed to miss
> the subtlety that there were two `advice_set()` calls strewn in.
> 
> As I pointed out in my review of patch 1/4, I would much prefer to have
> the addition of those callers in 1/4 along with the introduction of said
> function.
> 
> However, now that I write this, I would like to correct my advice (pun
> intended) from 1/4 to leave the removal of the assignment of the global
> `advice_*` variables here: It would make much more sense to move them to
> patch 4/4.

Sounds good. I'll still keep it as a separate patch, but just have one
for the read-only side, then a new patch which adds the write API and
transforms the 2 write instances. The final patch can then stay the
same. In short:

  - 2/4 (read-only bits) -> v2 1/4
  - 3/4 -> v2 2/4
  - 1/4 + 3/4 (write bits) -> v2 3/4
  - 4/4 -> mostly the same

Thanks,

--Ben