Re: [PATCH] Re: Should we be avoiding "zstyle -m" ?

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

On Tue, Jan 9, 2024 at 1:17 AM Mikael Magnusson <mikachu@xxxxxxxxx> wrote:
>
> I don't understand what the \ does in your context argument for zstyle -m

That is in reference to url-quote-magic.

The more typical usage would be in completion, e.g., in _sysrc
  zstyle -m ":completion:${curcontext%:*}:values" cache-policy '*'
where the context argument is built up by the chain of calls below
_main_complete and actually means something well-defined

The \* in the url-quote-magic context is just there so that the part
after the last colon is neither empty nor a legal value for a URL
scheme.

> This makes me suspicious that maybe the code doesn't at all do what
> the author expected it to do?

It's expected to return false for e.g.
  zstyle 'zstyle ':url-quote-magic:http*' url-metas '+*?[]^(|)#{}='
because that context is more specific than
  zstyle ':url-quote-magic:*' url-metas '*?[]^(|)~#{}='
for which it should return true (and does as you found).

The idea is to install the least-specific context whenever the user
has not already defined that, so that the user doesn't need to define
contexts other than the most specific.  So it needs a context that
will never match something more specific.  The actual lookup, later,
always will be in a full context, so is expected to skip the
least-specific default when the user has defined something more
specific that matches.

> I am slightly more confident in my view after the following experiment
> (again a new shell):
> % zstyle ':url-quote-magic:\\*' url-metas '*?[]^(|)~#{}='
> % zstyle -m ':url-quote-magic:\*' url-metas '*' && echo true
> true
>
> So if I've set the url-metas style for the \\* context, the code will
> now override whatever my url-metas style was set for in the * context

No, it won't ... it would only override when zstyle -m returns false
-- the tests in url-quote-magic are all || rather than &&

This is a different bug in that asserting a style for the '...::\\*'
context ends up being equivalent to setting the default url-metas to
unset, but given there are no sensible contexts that contain a leading
backslash, this would never occur unless someone were deliberately
trying to break it.

> On a more general note, I think the whole approach is wrong and we
> should never set a style if we think it's empty; just use the fallback
> value when you look up the style and it was empty later on in the
> code, that will never have weird subtleties like this.

True, but it considerably muddies the code that actually needs the
style values.  In url-quote-magic, that also means pushing the
defaults for some styles down into helper functions so they
potentially end up being mentioned more than once, and it prevents
setting up the defaults at autoload time so they don't have to be
re-asserted every time the function runs.

This could be a bit less messy now by using namespace parameters to
hold the defaults, but at the time zstyle was the only
explictly-scoped global cache.