Re: [PATCH] New :P history modifier.

[Daniel Shahaf <d.s@daniel.shahaf.name>]

Bart Schaefer wrote on Wed, Aug 17, 2016 at 00:02:40 -0700:
> On Aug 16, 11:54pm, Daniel Shahaf wrote:
> }
> } WDYT?
> 
> The code changes are essentially OK; my only thought is, maybe we
> should just remove the dependency on the POSIX realpath() call even
> from :A, and use xsymlink() everywhere?
> 

Two thoughts:

- Does anyone actually build zsh on a system that doesn't have realpath()?

- xsymlink() is not a drop-in replacement: it tolerates trailing
  non-existing path components.  The single callsite in current master
  wouldn't care, though.

> Picking at the docs ...
> 
> Given that we went to the trouble of hashing it out, it is probably
> worth noting that :a is intended to result in the path to along which
> "cd" would change under the default setopts (no_chase_dots),

38945 made such a change; do you think further changes are needed?

> and :A is meant to result in the physical directory at the end of that
> path.

No wording comes to mind, here.

> 
> } +	"a:absolute path, resolve '..' logically"
> } +	"A:same, then resolve symlinks"
> } +	"P:realpath, resolve '..' physically"
> 
> I'm not sure "logical" and "physical"
> are the right words here, but "by text replacement" and "by filesystem
> reference" seem a bit too verbose, so I don't have a suggestion, just
> calling attention.

They might work as parentheticals: "a:... logically (by text
replacement)" and "P:... physically (by filesystem lookup)"?

Some other options:

- :a   syntactically / lexically / "like 'cd'"
- :P   semantically / "like stat(2)"

(yes, "like 'cd'" is inaccurate if CHASE_* are set)

> You should spell out what "same" means, because the two descriptions
> may not always appear together.

I'll change "same" to "as ':a'".

I wonder if there's a way to make the :A completion text convey "Unless
you're trying to predict what 'cd' is about to do, you probably want :P,
not :A"...

> } +The new word modifier ':P' computes the realpath() of the argument.
> } +It is different from the existing ':a' modifier which does not resolve
> } +symlinks, and from the existing ':A' modifier which always resolves
>    ^^^^^^^^
> 
> If you're going to compare to both :a and :A, symlinks aren't the largest
> difference vs. :a -- following ".." is.  In fact I'd say following ".."
> is more important to the distinction than symlinks are.
>
> } +/before/here/../after to /before/after --- even if /before/here is
> } +a symbolic link.  It is recommended to review uses of ':A' and, if
> } +appropriate, convert them to ':P' as soon as compatibility with 5.2 is
> } +no longer a requirement.
> 
> So how about e.g.:
> 
>  The new word modifier ':P' computes the physical path of the argument.
>  it is different from the existing ':a' modifier which does always
>  resolves /before/here/../after to /before/after, and differs from the
>  existing ':A' modifier which resolves symlinks only after here/.. is
>  removed even when /before/here is itself a symbolic link.
> 

Looks good to me.

What about the "It is recommended [to audit uses of :A and change them
to :P]" sentence that the original patch had, should it be kept or
removed?

> That's all.

Thanks for the review!

Daniel