Re: Filename generation: instruct zsh not to follow symbolic links?

[Mikael Magnusson <mikachu@xxxxxxxxx>]

On Fri, Jul 17, 2026 at 12:10 PM Vincent Lefevre <vincent@xxxxxxxxxx> wrote:
>
> On 2026-07-17 02:47:20 +0200, Mikael Magnusson wrote:
> > On Fri, Jul 17, 2026 at 1:36 AM Vincent Lefevre <vincent@xxxxxxxxxx> wrote:
> > >
> > > On 2026-07-16 18:04:54 +0200, Mikael Magnusson wrote:
> > > > On Thu, Jul 16, 2026 at 11:35 AM Vincent Lefevre <vincent@xxxxxxxxxx> wrote:
> > > > >
> > > > > On 2026-07-16 02:57:46 +0200, Mikael Magnusson wrote:
> > > > > > On Thu, Jul 16, 2026 at 1:47 AM Vincent Lefevre <vincent@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > In filename generation, is there a way to instruct zsh not to follow
> > > > > > > symbolic links?
> > > > > > >
> > > > > > > Unfortunately, it is not possible to use glob qualifiers on
> > > > > > > pathname components (except the last one, i.e. ending the pattern),
> > > > > > > e.g. *(/)/foo, even though there would be no ambiguity in parsing
> > > > > > > (because a slash cannot appear in a pathname component).
> > > > > >
> > > > > > No, but you can do things like
> > > > > > () { mycmd $^@/foo } *(/)
> > > > > > which conveniently doesn't need any named temporary parameters.
> > > > >
> > > > > I get "No such file or directory" errors for some directories.
> > > > > But this is too complex anyway, and perhaps not usable in some
> > > > > situations.
> > > >
> > > > Right, you might need
> > > > () { setopt localoptions cshnullglob; mycmd $^@/foo } *(/)
> > > > :)
> > >
> > > I still get "No such file or directory" errors; or if I use "echo"
> > > as the command, I get non-existing files in the output.
> >
> > It was implied that foo is a pattern in the above, in which case the
> > cshnullglob wil filter out the nonexistant ones.
>
> In my case, "foo" is not a pattern. The goal was to get all the
> files named "foo" in each subdirectory (which are not symlinks).

In that case you can for example do this:
() { setopt localoptions cshnullglob; mycmd $^@/foo(#q) } *(/)
the #q does nothing here, except make the word a glob pattern.

> > > > > I would rather think that a (/ sequence could have a special meaning.
> > > > > That should be easy to parse.
> > > >
> > > > Sure, but that'd be an extreme special case then. You wouldn't be able
> > > > to use any other glob qualifiers there.
> > >
> > > Why not?
> >
> > Because the other glob qualifiers are not '/' and hence will not match
> > the (/ sequence.
>
> This is not true:
>
> qaa% mkdir a b
> qaa% chmod 500 b
> qaa% ls -l
> total 8
> drwxr-xr-x 2 vinc17 vinc17 4096 2026-07-17 11:35:03 a
> dr-x------ 2 vinc17 vinc17 4096 2026-07-17 11:35:03 b
> qaa% ls -ld *(/)
> drwxr-xr-x 2 vinc17 vinc17 4096 2026-07-17 11:35:03 a
> dr-x------ 2 vinc17 vinc17 4096 2026-07-17 11:35:03 b
> qaa% ls -ld *(/w)
> drwxr-xr-x 2 vinc17 vinc17 4096 2026-07-17 11:35:03 a

'w' is not a '/', so I'm not sure what this is supposed to
demonstrate? How would you do the opposite of what you're currently
trying to do, finding foo only in symlinked directories? (^/ and (@
both don't start with (/.

-- 
Mikael Magnusson