On 2022-04-07 14:28, Maxime Devos wrote:

> Andrew Tropin schreef op do 07-04-2022 om 11:22 [+0300]:
>> +         (define (no-follow-file-exists? file) 
>> +           "Return #t if file exists, even if it's a dangling
>> symlink."
>> +           (or (file-exists? file)
>> +               (and=> (false-if-exception (lstat file))
>> +                      (lambda (x)
>> +                        (equal? (stat:type x) 'symlink)))))
>
> Can't this be simplified to
>
>   (define (no-follow-file-exists? file)
>     (false-if-exception (lstat file)))
>

Idk how file-exists? works internally, but still expect it to be more
efficient than lstat.  That's why I decided to use lstat only as a
"fallback" option in `or` statement.

> ?  Also, do you want to ignore _all_ exceptions, or only the ENOENT and
> maybe ENOTDIR system-error?
>
> (catch 'system-error
>   (lambda () (lstat file) #t)
>   (lambda e
>     (if its-a-ENOFILE
>         #f
>         (apply throw e))))
>
> More concretely, why is ENOMEM, ENAMETOOLONG and EACCESS ignored here?

You are right, we are interested only in ENOENT here.  AFAIK, ENOMEM
shouldn't happen here, but some other can, still we do not handle them
and make function behave the same way as file-exists? do and just return
#f in such cases.  However, I'm not sure if file-exists? is a good
example to follow.

Anyway in current implementation handling other codes in
no-follow-file-exists? will not save the day, in case of EACCESS it
doesn't really matter if it will be thrown during the check of existence
or during creation of symlink.  But we probably can later rework the
whole symlinking process: check that we have premissions to all files,
we are interested in, and only after that cleanup, backup and symlink.

>
> Greetings,
> Maxime.

-- 
Best regards,
Andrew Tropin