Re: [Sbcl-bugs] make-pathname fails when #\* in name

[Martin M. <ma...@ma...>]

Often Christophe, Email does not quite communicate the intent!
I did not mean anything by the phrase "of course" .. it was just my way
of saying it.
No disparagement intended...

Another way that I can express this is that I expected that this
identity would hold

(let ((pn (make-pathname :name "test*" :type "dat"))) (equalp pn
(pathname (namestring pn))))
=> T on ABCL and ACL, NIL on SBCL

Consequently we can see that SBCL is somehow not consistently managing
the presence of the * in the name slot.
In fact it is parsing the name into a directory as we can see if we do this:
This is SBCL:

* (let ((pn (make-pathname :name "test*" :type "dat"))) (describe pn)
(describe (pathname (namestring pn))))

#P"test\\*.dat"
  [structure-object]

Slots with :INSTANCE allocation:
  HOST       = #<SB-IMPL::WIN32-HOST {22472481}>
  DEVICE     = NIL
  DIRECTORY  = NIL
  NAME       = "test*"
  TYPE       = "dat"
  VERSION    = NIL
#P"test/*.dat"
  [structure-object]

Slots with :INSTANCE allocation:
  HOST       = #<SB-IMPL::WIN32-HOST {22472481}>
  DEVICE     = NIL
  DIRECTORY  = (:RELATIVE "test")
  NAME       = :WILD
  TYPE       = "dat"
  VERSION    = :NEWEST


This for comparison is ACL:

cl-user(54): (let ((pn (make-pathname :name "test*" :type "dat")))
(describe pn) (describe (pathname (namestring pn))))
#P"test*.dat" is a structure of type pathname.  It has these slots:
  excl::host           nil
  excl::device         nil
  directory            nil
  excl::name           "test*"
  type                 "dat"
  excl::version        :unspecific
  namestring           "test*.dat"
  excl::hash           nil
  excl::dir-namestring nil
  excl::plist          nil
#P"test*.dat" is a structure of type pathname.  It has these slots:
  excl::host           nil
  excl::device         nil
  directory            nil
  excl::name           "test*"
  type                 "dat"
  excl::version        :unspecific
  namestring           "test*.dat"
  excl::hash           nil
  excl::dir-namestring ".\\"
  excl::plist          nil

Best wishes,
Martin M

On 2/17/2014 8:19 AM, Christophe Rhodes wrote:
> Martin Mallinson <ma...@ma...> writes:
>
>> Here is a little more about this issue.  Run this little code example to
>> see the difference:
>>
>> (progn
>>     (with-open-file (s "test00.dat" :direction :output :if-exists
>> :supersede))
>>     (equalp (directory "test*.dat") (directory (make-pathname :name
>> "test*" :type "dat"))))
>>
>> You will see that it returns NIL on SBCL and returns T on other Common
>> LISP implementations, it should of course return T.
> Why do you say "it should of course"?
>
> The object created from (pathname "test*.dat") involve the parsing of a
> namestring into a pathname; that parsing happens in an
> implementation-defined way, and SBCL defines it to parse in a relatively
> friendly way the asterisk into the equivalent of a shell glob.  If you
> request the PATHNAME-NAME of #p"test*.dat", you should get an
> SB-IMPL::PATTERN object.
>
> By contrast, MAKE-PATHNAME in SBCL does no parsing at all; the pathname
> object it constructs has a PATHNAME-NAME being equal to "test*".
>
> When the filesystem is interrogated about these files, the first one
> (with a PATTERN) is wild, and so matching rules are applied to find all
> those files matching the pattern.  The second, with the string
> PATHNAME-NAME, is not a wild pathname; it will match exactly the file
> whose five-character name is "test*" (and whose extension is "dat"), and
> no other files.
>
>> It is not that the * is in the name, it is in the name in both cases, it
>> is that some backslashes seem to get in there as well in SBCL.
> The backslash is an artifact of print-read consistency requirements: if
> you print a pathname, reading it back in must produce the same.  Since
> #p"test*.dat" is namestring syntax for a pattern, there must be an
> alternative syntax for the name with a literal asterisk; that
> alternative syntax escapes it with a backslash, thus: "test\\*".  (The
> backslash is doubled because in string syntax, the backslash is the
> escape character; the string "test\\*" has six character elements: #\t,
> #\e, #\s, #\t, #\\, #\*.
>
> Best wishes,
>
> Christophe
>