Steel Bank Common Lisp

Russell McManus <russell_mcmanus@...> writes:

> While writing some code using run-program, I came upon the need for
> the following:
>
> (defmacro with-current-directory ((new-dir) &body body)
>   (let ((start (gensym)))
>     `(let ((,start (sb-unix:posix-getcwd)))
>       (unless (char= #\\ (aref ,start (1- (length ,start))))
> 	(setf ,start (concatenate 'string ,start "/")))
>       (setf ,start (parse-namestring ,start))
>       (unwind-protect
> 	   (progn
> 	     (sb-posix:chdir ,new-dir)
> 	     ,@body)
> 	(sb-posix:chdir ,start)))))
>
> Is there an already existing way to do this sort of thing?

No.  There is at least a plausible argument that RUN-PROGRAM should
respect *DEFAULT-PATHNAME-DEFAULTS* by chdir()ing to the directory
component, which would allow the user to request a different directory
for run-program simply by binding it.  On the other hand, RUN-PROGRAM
is already a deeply scary mess of options, and adding another doesn't
really appeal...

Cheers,

Christophe

Christophe Rhodes <csr21@...> writes:

> Russell McManus <russell_mcmanus@...> writes:
>
>> While writing some code using run-program, I came upon the need for
>> the following:
>>
>> (defmacro with-current-directory ((new-dir) &body body)
>>   (let ((start (gensym)))
>>     `(let ((,start (sb-unix:posix-getcwd)))
>>       (unless (char= #\\ (aref ,start (1- (length ,start))))
>> 	(setf ,start (concatenate 'string ,start "/")))
>>       (setf ,start (parse-namestring ,start))
>>       (unwind-protect
>> 	   (progn
>> 	     (sb-posix:chdir ,new-dir)
>> 	     ,@body)
>> 	(sb-posix:chdir ,start)))))
>>
>> Is there an already existing way to do this sort of thing?
>
> No.  There is at least a plausible argument that RUN-PROGRAM should
> respect *DEFAULT-PATHNAME-DEFAULTS* by chdir()ing to the directory
> component, which would allow the user to request a different directory
> for run-program simply by binding it.  On the other hand, RUN-PROGRAM
> is already a deeply scary mess of options, and adding another doesn't
> really appeal...

I agree about *DEFAULT-PATHNAME-DEFAULTS*, and indeed I tried this
first, hoping that it would "just work".  I would be willing the
change run-program to work in this way: is this of interest?  The
reason I didn't do this first is because I wasn't sure whether there
was an existing way to achieve what I wanted.

I agree that this probably should not be an additional keyword option
to RUN-PROGRAM.

If a patch to RUN-PROGRAM is not desired, then I put the above code in
the public domain, hopefully to be included with sbcl.

-russ

Today, Russell McManus <russell_mcmanus@...> wrote:
>
> While writing some code using run-program, I came upon the need for
> the following:

autobench uses a similar macro, so I have a few questions about your
code.

One, wouldn't it be better if the macro also bound
*default-pathname-defaults* to the new-dir? (My implementation assumes
that the user(me) will consistently use with-current-directory and so
skips the whole getcwd thing in favor of just chdir-ing back into
*d-p-defaults*)

> (defmacro with-current-directory ((new-dir) &body body)
>   (let ((start (gensym)))
>     `(let ((,start (sb-unix:posix-getcwd)))
>       (unless (char= #\\ (aref ,start (1- (length ,start))))
> 	  (setf ,start (concatenate 'string ,start "/")))

Two, what's with that #\\ check? In which cases does getcwd return the
\ character as the last element?

>       (setf ,start (parse-namestring ,start))
>       (unwind-protect
> 	   (progn
> 	     (sb-posix:chdir ,new-dir)
> 	     ,@body)
> 	(sb-posix:chdir ,start)))))

Cheers,
-- 
Andreas Fuchs, <asf@...>, asf@..., antifuchs

* Russell McManus:

> 	(sb-posix:chdir ,start)))))

I think you should use fchdir instead because it works in more cases
(deleted directory, unmount is prevented).

Florian Weimer <fw@...> writes:

> * Russell McManus:
>
>> 	(sb-posix:chdir ,start)))))
>
> I think you should use fchdir instead because it works in more cases
> (deleted directory, unmount is prevented).

I'm not a unix guru; how do you recommend to acquire the necessary fd
to pass to fchdir?

-russ

* Russell McManus:

>> * Russell McManus:
>>
>>> 	(sb-posix:chdir ,start)))))
>>
>> I think you should use fchdir instead because it works in more cases
>> (deleted directory, unmount is prevented).
>
> I'm not a unix guru; how do you recommend to acquire the necessary fd
> to pass to fchdir?

Ah, this is somewhat messy.  According to POSIX, the open syscall
would do it (in O_RDONLY mode), but unfortunately it breaks on Linux,
where O_DIRECTORY is required.

Maybe it's not such a great idea to use fchdir after all.

Andreas Fuchs <asf@...> writes:

> Today, Russell McManus <russell_mcmanus@...> wrote:
>>
>> While writing some code using run-program, I came upon the need for
>> the following:
>
> autobench uses a similar macro, so I have a few questions about your
> code.
>
> One, wouldn't it be better if the macro also bound
> *default-pathname-defaults* to the new-dir? (My implementation
> assumes that the user(me) will consistently use
> with-current-directory and so skips the whole getcwd thing in favor
> of just chdir-ing back into *d-p-defaults*)

I can easily be convinced that *default-pathname-defaults* should be
magical in some sneaky way.  I just threw together the macro that I
wrote based on an immediate need, and wondered whether someone else
had a fully baked solution.

>> (defmacro with-current-directory ((new-dir) &body body)
>>   (let ((start (gensym)))
>>     `(let ((,start (sb-unix:posix-getcwd)))
>>       (unless (char= #\\ (aref ,start (1- (length ,start))))
>> 	  (setf ,start (concatenate 'string ,start "/")))
>
> Two, what's with that #\\ check? In which cases does getcwd return the
> \ character as the last element?

This is an error in my code, the constant should be #\/, because
getcwd sometimes returns a path without a trailing slash, which when
parsed with parse-namestring will put the last path component into the
pathname's name field rather than the final component in the directory
field.

-russ