Re: [Sbcl-devel] *default-pathname-defaults* and SB-POSIX

[Kevin R. <kp...@ma...>]

On Oct 12, 2009, at 5:43, Leslie P. Polzer wrote:

> I know that CLHS recommends this to be the directory in which
> the runtime has been started, but the current directory actually
> seems much more intuitive in a POSIX environment.
>
> Has there been a discussion in the past about making sb-posix:chdir
> change *default-pathname-defaults*? Would it be desirable to do so?

There are several mismatch issues here:

   1. CL has dynamic bindings, where the exact previous value is
      expected to be restored on exit.

   2. In POSIX the current directory is not a pathname but (roughly)
      an open directory, and pathname/directory associations are neither
      fixed nor one-to-one.

   3. *default-pathname-defaults* can be more than just a directory
      name. (I believe this is rarely used.)

It seems to me that the interactions of these two facts are  
sufficiently hairy that "just" changing both is a bad idea.

Two ideas for refinements:

1. Do it as a separate “high level” function (call it, say, sb- 
ext:change-directory) which both chdirs and assigns *d-p-d*.  
(Disadvantage: would interact surprisingly with rebinding *d-p-d*.)

2. An *option* which causes all changes of the binding of *d-p-d*  
(whether assignment, rebinding, or leaving a dynamic scope) to chdir,  
and possibly vice versa. This might perhaps also eliminate the actual  
storage of *d-p-d* in favor of "whatever getcwd(3) returns". This mode  
would not work for POSIX-in-all-its-generality (e.g. unreadable parent  
directories) but would seamlessly join the intersection of CL and  
POSIX, which would probably be greatly appreciated by those users  
which can work within it.

-- 
Kevin Reid                                  <http://switchb.org/kpreid/>