() Andy Wingo <wi...@pobox.com>
() Tue, 15 Jun 2010 23:07:47 +0200
Just for readability I have rebased the commits. The gnulib commit will
get overwritten at the next gnulib import. GUILE_CONFIG_SCRIPT is fine.
OK, thanks for cleaning up my mess.
It's good to have tmpfile, but I wonder about making the port-filename
not a string or #f. 'tmpfile seems too magical to me; if you're working
from scheme you can always associate a tmpfile object property on the
port, and I wouldn't want people to start asking if it's a tmpfile. So I
have changed it to return #f.
Why wouldn't you want people to start asking if it's a tmpfile?
People are curious.
If you really think that 'tmpfile is the right thing, let's talk about
it :)
One way to rationalize 'tmpfile is to consider #f to denote "invalid", in
which case #f for the ‘tmpfile’ (the port-returning proc) does not ring true.
On the other hand, until another file-port with non-string filename type
rears, i have no problem mentally keeping track of:
(port-filename P) => 'tmpfile
===
(and (file-port? P) (not (port-filename P)))
except that ‘file-port?’ does not exist in Guile 1.4.x (another reason to hoof
it over to official Guile :-).
The bigger question is (to touch upon a past discussion) the separation of
"file name" into "directory component + base name". If that ever comes to
pass (fundamentally), i think ‘port-filename’ won't mind transparently
passing the (richer) information to the user, with rv type most likely in
the set {#f, string, location (d+b)}, with some distinguished locations
symbolic. Then we can welcome the return of 'tmpfile.
thi
