Use ENSURE-PATHNAME on its argument.

These from Spencer Truex, with tests by me.

Neil found that the pathname argument would be misinterpreted if it was a directory pathname without a trailing slash.
Applied his fix and added a test.

These changes require a large set of patches to Genera which are, presently, only
available in the world used by the Intel and ARM versions of Open Genera called
Portable Genera.  (If there's ever a new release of Genera, ...)

Elias Pipping authored Sep 21, 2016 and Robert Goldman committed Oct 05, 2016

In particular:
 - clasp does not define :ecl
 - use :os-unix instead of :unix
 - use :os-windows instead of :mswindows or :windows
 - use :os-macosx instead of :darwin
 - use :lispworks7+ instead of :lispworks7

Make ENSURE-DIRECTORY-PATHNAME and TRUENAME* more robust in case
MAKE-PATHNAME throws an error when trying to synthetize a pathname
with components that the implementation considers invalid.

Uncomment the proper definition of LISP-IMPLEMENTATION-DIRECTORY on CLISP,
which now doesn't break asdf-pathname-test anymore.

When using DIRECTORY, the proper pattern to use to get all files
is #p"*" and not #p"*.*" on CLISP and GCL.
This is the opposite of other CL implementations, that
match everything with #p"*.*" and only files without extension with #p"*",
which seems more in line with the CLHS, though counter-intuitive wrt to POSIX.

Now the proper pattern to match all files with TRANSLATE-PATHNAME and
with logical pathnames is still #p"*.*" or #p"*.*.*".

Therefore, we introduce a new parameter *wild-file-for-directory*,
as distinguished from *wild-file*, for notable use with DIRECTORY-FILES.

This make test-logical-pathname.script work on CLISP,
despite CLISP bug 677.

Mistakenly committed after refilling a comment and emacs refilled it
wrong, so that comment was incorrectly not all prefixed by comment
characters.

Bug found by @axionic on github: on very large directories,
directory-files was slow, because it was using remove-duplicates
with non-hashable test function #'pathname-equal.
But this functionality was only really needed for the purpose
of dealing with logical-pathnames, and having a few duplicates
in case of wanton symlinks isn't much of an issue.

In some cases a subdirectory pathname returned from this function was
not a DIRECTORY-PATHNAME (i.e., leaf directory name was in the
PATHNAME-NAME, not PATHNAME-DIRECTORY).  I used
ENSURE-DIRECTORY-PATHNAME to avoid this happening.

Also add one for SCL, but it's untested.

Fix DIRECTORY* on Genera (contribution by Gary Palter, lp#1518044):
UIOP uses FS:DIRECTORY-LIST on Genera to enumerate the contents of the
source repositories. However, if the directory doesn't exist,
FS:DIRECTORY-LIST signals an error, whereas UIOP expects to get NIL.

A few are kept, for backward-compatibility with systems in quicklisp.

It's been the official way of calling CMUCL since 2011.

I initially introduced all that make-pathname* madness on the road to
ASDF 2.27, back when I was trying to normalize pathname components so
I may have pathname-indexed entries in the asdf-cache (to minimize
costly filesystem access from file-write-date and such, that could
also introduce inconsistency in the build plan).

Problem is, there was no good way (even non-portable) to normalize
pathnames in a way that works when you use all of parse-namestring,
make-pathname, directory, probe-file, translate-logical-pathname,
merge-pathnames, etc. In the end (as late as b4dc78de, before 3.0.2.13!),
I conceded defeat, for it was an uphill battle, and instead asdf-cache
is indexing things using namestrings instead, which is more stable (see
asdf/cache:normalize-namestring).

The hacks in make-pathname* are therefore not needed anymore,
and the function can be done away with.

This commit guts the contents of make-pathname* and marks it deprecated.
Just use make-pathname instead. It also updates uiop and asdf to not
use make-pathname* but make-pathname. Some future commits may move it to
backward-driver and eventually start issuing warnings of some sort when
it's used.

See https://bugs.launchpad.net/asdf/+bug/1502323

Make it support more platforms.

No code change.

Fix PROBE-FILE*, make the (executable) image suffix .exe on Windows.

Punt on making rename-file-overwriting-target atomic.
Punt on trying to run cmd with anything that SBCL would escape.

Based on a contribution by Christian Shafmeister.

actually enforce empty-is-nil in getenv-pathnames. Reindenting a few forms.

Remove early check for absolute pathname, deferring to ENSURE-PATHNAME.

Removed some commented-out material to tidy the code.
Also checked to reject any :EMPTY-IS-NIL NIL cases from code that calls
into SPLIT-NATIVE-PATHNAMES-STRING, since that function always turns
empty entries into NILs.

1. Put in place Fare's proposed solution to
SPLIT-NATIVE-PATHNAMES-STRING.
2. Check for absolute pathnames in PARSE-NATIVE-NAMESTRING.  This was a
check whose need was recognized as a side-effect of getting the first
bugfix in place.

By default, this function now simply drops empty entries in the
environment variable list. But this is controllable by the
DISCARD-EMPTY-ENTRIES keyword argument.

to speed up search in a source-registry :tree. If you use one, don't let it go out-of-date!

Even if the root check failed, it was still recursing.