Extend the setter for UIOP:GETENV to do this.

Set the test to expect the check for setting an environment variable to NIL to fail on clasp.

Raise `not-implemented-error` if clasp tries to set environment variable to NIL.

This is necessary to avoid ASDF putting single- and multi-threaded SBCL fasls in the same directory.  That's the wrong thing, since they are not compatible. By giving such builds different implementation-identifiers, they are kept apart.

 * Add "Package: CL-USER" to the modeline so a user could compile ASDF on Genera without first setting their Lisp context to ANSI-CL.

Replace `:haiku`, `:mezzano`, and `:genera` with `:os-` prefixed
alternatives.  This fixes a problem on SBCL where we were
inadvertently *removing* the `:haiku` feature.

Treat `:haiku` as a unix variant (fix `os-unix-p`).

Tested on SBCL, ECL, and ABCL. The :|aarch64| is due to a bug in ABCL that will
be fixed in the next release.

Gary Palter authored Feb 15, 2021 and Robert Goldman committed Feb 16, 2021

macOS has implemented a separate read-only system volume and /usr/share is on that volume.
Consequently, /usr/share/common-lisp does not exist nor can it be created. Further, Genera
accesses its host filesystem through NFS and there is no way to export the root (/) or
/usr/share that will not lead Genera into an infinite loop trying to access those mounts.
Instead, just search /usr/local/share which is on the writable Data volume.

Also, change the value return by UIOP/OS:ARCHITECTURE to :VLM when running on a VLM.

Thanks to Martin Simmons.
Fix for issue #18 -- getenv did not work on Windows + Lispworks.

These changes add support for the Mezzano operating system and its
Common Lisp implementation. They are based on changes originally
developed by Henry Harrington against ASDF 3.1.7.

Update comment based on Chun Tian's RMCL test.
No code change.

Shaving away "unknown" from the implementation-identifier on ECL,
where it appears, which can be very precious for FASLs on Windows
due to pathname size limits.

This partially addresses ECL issue 360
https://gitlab.com/embeddable-common-lisp/ecl/issues/360

Shaving away "unknown" from the implementation-identifier on ECL,
where it appears, which can be very precious for FASLs on Windows
due to pathname size limits.

This partially addresses ECL issue 360
https://gitlab.com/embeddable-common-lisp/ecl/issues/360

Recent patch relied on MKCL:GIT-DESCRIBE-THIS-MKCL, which is not
uniformly available.

Rather than "1.1.10" for the current HEAD, which is 155 commits head
of 1.1.9 and will probably eventually be called 1.1.10, the output
will now be "1.1.9.155-fc50d00"; the same git-described-inspired
format that is used by SBCL.

The function mkcl:git-describe-this-mkcl is only available in very
recent versions of MKCL; in particular, not in vanilla 1.1.9

The way that lisp-version-string is written, it will fall back to
lisp-implementation-version unless a platform-specific override is
provided. Since the body of lisp-version-string uses

  (car (list #+foo exp1 exp2))

rather than

  (or #+foo exp1 exp2)

it is theoretically possible that lisp-version-string would return NIL
if that's what exp1 evaluates to, which would not also be returned by
lisp-implementation-version, which it previously would not have.

Such behaviour would qualify as a bug in lisp-version-string, though.

The snippet

  (format nil "~A-~A" (lisp-implementation-version)
                      (core:lisp-implementation-id))

would generate something as superfluous as

  "cclasp-boehm-0.4.0-1024-g3533831-3533831"

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

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

Thus, you can test the *features* as locally bound from some target system.

The JVM doesn't give access to native chdir without some extensions that
aren't available with ABCL. There's a user.dir system property whereby
some Java programs kind of fake it, but ABCL doesn't play nice with it.
Therefore, use *default-pathname-defaults* as our "current directory",
and be sure to chdir to it before to run commands with run-program.
Try to make the run-program chdir feature work on Windows.

Francois-Rene Rideau authored Mar 28, 2015 and Robert Goldman committed Mar 29, 2015

On ABCL, os testing is at runtime.
On all other implementations, it's at compile-time.
SBCL can optimize some runtime tests and then complain about unreachable code.
os-cond abstracts over OS conditionalization and makes everyone happy.
Needed by an upcoming configuration change.

Plus minor typographical fixes.

Based on a contribution by Christian Shafmeister.

The previous version did not seem to update after setting environment
variables.

Needed for some testing.

Thanks to Christophe for patch.

Previously, we would detect Unix first (correctly), then short-circuit,
and never detect Mac OSX.

I would like to rewrite DETECT-OS.  The code seems opaque.  But this is
correct, so committing it now, since it used to be broken.

ensure-function: a hash-table can be seen as a function (to be used by ASDF).

CMUCL: hush some warnings