Note that this implementation does not capture values printed to
standard output. It is not clear that dynamically rebinding the
standard output port is possible in Kawa.

sb-kernel:%simple-fun-self is no more, but it's useless as it points
to itself.

Instead of checking for interrupts every N seconds notify the
condition variable that is used by RECEIVE-IF.

This requires an optional WAKE-THREAD implementation function, which
is added for SBCL, CCL and CLASP.

tempnam(NULL, NULL) uses "tmp." as a prefix, which made fasls compiled
from that file name all go into tmp.fasl.

more-count and more-context are used to get the actual &more value, no
need to show them.

@dkim points out that some contribs like slime-editing-commands and
slime-indentation are useful before (or without) connecting to a swank
server. Therefore, let's try and load them a bit earlier.

Partially reverts 9055a7e5.

The "lisp-loop-body-forms-indentation" customization variable is of type
integer but was wrongly specified as boolean.

The code header layout has changed.

This is more friendly towards custom readtables as ! is a common reader
macro character. (Nevertheless, users shouldn't be loading SWANK with a
custom readtable...)

Now handles jar files and buffers.

This fixes issue #339

This way (asdf:load-system :swank :force t) actually forces SWANK to be
reloaded.

It expected a symbol for the second argument and (inst (x)) failed.

This function is being removed from ASDF.

Furthermore, it always returns T, so simply replacing the call to
(ASDF::OPERATION-FORCED O) with T would preserve the behavior of this
code.  However, that is a dangerous default because if executed when
SWANK is already running, it will kill existing connections, so let's
default to NIL.

Fix byte-compiler warnings

ElDoc mode has been using its own eldoc-highlight-function-argument face
for the argument at point in a function's argument list.

This function was added back in d9e86fbe, but in this form it is
effectively a no-op as slime-extended-operator-name-parser-alist was not
brought back (from 67b8bb2e) as well.

Fixes issue #332.

Otherwise it'd trigger a SWANK event while in slime-repl-read-mode which
would block indefinitely.

Fixes issue #329.

Now that slime-setup is delayed until a connection is opened, we should
explictly move the load-path setup to load-time so that contribs that
depend on other contribs can find them and be properly byte-compiled.

Fixes issue #326.

That way the correct readtable can be selected.

For a while now, the party line has been that slime-contribs is the way
to select which contribs are to be loaded. However, there were two
common scenarios that didn't work properly:

  1. (require 'slime)
     (setq slime-contribs <something>)

  2. Installing SLIME via package.el, then setting slime-contribs.

In both cases, the problem was that, after loading SLIME, changing
slime-contribs has no effect since it was only being taken into account
at load-time. This commit makes it so that we now look at slime-contribs
each time a new SLIME connection is launched.

When byte compiling a contrib via the Makefile, slime.el is loaded (via
`require') but not compiled. That causes `byte-compile-current-file' to
point to the contrib file, rather than slime.el, generating confusion
when setting `slime-version'. To avoid this, let's look at both
`load-file-name' and `byte-compile-current-file' to see if one of them
points to slime.el. (Sadly, this involves hard-coding the file name.)