try to set the source location information.  We try to print a warning
in such cases.

CLHS.

From ticket 81, the tests are now:

{{{
(time (prog1 t (time-rev *s*)))
; Evaluation took:
;   0.49 seconds of real time
;   0.481813 seconds of user run time
;   0.003624 seconds of system run time
;   1,490,776,936 CPU cycles
;   [Run times include 0.13 seconds GC run time]
;   0 page faults and
;   200,073,704 bytes consed.

(time (prog1 t (time-rev *s2*)))
; Evaluation took:
;   0.97 seconds of real time
;   0.965893 seconds of user run time
;   0.005139 seconds of system run time
;   2,980,415,911 CPU cycles
;   [Run times include 0.23 seconds GC run time]
;   0 page faults and
;   400,005,560 bytes consed.
}}}

So the new string-reverse* is 20 times faster for strings without
surrogates and 10 times faster for strings containing only surrogates.

Wrap exports in eval-when for x86 as was done for sparc and add
CHAR-BYTES to x86-x86 cross-compile script.

CHAR-BYTES to x86-x86 cross-compile script.

Only the change for CHAR-BYTES is needed for cross-compiling from x86
to x86, but I think the other changes will be needed if you do a
cross-compile from some other arch to x86.

CHAR-BYTES to ppc cross-compile script.

This is untested.

 src/compiler/generic/objdef.lisp::
 src/compiler/generic/utils.lisp:;
 src/compiler/generic/vm-macs.lisp::
 src/compiler/sparc/c-callback.lisp::
 src/compiler/sparc/parms.lisp::

  When export lost its compile-time effects and became a normal
  function, the exports in many files no longer took affect while
  compiling the file.  This change makes the compile-time effects
  happen as before in selected files. With this change, the sparc port
  can be cross-compiled from x86 again.

 src/tools/cross-scripts/cross-x86-sparc.lisp::
  Need to frob CHAR-BYTES, which is needed by BYTE-BASH-COPY. .

 * GNUmakefile
 * Config.solaris_sunc
   * Remove assignment of CC, letting it default.
 * Config.solaris
   * Force CC = gcc in case cc is Sun C

 * Config.sparc_common
   * OS_LINK_FLAGS is not common, so remove it.
 * Config.sparc_gcc
 * Config.sparc_sunc
   * Set OS_LINK_FLAGS appropriately.

 * GNUmakefile
   * Replace $(DEPENDS) with $(CC)
 * Config.x86_common
   * Set DEPENDS_FLAG to -MM

cc is Sun's C compiler, not an alias to gcc.

older systems or clang on newer ones.

CC if necessary.

 Config.sparc_common::
 * Use += instead of = for CPPFLAGS and CFLAGS

 Config.sparc_gcc:
 * Move options from CC to CFLAGS.

 Config.sparc_gcc:
 * Move options from CC to CFLAGS.
 * Add DEPEND_FLAGS with appropriate options.

utf16_output being undeclared.

In gc_free_heap, the check that unallocated pages are zero can only be
done if gencgc_unmap_zero is MODE_MAP or MODE_MEMSET.  The other modes
are known not to zero memory.

Also, don't limit the check to the first 16K pages, and also check the
entire page of GC_PAGE_SIZE bytes, not just the first 1024 words.

 * Remove PAGE_MADVISE_MASK
 * Remove code using PAGE_MADVISE_MASK
 * Move some #defines from gencgc.c to gencgc.h
 * Add new or better comments

This should only be temporary so we can get some testing with lazy
mode.

 * Disable gencgc_debug_madvise.

 * Add MODE_LAZY
 * Rename PAGE_MADVISE flag to PAGE_MADVISE_MASK
 * Remove some #ifdef'ed out code.
 * Change default gencgc_unmap_zero to MODE_MEMSET for all OS/archs.

   madvise, though.  I get an error about no transport function for
   some object.

 * Actually enable call to madvise instead of doing memset.

out freed pages.

 * Increase the size of *powers-of-ten* a bit.
 * In expt-ten, handle the case where the exponent exceeds the size of
   the *powers-of-ten* array.