I don't know any Unix that treats two leading slashes differently, so
I won't reject that case. However, maybe the pathname abstraction
needs to handle it differently if there are such Unices.
Update terminal drawing operation costs on resize.
Fixes random errors during redraw after expanding your terminal.
Evidently nobody uses Edwin in a terminal, at least since I broke
this five years ago...
Matt Birkholz [Tue, 19 Apr 2011 07:49:14 +0000 (00:49 -0700)]
New compile-system procedure. Replaced ffi.sf and ffi.cbf.
* src/compiler/base/toplev.scm (compile-file): Only declare
integrate-external when dependencies have no type. Thus a dependency
on e.g. "gtk-const.bin" might cause the dependent file to be
re-compiled without a spurious warning about a missing .ext file.
* src/cref/: butils.scm, cref.pkg: Added new file butils.scm,
containing a compile-system procedure that applies compile-file to
each file mentioned in a .pkg file.
* src/ffi/: compile.scm, ffi.cbf, ffi.sf: Replaced the old .cbf and
.sf files with a new, modern compile.scm file.
Matt Birkholz [Mon, 18 Apr 2011 15:25:25 +0000 (08:25 -0700)]
Fixed ffi/Makefile-fragment install target.
* src/Setup.sh: Create src/lib/ links to the ffi-test FFI.
* src/ffi/Makefile-fragment: Punt installing the ffi-test FFI, which
is intended for `make check', which developers run in the source tree.
The links in src/lib/ are enough; no need to copy files there.
Matt Birkholz [Mon, 18 Apr 2011 00:08:38 +0000 (17:08 -0700)]
Added FFI test.
* doc/ffi/ffi.texinfo: Punt the Makefile fragments for the example,
and the logic behind the leak-proof malloc/free procedures.
* doc/user-manual/user.texinfo: Needlessly picked some consistency nits.
* src/Clean.sh, src/Setup.sh: Link config.sub and config.guess to
microcode/*, and clean them too. These scripts are used by
src/configure.ac, which uses AC_CANONICAL_HOST to configure the
compilation and linking of FFIs.
* src/Makefile.in (FFIS, build-ffis): Added FFIS, the list of
configured foreign library interfaces (subdirs). These are built by
the new build-ffis target, which follows build-bands in most places,
and does `make build' in each ffi subdir.
* src/configure.ac: Added --enable-debugging, FFIS, CCLD, DEFS,
CFLAGS, CPPFLAGS, LDFLAGS, SHIM_CFLAGS and SHIM_LDFLAGS. Include
microcode/achost.ac. Added FFIS to the bundles list.
* src/etc/compile.scm (compile-ffi): New. This procedure arranges to
compile an FFI subdir as a LIARC bundle when LIARC is in use.
* src/ffi/Clean.sh: Punt prhello example. Clean up ffi-test-* build
products.
* src/ffi/Makefile-fragment: Punt prhello example. Build and install
an FFI test lib instead.
* src/ffi/: ffi-test.c.stay, ffi-test.cdecl, ffi-test.h: The new test
foreign library and interface.
* src/ffi/: prhello.cdecl, prhello.scm: Removed. Now in doc/;
replaced in src/ by ffi-test*.*.
* src/microcode/: achost.ac, configure.ac: Moved the host
configuration from configure.ac to the new achost.ac, to share with
src/configure.ac.
* src/runtime/ffi.scm (guarantee-alien, error:not-alien): Use the
standard arglist (object operator), not (operator object #!optional
ctype), and do not loop insisting on a qualifying object. Integrate
the test, separated from the restart code, which goes in the new,
global error:not-alien procedure.
(guarantee-alien-function, error:not-alien-function): New, like
guarantee-alien and error:not-alien. Use them in call-alien.
(make-alien-to-free): New, for interfacing to C functions like
malloc() that return something that needs to be freed with free().
Use it in the Scheme malloc procedure, and the test FFI.
* src/runtime/runtime.pkg: Added error:not-alien,
guarantee-alien-function, error:not-alien-function and
make-alien-to-free.
* tests/Clean.sh: New, for cleaning up after tests.
* tests/check.scm: Added tests/ffi/test-ffi.scm. Hacked handling of
test-flonum-casts.com so that `make check' works on portable C.
* tests/ffi/test-ffi-wrapper.scm: New. Code that needs to be
syntaxed/compiled as part of the FFI test.
* tests/ffi/test-ffi.scm: New. A test of the C/Unix FFI.
Make `--batch-mode' without `--interactive' imply non-interactive.
In this case, don't check whether stdin/stdout/stderr are ttys to
determine whether we are interactive.
This keeps `scheme --batch-mode' from futzing with the tty interrupt
characters, so that if you run it from a script, there is no window
during which ^G becomes the terminal's interrupt character and ^C its
quit character. E.g., if you have set `stty intr ^C', then Scheme
won't interfere with that while you run your script.
Use the closefrom system call if available to close all fds.
If not available, fall back to fcntl(F_CLOSEM); if that's not
available, fall back to closing all fds up to and including
fcntl(F_MAXFD); if that's not available, fall back to closing all
fds below UX_SC_OPEN_MAX like before.
Each case tested under NetBSD by running runtime/test-process.scm,
evaluating (RUN-SHELL-COMMAND "ls"), and ktracing Scheme during
The implementation is totally kludgey, and the tests take an
unreasonably long time to run (since we've mixed stress tests with
simple tests), but this is better than nothing, and may help to
complement firing up Edwin as the general way to test Scheme.
Advise operating system about memory access during GC.
- Expect sequential access to oldspace during gc_scan_oldspace.
- Free heap pages (newspace) before copying tospace over them.
- Free tospace pages after saving tospace.
The OS hooks are implemented only for Unix, in terms of posix_madvise
and madvise. Implementing them for Windows and/or OS/2 is left as an
exercise for the reader.
Chris Hanson [Sun, 20 Mar 2011 09:25:38 +0000 (02:25 -0700)]
For some reason, fatal_buf doesn't have a NUL under certain circumstances on my Mac. Instead of calling strlen, do a more careful scan that stops at the end of the buffer.
This returns the number of seconds, with at most nanosecond
resolution, that have elapsed since the start of 1972 in UTC.
Representation is fixed-point, stored in a pair given as input.
Thanks to Yoichi Hirai and Kazuhiko Yamamoto for analyzing the nature
of the bug and finding a good fix.
The trivial tests are enough to catch trivial errors, and one
manifestation of the balancing bug, but they should be replaced by a
more comprehensive test suite later.
Check URI record arguments more carefully in ->MUMBLE-URI.
->ABSOLUTE-URI would formerly accept URI records representing
relative references, even though it would reject their string
representations. Now it rejects both.