- Mar 22, 2003
-
-
gerd authored
= pcl:class part. To get it booted from 18e, cross-compile using boot1.lisp as bootstrap.lisp in pmai's build scripts, then do a normal compile with boot2.lisp as bootstrap.lisp with the resulting Lisp. * code/byte-interp.lisp, code/defstruct.lisp, code/describe.lisp: * code/error.lisp, code/exports.lisp, code/hash-new.lisp: * code/hash.lisp, code/macros.lisp, code/misc.lisp: * code/package.lisp, code/pred.lisp, code/sharpm.lisp, code/type.lisp: * compiler/dump.lisp, compiler/fndb.lisp, compiler/globaldb.lisp: * compiler/proclaim.lisp, compiler/typetran.lisp, compiler/xref.lisp: * compiler/generic/primtype.lisp, compiler/generic/vm-type.lisp: Changes for to use kernel::class etc. * code/class.lisp (toplevel): Shadow class, built-in-class etc. (class): Give it conc-name %class-. (toplevel) [#+bootstrap-lisp-class=pcl-class]: Define old accessors. (everywhere): Use new class accessors. * compiler/generic/vm-fndb.lisp (%make-instance): Change from unsafe to flushable and movable. * code/ntrace.lisp (expand-trace, untrace): Changes for method tracing. * code/profile.lisp (profile, profile-all, unprofile): Method profiling. * pcl/*.text, pcl/bench.lisp, pcl/extensions.lisp: * pcl/fast-init.lisp, pcl/precom1.lisp, pcl/precom4.lisp: * pcl/structure-class.lisp, pcl/user-instances.lisp: Removed. * tools/pclcom.lisp: Changes for my PCL and lisp:class = pcl::class.
-
pmai authored
-
- Mar 21, 2003
- Mar 19, 2003
- Mar 18, 2003
-
-
toy authored
bug cons-union-method-inaccurate.
-
- Mar 17, 2003
- Mar 10, 2003
-
-
gerd authored
Treat NIL as a symbol, to produce the same result as the DEFTRANSFORM for sxhash/symbol in c:generic/vm-tran.lisp.
-
- Mar 08, 2003
-
-
pmai authored
the layout of struct dirent64 returned from readdir64. Also added readdir64 to the list of stubs, in order to force usage of the same readdir64 definition at runtime as was used at compile-time.
-
- Mar 06, 2003
-
-
pmai authored
Attached is the full patch containing the addition of the ISTREAM-MEMORY-BARRIER VOP, the change to GENTRAP to use the bugchk opcode, and also a boot file to allow the patched sources to be compiled with an unpatched binary.
-
emarsden authored
This is to avoid ending up with the same initializer present twice, and also to give the initializer a clearer print-name.
-
- Mar 05, 2003
-
-
emarsden authored
(values nil t). This fixes a silly mistake in the recent LFS changes.
-
- Mar 04, 2003
- Mar 03, 2003
- Mar 02, 2003
-
-
toy authored
-
emarsden authored
files to be visible via DIRECTORY and friends: added stat64, fstat64, lstat64, statfs64, readdir64. This requires additional transitions to 64-bit wide data types, and additions to linux-stubs.S, given that some of the 64-bit stat variants are not accessible via dlsym(). Mostly from Pierre Mai.
-
- Feb 28, 2003
-
-
emarsden authored
-
- Feb 25, 2003
-
-
emarsden authored
:if-output-exists and :if-error-exists. The two forms below currently signal errors if the file /tmp/inexist does not exist, due to the defaulting value of the :if-exists argument to OPEN. With this change, the output/error file is created if it does not already exist. (ext:run-program "ls" '("/tmp") :output "/tmp/inexist" :if-output-exists :append) (ext:run-program "ls" '("/zob") :error "/tmp/inexist" :if-error-exists :append)
-
toy authored
-
emarsden authored
uses the CPUID + RDTSC instructions on Pentium, and reads the %TICK register on UltraSPARC. Accessible via the VM::READ-CYCLE-COUNTER VOP that returns two (unsigned-byte 32) values, that are the lower and upper components of a 64-bit cycle count (actually 63 bits for UltraSPARC). Basic support for counting the number of CPU cycles has been added to the TIME macro.
-
emarsden authored
-
- Feb 24, 2003
-
-
emarsden authored
is necessary to support access to files that are larger than 2GB. It is implemented by calling 64-bit variants of certain library functions, such as open64 instead of open. Certain library functions such as lseek are called with a 64-bit offset, instead of a 32-bit offset.
-
gerd authored
form evaluating the result form, for the case where VAR is declared special.
-
emarsden authored
from generic errors to conditions of type PROGRAM-ERROR.
-
emarsden authored
number of arguments was being disabled in the presence of a dotted lambda-list or when a &rest keyword is present. Change this so that the test is only disabled for dotted lambda-lists. This modification fixes a number of bugs in source-transforms used by the compiler, since the macro lambda-list parsing code is used to determine whether a given source-transform can be applied to the form being compiled. When source-transforms were defined for functions with &rest parameters, the macro-lambda-list checking was not signalling an error when an inappropriate number of arguments were present in the form being compiled. This resulted in inappropriate use of the source-transform. This change reveals a problem in the gray-stream class definitions (now fixed).
-
- Feb 23, 2003
-
-
gerd authored
64-bit off_t on BSD/x86.
-
- Feb 21, 2003
-
-
toy authored
value and key to NIL on the last iteration, which doesn't work if we have type declarations for the value/key.
-
- Feb 20, 2003
-
-
gerd authored
-
- Feb 18, 2003
- Feb 16, 2003
-
-
emarsden authored
VALID-FUNCTION-NAME-P, instead of checking for (setf xxx) names. Declare two new valid function name classes: names of the form (:macro foo) (:compiler-macro foo) that are now used to name lambda nodes in IR1.
-
- Feb 15, 2003
-
-
pmai authored
This change causes condition printing (generally) and condition reporting (in particular) to go via print-object, once PCL is loaded, as required by ANSI. Note that the :report option of define-condition still goes through the old non-print-object mechanism (which is allowed by ANSI), but that this mechanism is now invoked via a default method of print-object specialized on condition. Hence user-defined methods on print-object are now effective, instead of being bypassed through the old mechanism.
-
- Feb 14, 2003
-
-
toy authored
has a relative directory component, it is taken as relative to DEFAULTS instead of signaling an error that it can't be represented relative to DEFAULTS.
-
- Feb 12, 2003
-
-
cracauer authored
Fix a problem with the consing-free allocation counter. If you allocate more than most-positive-fixnum bytes between two GCs, do a normal, possibly consing addition instead of the fast dfixnum increment. Also clean up the exports of the dfixnum package and define a new operator to increment a dfixnum by any integer. Tested: full ITA testsuite, tried the profiler, tried various allocation patterns which failed before this fix.
-
- Feb 11, 2003
-
-
toy authored
initializer given on the mt19937 website. o Allow INIT-RANDOM-STATE to take either an integer or a vector to initialize the state of a random-state object o Added GENERATE-SEED to generate a seed suitable for use by INIT-RANDOM-STATE. Use /dev/urandom if available. (Based on work by Eric Marsden).
-