- Dec 04, 1999
- Dec 03, 1999
-
-
dtc authored
maybe-convert-to-assignment, aborting the conversion if such a ref is found. Problem noted by Peter Van Eynde; patch based on a suggestion from Rob. Add a similar check in maybe-let-convert which is bound to have similar problems. This will most likely need more work as aborting is not optimal, and there appear to be other similar problems.
-
dtc authored
%lexical-exit-breakup for each associated entry, as expected by the interpreter. The generation of multiple breakups had been causing invalid results from interpreted code with multiple non-local exits - as noted for the series extension.
-
dtc authored
generating an error when passed a Nil argument due to an attempt to use ELT to access the name; the name is now accessed via NTH for list based structures.
-
- Nov 29, 1999
-
-
dtc authored
-
- Nov 25, 1999
-
-
dtc authored
the global *adjustable-vectors* resource, rather allocate an new vector for each use in emit-error-break. These vectors are small and in most cases the required length is less than 8 bytes which is the new lower default.
-
dtc authored
the global special *elsewhere-label* by dynamically binding *elsewhere-label* in native-compile-component, otherwise label-elsewhere-p breaks if code generation is reentered while code dumping is still in progress.
-
dtc authored
global tables and counters for the continuation, TN, and label debug IDs; rather dynamically bind new tables and counters in the functions compiler and sub-compiler-file with the aid of the new macro with-debug-counters.
-
dtc authored
the following global tables for compiler debug code by dynamically binding newly allocated tables when necessary: *seen-blocks*, *seen-functions*, and *list-conflicts-table*.
-
- Nov 19, 1999
-
-
dtc authored
found by Akira Kurihara. The bug was caused by (- xx (float a x)) eventually becoming zero; an extra test has been added to exit the do loop when this is true. From Raymond Toy.
-
dtc authored
Sparc V9; replace them with mulx for the Sparc V9. Now use only one of :sparc-v7, :sparc-v8, or :sparc-v9 when building to get the desired features. From Raymond Toy.
-
dtc authored
get the desired features. From Raymond Toy.
-
- Nov 18, 1999
-
-
dtc authored
instructions on the Sparc V9. From Raymond Toy.
-
- Nov 13, 1999
-
-
dtc authored
-
dtc authored
numerator is positive. The result should be still be positive, but interval-div was incorrectly returning an unbounded interval. From Raymond Toy. For two-arg-derive-type, when both arguments are member types, convert both to numeric types and call the deriver function which often results a tighter, less conservative, type. For example the division of two positive numbers that include zero now returns a positive type. The older behavior of calling the function with the two member values is maintained but is now conditional on the :conservative-float-type feature; also ensure that the :invalid trap is masked out when calling the function even under the negative-zero-is-not-zero feature. From Raymond Toy.
-
- Nov 11, 1999
-
-
dtc authored
objects being EQ because they may be NaN in which case Nil must be returned.
-
dtc authored
From Raymond Toy.
-
dtc authored
bignum-logbitp as needed.
-
dtc authored
port. Do not use the popc instruction which isn't implemented in hardware on an Ultrasparc. From Raymond Toy.
-
dtc authored
otherwise the set of masked exceptions may be lost. Further, when restoring the FPU control word restore the rounding mode to its required default of round-to-nearest.
-
dtc authored
-
dtc authored
-
dtc authored
for which it provides a faster stack pop.
-
dtc authored
been removed and this produced a compile error.
-
dtc authored
instruction when the argument is a descriptor-reg.
-
dtc authored
-
- Oct 07, 1999
-
-
dtc authored
-
- Oct 06, 1999
-
-
dtc authored
'jmp *offset(%eax)' rather than 'jmp offset(%eax)'. Peter Van Eynde reports that the older syntax is incorrectly assembled on recent linux systems. The new syntax does work on recent FreeBSD systems.
-
- Sep 28, 1999
-
-
dtc authored
constant TN as an argument, but which failed when comparing a constant TN and integer arguments. For example this can occur when the integer is the static symbol Nil. Noted by Andy Freeman.
-
- Sep 25, 1999
-
-
dtc authored
timeout was given. Fixes CLX event handling.
-
- Sep 23, 1999
-
-
dtc authored
-
- Sep 22, 1999
-
-
dtc authored
the defstruct-slot-descriptor which was broken for inherited slot as the dsd-accessor was set to Nil in this case. Rather than setting dsd-accessor to Nil for inherited slots, add a new function dsd-inherited-p to test if an accessor has been inherited, and always setup the dsd-accessor slot with the name of the accessor function. This allows the structure printer and inspect to easily obtain a slot accessor function.
-
dtc authored
it here to be sa_handler. This is necessary for older Linux distributions which do not define sa_sigaction.
-
- Sep 20, 1999
-
-
pw authored
included other defstructs. An example of a failing case is: (defstruct a1 s1) (defstruct (a2 (:include a1)(:conc-name a1-)) s2) as seen in Hemlock source. The problem was that the dsd-accessor slot is documented to sometimes contain NIL but default-structure-print blindly called fdefinition on the sometimes NIL value. The fix here is to compute the slot accessor name from the conc-name and slot-name. NOTE: There is code in ir1tran that also blindly calls fdefinition on the contents of dsd-accessor. Don't know if this is a latent bug.
-
- Sep 17, 1999
-
-
dtc authored
compute-lra-data-from-pc as it has some object pointer hacks.
-
- Sep 16, 1999