Chris Hanson [Wed, 3 Oct 1990 04:56:28 +0000 (04:56 +0000)]
* X terminal interface extensively changed to use new event-handling
model -- this requires microcode 11.45 or later.
* `delete-screen' command now works. `create-new-screen' command
removed.
* `save-buffers-kill-scheme' command now prompts for confirmation.
* Changed all "event" hook variables to have event distributors as
their values. Users get access to an event by adding themselves to
the event distributor.
* X screens now update their window and icon names under the control
of the new variables `x-screen-name-format' and
`x-screen-icon-name-format'.
* `typein-window?' is now true if the window is a typein window of any
screen. Previously it was only true of the typein window of the
selected screen.
* `select-window' and `select-cursor' now work for windows in any
screen, rather than just for the selected screen.
* Changed `make-buffer' to use `editor-default-mode' for new buffers,
rather than `fundamental-mode'.
* New procedure `typein-edit-other-window' returns the non-typein
window that was active immediately before the current typein edit
began.
* New procedure `format-modeline-string' permits more general use of
the formatting language used for `mode-line-format'.
* Implemented `command-reader/reset-and-execute', which aborts to the
top-level command reader and executes a thunk there before reading
the next command. This mechanism replaces the
`set-reader-do-before-next-read!' procedure.
Internal changes:
* Modeline events for each window are now passed to the window's
screen, in case the screen needs them.
* Guaranteed that `buffer-modified' modeline event is only signalled
when the "modified" bit of the buffer changes.
* Eliminated `cursor-moved' modeline event.
* Screens now cache their dimensions. X screens use this cached
information to determine when a screen's size has changed.
* Removed `reader-continuation' stuff, which was commented out anyway.
Current assumption is that, no matter how many screens there are,
there is only one input port. For X, this means that all screens
are on the same display.
* Editor-frame operations are now subsumed by screen operations.
There are no editor-frame operations in the (edwin) package.
* Many references to `update-screens!' changed to be calls to
`update-selected-screen!'.
* Incremental search bullet-proofed to remove its message when it is
aborted by any means.
* `standard-editor-initialization' bullet-proofed against aborts.
* Typein edits must consistenly use `current-typein-bufferset' for
typein buffers.
* Eliminated "rescrn" by moving `toggle-screen-width' into "wincom".
Add nbits = 0 case to copy_bits to compensate for division bug in gcc.
The bug arises when dividing by OBJECT_LENGTH inside
BIT_STRING_INDEX_TO_WORD. The division is optimized into an unsigned
left shift that works for non-negative values, but not for -1, used
when the number of bits is 0!
Fix recaching bug in extend_frame when the root of the chain is not
the global environment and there is no cell being shadowed.
Compiler_uncache was not called so shadowing_recache called from
higher levels was called on the wrong shadowed cell. There is no
shadowed cell in this case so we use the cell containing the
unbound_trap_object. This makes compiler_uncache and compiler_recache
punt the caching operation.
Chris Hanson [Tue, 11 Sep 1990 20:46:01 +0000 (20:46 +0000)]
* Advice package now signals error if the user attempts to advise
anything but an compound procedure.
* Continuation parser keeps track of the type of the previous stack
frame. This information is used as context in some situations.
This mechanism replaces special-purpose flag `allow-next-extended?'.
* Continuation parser slightly reorganized and commented to make it
easier to understand.
* Debugger modified to provide more flexible control over use of
history information, to provide more detailed information about
stack frames, and to make it more self-explanatory. Also knows
about "simulated" compiled-code environment frames, and ignores
them.
* The environment inspector has been modified to make it more
self-explanatory. The N command has been replaced by an O command
like that of the debugger.
* `pretty-print' now has additional optional argument that specifies
an indentation for the printed expression. If given, the output is
indented by that many columns.
* The emacs interface now has a hook for evaluating arbitrary
emacs-lisp expressions. This is used to provide a better debugger
interface.
**** This requires "xscheme.el" version 1.26 or later. ****
* `stack-frame/debugging-info' now returns a third value,
"subexpression", which indicates the subexpression of the expression
that the next later subproblem is evaluating.
* The lambda abstraction now forces the use of internal lambda
expressions for auxiliary variables. This is required for correct
semantics of `letrec'.
* `make-lambda' now does error-checking on its parameter-list
arguments, which disallows duplicates in the parameter lists.
* The `procedure' abstraction has been split off into a separate file.
`procedure-arity' has been modified to handle entities correctly.
A new datatype, `apply-hook', is like entities except that it
doesn't pass itself to the handler. `compound-procedure' operations
have been removed from the global environment; use generic
operations instead.
* The unsyntaxer has a new entry point, `unsyntax-with-substitutions',
which allows subexpressions of an expression to be replaced in the
output with arbitrary objects.
* Removed `dynamic-state-let' from `system-global-syntax-table'.
* The syntaxer now disallows the use of syntactic keywords as
variables. This applies to references, bindings, and definitions.
* The syntaxer signals an error if the name of a named `let' is also
one of its bound variables.
* The syntaxer signals an error if there are duplicates in the
parameters of a lambda expression.
* Compiled-code environments that do not have interpreter-compatible
ancestors now simulate such ancestors for debugging convenience.
The simulated ancestor is the closing environment of the compiled
code, if known, otherwise it is the system global environment.
* New macro UNSIGNED_SHIFT_BUG has opposite meaning of old
UNSIGNED_SHIFT -- it says that the compiler uses arithmetic right
shift on the `unsigned long' type.
* Redo MIPS configuration to handle non-DEC machines.
* Add configurations for i386 and NeXT.
* Eliminate `term_type' definition -- HP's compiler no longer has this
problem.
* Change `FIXNUM_TO_LONG' to use more efficient conversion suggested
by Aubrey Jaffer.
Fix a bug in the interaction between multi-closures and frame reusing.
The shared closure block was not considered when determining whether a
slot could be rewritten or not.
- non-canonical multi-closure entries appearing in letrecs were
computed incorrectly. The context offset is not the context after
binding since the code has not finished pushing.
- referencing a closure from within itself cannot bypass the variable
lookup code if the closure for environment is not the closure for
value. This only affects machines where closures are canonicalized
for environment.
Replace internal code words with return address code words so the
debugger will not get confused about internal procedures and interrupt
frames in compiled code.
Split cache flusing code from STORE_EXECUTE_CACHE_ADDRESS, since it is
not needed at GC time because the whole cache is flushed immediately
afterwards.
The flushing code is now in FLUSH_I_CACHE_REGION, used explicitly by
store_uuo_link in cmpint.c
Chris Hanson [Thu, 16 Aug 1990 20:13:17 +0000 (20:13 +0000)]
Don't use `event:before-exit' to tell the microcode to clean itself
up. There is now a new hook at the microcode level that delays this
cleanup until the `disk-restore' is committed.