 |
Description  |
|
|
BACKGROUND OF THE INVENTION
1. Field of the Invention
The techniques disclosed herein concern user interfaces to computers
generally and more specifically interactive user interfaces which employ
windows and pointing devices.
2. Description of the Prior Art
The problem of designing appropriate graphical user interfaces to the
UNIX.RTM. system is vexing and largely unsolved, even today, ten years
after bitmap displays were first attached to UNIX systems. In those ten
years, graphical applications have become major subsystems that sidestep
or even subvert some of the properties of UNIX that helped make it
popular, in particular its piece-parts, tool-based approach to
programming. Although there have been some encouraging recent attempts, in
particular ConMan, described in Paul Haberli, "ConMan: A Visual
Programming Language for Interactive Graphics", Comp. Graph., Vol. 22, #4
August 1988, pp. 103-110, and Tcl, described in John Ousterhout, "Tcl: An
Embeddable Command Language", Proc. USENIX Winter 1990 Conf., pp. 133-146,
they have taken the form of providing interprocess communication within
existing environments, permitting established programs to talk to one
another. None has approached the problem structurally. Moreover, they are
minor counterexamples to the major trend, which is to differentiate among
systems by providing ever larger, fancier, and more monolithic graphics
subsystems rather than by increasing the functionality or programmability
of the overall system. To the software developer, that trend is
problematical; modem user interface toolkits and window systems are as
complex as the systems UNIX displaced with its elegant, simple approach.
The system of the invention, termed herein Help is an experimental program
that combines aspects of window systems, shells, and editors to provide an
improved user interface for textual applications. It is not a `toolkit`;
it is a self-contained program, more like a shell than a library, that
joins users and applications. From the perspective of the application, it
provides a universal communication mechanism, based on familiar UNIX file
operations, that permits small applications--even shell procedures--to
exploit the graphical user interface of the system and communicate with
each other. For the user, the interface is extremely spare, consisting
only of text, scroll bars, one simple kind of window, and a unique
function for each mouse button--no widgets, no icons, not even pop-up
menus. Despite these limitations, help is an effective environment in
which to work and, particularly, to program.
Help's roots lie in Wirth's and Gutknecht's Oberon system, described in N.
Wirth and J. Gutknecht, "The Oberon System", Software Practice and
Experience, September 1989, vol 19, no. 9, pp 857-894 and in Martin
Reiser, The Oberon System, Addison Wesley, New York 1991. Oberon is an
attempt to extract the salient features of Xerox's Cedar environment,
described in W. Teitelman, "A Tour through Cedar", IEEE Software 1, no. 2,
pp. 44-73, and implement them in a system of manageable size. It is based
on a module language, also called Oberon, and integrates an operating
system, editor, window system, and compiler into a uniform environment.
Its user interface is especially simple: by using the mouse to point at
text on the display, one indicates what subroutine in the system to
execute next. In a normal UNIX shell, one types the name of a file to
execute; instead in Oberon one selects with a particular button of the
mouse a module and subroutine within that module, such as Edit.Open to
open a file for editing. Almost the entire interface follows from this
simple idea.
A major difficulty with the Oberon system is that it has a single process
and is language oriented. Modern computing systems are typically
multi-process systems and are file oriented instead of language oriented.
It is thus an object of the invention to secure the advantages of Oberon's
user interface in such multi-process and file-oriented computing systems.
SUMMARY OF THE INVENTION
In one aspect, the invention is a method for defining an input to a
computer system having at least one window. The method has steps including
associating the window with a context; accepting a string of text selected
from anywhere in the window; and providing the string of text together
with the context as the input. This and other aspects and advantages of
the invention will be clear to those of ordinary skill in the art upon
perusal of the Drawing and Detailed Description, wherein:
BRIEF DESCRIPTION OF THE DRAWING
FIG. 1 shows the display provided by the invention;
FIG. 2 shows how a command is input from the display;
FIG. 3 shows how contexts are defined for names;
FIG. 4 shows how the operation of reading mail headers is defined;
FIG. 5 shows how the headers are displayed in a window;
FIG. 6 shows how an operation of viewing a process's stack is defined;
FIG. 7 shows how a portion of the stack is displayed and how a portion of
the code for a function is selected for viewing;
FIG. 8 shows how the selected portion of the code is displayed in a window;
FIG. 9 shows how an operation of finding places where a variable is used is
defined;
FIG. 10 shows how the places where the variable is used are displayed in a
window; and
FIG. 11 shows how the code at the places where the variable is used is
displayed in windows.
Reference numbers employed in the following Detailed Description have two
parts. The two right-most digits are reference numbers within a figure;
the remaining digits to the left are figure numbers; thus a reference
number 1003 refers to an item which appears for the first time in FIG. 10.
DETAILED DESCRIPTION
In the following, the basics of the user interface provided by Help are
first described; thereafter, an example is given of how the user interface
might be used in locating and fixing a bug in a program.
Basic User Interface: FIGS. 1-3
In a presently-preferred embodiment, Help operates only on text which is
displayed in a window; in other embodiments, the techniques set forth here
may be used with icons or other non-text representations of entities. In
the preferred embodiment, the interface to Help are a three-button mouse
and keyboard. In other embodiments, other pointing devices may be
employed, including one and two button mice and pen devices. The
fundamental operations are to type text with the keyboard and to control
the screen and execute commands with the mouse buttons. Text may be
selected with the left and middle mouse buttons. The middle button selects
text defining the action to be executed; the left selects the object of
that action. The right button controls the placement of windows. Note that
typing does not execute commands; newline is just a character.
Several interrelated rules were followed in the design of the interface.
These rules are intended to make the system as efficient and comfortable
as possible for its users. First, brevity: there should be no actions in
the interface--button clicks or other gestures--that do not directly
affect the system. Thus help is not a `click-to-type` system because that
click is wasted; there are no pop-up menus because the gesture required to
make them appear is wasted; and so on. Second, no retyping: it should
never be necessary or even desirable to retype text that is already on the
screen. Many systems allow the user to copy the text on the screen to the
input stream, but for small pieces of text such as file names it often
seems easier to retype the text than to use the mouse to pick it up. As a
corollary, when not typing genuinely new text, such as when browsing
source code or debugging, it should be possible to work efficiently and
comfortably without using the keyboard at all. Third, automation: let the
machine fill in the details and make mundane decisions. For example, it
should be good enough just to point at a file name, rather than to pass
the mouse over the entire textual string. Finally, defaults: the most
common use of a feature should be the default. Similarly, the smallest
action should do the most useful thing. Complex actions should be required
only rarely and when the task is unusually difficult.
As shown in FIG. 1, the help screen 101 is tiled with windows 103 of
editable text, arranged in (typically) two side-by-side columns 115(a) and
(b). In other embodiments there may be only a single column or more than
two columns FIG. 1 shows help screen 101 in mid-session. Each window 103
has two subwindows, a single tag line 105 across the top and a body of
text 107. Tag 105 typically contains the name of the file whose text (or a
copy thereof) appears in body 107.
The text in each subwindow (tag 105 or body 107) may be edited using a
simple cut-and-paste editor integrated into the system. The left mouse
button selects text; the selection is that text between the point where
the button is pressed and where it is released. Each subwindow has its own
selection. One subwindow--the one with the most recent selection or typed
text--is the location of the current selection 109 and its selection
appears in reverse vide. The selection in other subwindows appears in
outline, as seen at PS 111.
Typed text replaces the selection in the subwindow under the mouse. The
right mouse button is used to rearrange windows 103. The user points at
tag 105 of a window 103, presses the tight button, drags the window to
where it is desired, and releases the button. Help then does whatever
local rearrangement is necessary to drop the window to its new location
(the rule of automation). This may involve coveting up some windows or
adjusting the position of the moved window or other windows. Help attempts
to make at least the tag of a window fully visible; if this is impossible,
it covers the window completely.
A tower of small black tabs 113, one per window, adorns the left edge of
each column 115. These tabs 113 represent the windows in that column 115,
visible or invisible, in order from top to bottom of the column, and can
be clicked with the left mouse button to make the corresponding window
fully visible, from its tag 105 to the bottom of the column it is in. A
similar row across the top of the columns allows the columns to expand
horizontally. These little tabs are a solution to the problem of managing
many overlapping windows.
Like the left mouse button, the middle button also selects text, but the
act of releasing the button does not leave the text selected; rather it
executes the command indicated by that text. For example, to cut some text
from the screen, one selects the text with the left button, then selects
with the middle button the word Cut 119 anywhere it appears on the
display. (By convention, capitalized commands represent built-in
functions.) As in any cut-and-paste editor, the cut text is remembered in
a buffer and may be pasted into the text elsewhere. If the text of the
command name is not on the display, one just types it and then executes it
by selecting with the middle button. Note that Cut 119 is not a `button`
in the usual window system sense; it is just a word, wherever it appears,
that is bound to some action. To make things easier, help interprets a
middle mouse button click (not double click) anywhere in a word as a
selection of the whole word (the rule of defaults). Thus one may just
select the text normally, then click on Cut with the middle button,
involving less mouse activity than with a typical pop-up menu. As shown at
201 in FIG. 2, the text selected for execution is underlined. Arrow 202
indicates the current cursor position. As a strict rule, if the text for
selection or execution is the null string, help invokes automatic actions
to expand it to a file name or similar context-dependent block of text; if
the text is non-null, it is taken literally.
As an extra acceleration, help has two commands invoked by chorded mouse
buttons. While the left button is still held down after a selection,
clicking the middle button executes Cut; clicking the right button
executes Paste, replacing the selected text by the contents of the cut
buffer. These are the most common editing commands and it is convenient
not to move the mouse to execute them (the rules of brevity and defaults).
One may even click the middle and then right buttons, while holding the
left down, to execute a cut-and-paste, that is, to remember the text in
the cut buffer for later pasting.
More than one word may be selected for execution; executing Open
/user/rob/lib/profile creates a new window and puts the contents of the
file in it. (If the file is already open, the command just guarantees that
its window is visible.) Again, by the rule of automation, the new window's
location will be chosen by help.
A typical shell window in a traditional window system permits text to be
copied from the typescript and presented as input to the shell to achieve
some son of history function: the ability to re-execute a previous
command. Help instead tries to predict the future: to get to the screen
commands and text that will be useful later. Every piece of text on the
screen is a potential command or argument for a command. Many of the basic
commands pull text to the screen from the file system with a minimum of
fuss. For example, if Open is executed without an argument, it uses the
file name containing the most recent selection (the rule of defaults).
Thus one may just point with the left button at a file name and then with
the middle button at Open to edit a new file. Using all four of the rules
above, if Open is applied to a null selection in a file name that does not
begin with a slash /), the directory name is extracted from the file name
in the tag of the window and prepended to the selected file name. An
elegant use of this is in the handling of directories. When a directory is
Opened, help puts the its name 121, including a final slash, in the tag
and just lists the contents 123 (i.e., the names of the files in the
directory) in the body. For example, as shown in FIG. 3, by pointing at
dat.h 301 in the source file /usr/rob/src/help/help.c and executing Open,
a new window 305 is created containing the contents of
/usr/rob/src/help/dat.h: all it takes is two button clicks. Making any
non-null selection disables all such automatic actions: the resulting text
is then exactly what is selected.
That Open prepends the directory name gives each window a context: the
directory in which the file resides. The various commands, built-in and
external, that operate on files derive directory 307 in which to execute
from tag line 105 of window 305. Help has no explicit notion of current
working directory; each command operates in the directory appropriate to
its operands.
The Open command has a further nuance: if the file name is suffixed by a
colon and an integer, for example help. c:27, the window will be
positioned so the indicated line is visible and selected. Help's syntax
further permits specifying general locations, although only line numbers
will be specified in the following examples.
It is possible to execute any external operating system command. If a
command is not a built-in like Open, it is assumed to be an executable
file and the arguments are passed to the command to be executed. For
example, if one selects with the middle button the text grep ` main`
/sys/src/cmd/help/*.c the traditional grep command will be executed.
Again, some default rules come into play. If the tag line 105 of the
window 103 containing the command has a file name and the command does not
begin with a slash, the directory of the file will be prepended to the
command. If that command cannot be found, the command will be searched for
in the conventional directory/bin. The standard input of the commands is
connected to /dev/null; the standard and error outputs are directed to a
special window, called Errors, that will be created automatically if
needed. The Errors window is also the destination of any messages printed
by the built-in commands.
Example of Operation: FIGS. 4-11
The interplay and consequences of these rules are easily seen by watching
the system in action. In the following example, the Help system is used to
fix a simple bug in a program. The bug has been reported in a mail message
sent by a user named Sean.
When help starts it loads a set of `tools` into the right hand column of
its initially two-column screen. These are files with names like
/help/edit/stf (the stuff that the help editor provides), /help/mail/stf,
and so on. Each is a plain text file that lists the names of the commands
available as parts of the tool, collected in the appropriate directory. A
help window on such a file behaves much like a menu, but is really just a
window on a plain file. The useful properties stem from the interpretation
of the file applied by the rules of help; they are not inherent in the
file. Turning to FIG. 4, mail is read by executing the program headers in
the mail tool, that is, I click the middle mouse button on the word
headers 403 in window 401 containing the file/help/mail/stf. This executes
the program /help/mail/headers by prefixing the directory name of the file
/help/mail/stf, 405, collected from tag 407, to the executed word,
headers. This simple mechanism makes it easy to manage a collection of
programs in a directory.
As seen in FIG. 5, Headers creates a new window 501 containing the headers
of the mail messages, and labels it /mail/box/rob/mbox. I know Sean has
sent me mail, so I point at header 503 of his mail (just pointing with the
left button anywhere in the header line will do) and click on messages
505. Help responds as shown in FIG. 6. It opens a new window 601 in
left-hand column 115(a). Message 603 is displayed in window 601. Message
603 indicates that a new version of help has crashed and a broken process
lies about waiting to be examined. The message includes process number 603
of the broken process; all that is needed to he done in order to look at
the process's stack is to point at process number 603 and execute stack
607 in the window 605 for the debugger tool, /help/db/stf.
The result of the execution of the stack program is shown in FIG. 7. A new
window, 701, has appeared in column 115(a). The contents 703 of window 701
are a a traceback of the calls made by the process. Notice that this new
window has many file names in it. These are extracted from the symbol
table of the broken program. One can look at the line (of assembly
language) that died by pointing at the entry
/sys/src/libc/mips/strchr.s:34 705 and executing Open, but the problem
probably lies further up the call stack. The deepest routine in help is
text insert 707, which calls strlen on line 32 of the file text.c. To look
at the place where the trouble seems to have occurred, one points at the
identifying text in the stack window and executes open to see the source,
which appears in window 801 of FIG. 8. The problem is coming to light: as
indicated at 807, s, the argument to strlen, is zero, and was passed as an
argument to textinsert by the routine errs, which apparently also got it
as an argument from Xdie2. To close the window on text.c, one hits Close!
809 in the tag of window 801. By convention, commands ending in an
exclamation mark take no arguments; they are window operations that apply
to the window in which they are executed. The next step is examining the
source of the suspiciously named Xdie2 by pointing at stack trace 703 and
Opening again.
Help responds as shown in FIG. 9. Window 901 contains line 252 (indicated
by the reverse video) of exec.c. Now the problem gets harder. The argument
passed to errs 903 is a variable, n 905 that appears to be global. Who set
it to zero? The c browser tool /help/cbr/stf, in window 909, lets me find
out. The program uses*.c in that window displays all references in a given
file to a given variable. To employ uses*.c to find all references to the
variable n in /usr/rob/src/help/*.c, one points at the variable in the
source text and executes uses *.c by sweeping both `words` with the middle
button in /help/cbr/stf. As shown in FIG. 10, uses creates a new window
1001 with a list in which the locations of all references to the variable
n in the files /usr/rob/src/help/*.c are indicated by file name and line
number. The implementation of the C browser is described below; in a
nutshell, it parses the C source to interpret the symbols dynamically.
The first item on the list in window 1001 is clearly the declaration in the
header file. It looks like help.c:35 1003 should be an initialization. To
find out, one uses Open help.c to examine that line and see that the
variable is indeed initialized. The result of the Open is shown in FIG.
11. Window 1101 shows line 35 of help.c at 1103. The line indicates that n
905 was set; some other use of n must have cleared it. Line 252 of exec.c
is the call; that's a read, not a write, of the variable. The next
possibility is exec. c:213; on pointing to that and again executing Open,
window 1107 appears; as shown there at reverse video line 1109, n is set
to 0. Sometime before Xdie2 was executed, Xdie1 cleared n. The bug is
fixed by selecting Cut 1111 to remove the offending line, selecting Put!,
which appears in the tag of a modified window, to write the file back out
and then selecting mk 1113 in /help/cbr to compile the program. It is
worth noting that the entire debugging operation of the example was done
without touching the keyboard.
This demonstration illustrates several things besides the general flavor of
help. It is easy to work with files and commands in multiple directories.
The rules by which help constructs file names from context and,
transitively, by which the utilities derive the context in which they
execute simplify the management of programs and other systems constructed
from scattered components. Also, the few common rules about text and file
names allow a variety of applications to interact through a single user
interface. For example, none of the tool programs has any code to interact
directly with the keyboard or mouse. Instead help passes to an application
the file and character offset of the mouse position. Using the interface
described in the next section, the application can then examine the text
in the window to see what the user is pointing at. These operations are
easily encapsulated in simple shell scripts, an example of which is given
in the next section.
The Interface Seen by Programs
help provides its client processes access to its structure by presenting a
file service, as described in Rob Pike, et at., "Plan 9 from Bell Labs",
Proc. of the Summer 1990 UKUUG Conf., London, July, 1990, pp. 1-9. Each
help window is represented by a set of files stored in numbered
directories. The number of each directory is a unique identifier, similar
to UNIX process id's. Each directory contains files such as tag and body,
which may be read to recover the contents of the corresponding subwindow,
and ctl, to which may be written messages to effect changes such as
insertion and deletion of text in contents of the window. The help
directory is conventionally mounted at /mnt/help, so to copy the text in
the body of window number 7 to a file, one may execute
cp/mnt/help/7/body file.
To search for a text pattern,
grep pattern/mnt/help/7/body
An ASCII file /mnt/help/index may be examined to connect tag file names to
window numbers. Each line of this file is a window number, a tab, and the
first line of the tag.
To create a new window, a process just opens /mnt/help/new/ctl, which
places the new window automatically on the screen near the current
selected text, and may then read from that file the name of the window
created, e.g. /mnt/help/8. The position and size of the new window is, as
usual, chosen by help.
Implementation of Components of the C Browser
The directory /help/cbr contains the C browser we used above. One of the
programs there is called decl; it finds the declaration of the variable
marked by the selected text. Thus one points at a variable with the left
button and then executes ctecl in window 909 for the file /help/cbr/stf.
Help executes /help/cbr/decl using the context rules for the executed text
and passes it the context (window number and location) of the selected
text through an environment variable, helpsel.
Decl is a shell (rc) script. Here is the complete program:
______________________________________
eval `{help/parse -c}
x=`{cat/mnt/help/new/ct1}
echo a
echo $dir/` Close!`
} .vertline. help/buf >/mnt/help/$x/ct1
{
cpp $cppflags $file .vertline.
help/rcc -w -g -i$id -n$line .vertline. sed 1q
}>/mnt/help/$x/bodyapp
______________________________________
The first line runs a small program, help/parse, that examines $helpsel and
establishes another set of environment variables, file, id, and line,
describing what the user is pointing at. The next creates a new window 103
and sets x to its number. The first block writes the directory name to the
tag line 105; the second runs the C preprocessor on the original source
file and passes the resulting text to a special version of the compiler.
This compiler has no code generator; it parses the program and manages the
symbol table, and when it sees the declaration for the indicated
identifier on the appropriate line of the file, it prints the file
coordinates of that declaration. This appears on standard output, which is
appended to the new window by writing to /mnt/help/$x/bodyapp. The user
can then point at the output to direct Open to display the appropriate
line in the source. (A future change to help will be to close this loop so
the Open operation also happens automatically.) Thus with only three
button clicks one may fetch to the screen the declaration, from whatever
file in which it resides, the declaration of a variable, function, type,
or any other C object.
A couple of observations about this example. First, help provided all the
user interface. To turn a compiler into a browser involved spending a few
hours stripping the code generator from the compiler and then writing a
half dozen brief shell scripts to connect it up to the user interface for
different browsing functions. Given another language, we would need only
to modify the compiler to achieve the same result. We would not need to
write any user interface software. Second, the resulting application is
not a monolith. It is instead a small suite of tiny shell scripts that may
be tuned or toyed with for other purposes or experiments.
Other applications are similarly designed. For example, the debugger
interface, /help/db, in window 605, is a directory of ten or so brief
shell scripts, about a dozen lines each, that connect adb to help. Adb has
a notoriously cryptic input language; the commands in /help/db package the
most important functions of adb as easy-to-use operations that connect to
the rest of the system while hiding the rebarbative syntax. People
unfamiliar with adb can easily use help's interface to it to examine
broken processes. Of course, this is hardly a full-featured debugger, but
it was written in about an hour and illustrates the principle. It is a
prototype, and help is an easy-to-program environment in which to build
such test programs. A more sophisticated debugger could be assembled in a
similar way, perhaps by leaving a debugging process resident in the
background and having the help commands send it requests.
The most important feature of help is that it provides a comfortable,
easy-to-program user interface that all applications may share. It greatly
simplifies the job of providing an interactive `front end` for a new
application. Help is able to serve as a promtyping system that provides
the user interface for a windowing system, freeing the programmer to
concentrate on the job at hand, the construction of new software tools.
Conclusion
The foregoing has disclosed how one may make and use an improved user
interface for computer systems in which users interact with the system by
means of a display device with windows and a mouse. In a preferred
embodiment, a three-button mouse is employed; however, in other
embodiments, other pointing devices may be employed. Similarly, in a
preferred embodiment, the windows are tiled onto the display device; in
other embodiments they may simply overlap. Further, while the preferred
embodiment deals only with text strings displayed in windows, other
embodiments may deal also with icons and other representations of entities
in windows. Finally, the techniques used to separate tag and body and to
indicate active locations may vary in other embodiments. That being the
case, the foregoing Detailed Description is to be regarded as being in all
respects exemplary but not restrictive, and the scope of the invention is
to be determined not from the Detailed Description, but rather from the
appended claims as interpreted in light of the Detailed Description and
the doctrine of equivalents.
* * * * *
|
|
 |
|
 |
Description |
|
