|
Testing Computer Software Second Edition
Common Software Errors - User Interface Errors
|
This is the appendix from the best-selling book Testing Computer Software, 2nd ed.
Copyright © 1988 by Cem Kaner Copyright © 1993 by Cem Kaner, Jack Falk, Hung Quoc Nguyen
This is part 2 of 13.
[ 1 ]
[ 2 ]
[ 3 ]
[ 4 ]
[ 5 ]
[ 6 ]
[ 7 ]
[ 8 ]
[ 9 ]
[ 10 ]
[ 11 ]
[ 12 ]
[ 13 ]
|
The user interface (UI) includes all aspects of the product that involve the user. The UI designer tries to strike a balance between
- functionality
- time to learn how to use the program
- how well the user remembers how to use the program
- speed of performance
- rate of user errors
- the user's satisfaction with the program
In seeking a good balance, the designer weighs the experience and needs of the people she expects to use the program against the capabilities of the equipment and available software technology. An error in the UI results in a suboptimal match between the user and the program.
Because tradeoffs are unavoidable in UI design, a good designer might deliberately make any of many of the "errors" listed below. Don't take this list as gospel. If you are at all unsure, listen to the designer's reasoning before condemning one of her choices. See Baecker & Buxton (1987), Helander (1991), Laurel (1990, 1991), Rubenstein & Hersh, (1984), Schneiderman (1987), and Smith and Mosier (1984) for excellent introductions to user interface design, including extended discussion of many of the issues raised in this Appendix.
Throughout this Appendix we write as if you were the user of the program. As a tester of it, you will certainly use it heavily. Realize that other people will also use the program and they will have different problems from you. Try to be empathetic.
FUNCTIONALITY
A program has a functionality error if something that you reasonably expect it to do is hard, awkward, confusing, or impossible.
Excessive functionality
This is an error (see Brooks, 1975). It is the hardest one to convince people not to make. Systems that try to do too much are hard to learn and easy to forget how to use. They lack conceptual unity. They require too much documentation, too many help screens, and too much information per topic. Performance is poor. User errors are likely but the error messages are too general. Here's our rule of thumb: A system's level of functionality is out of control if the presence of rarely used features significantly complicates the use of basic features.
Inflated impression of functionality
Manuals and marketing literature should never lead you to believe the program can do more than it can.
Inadequacy for the task at hand
Because a key feature isn't there, is too restricted or too slow, you can't use the program for real work. For example, a database management system that takes 8 hours to sort 1000 records can claim sorting ability but you wouldn't want to use it.
Missing function
A function was not implemented even though it was in the external specification or is "obviously" desirable.
Wrong function
A function that should do one thing (perhaps defined in a specification) does something else. Functionality must be created by the user "Systems that supply all the capabilities the user could want but that also require the user to assemble them to make the product work are kits, not finished products." (Rubenstein & Hersh, 1984, p. 45.)
Doesn't do what the user expects
For example, few people would expect a program written to sort a list of names to sort them in ASCII order. They wouldn't expect it to count leading blanks or distinguish between uppercase and lowercase letters. If the programmers insist that the function should work this way, get them to change its name or add the expected behavior as an option.
COMMUNICATION
This section describes errors that occur in communication from the program to the user. Our model is an interactive program, with the person sitting at a computer or terminal. However, batch programs also give information, such as error messages.
Missing information
Anything you must know should be available onscreen. Onscreen access to any other information that the average user would find useful is also desirable.
No onscreen instructions
How do you find out the name of the program, how to exit it, and what key(s) to press for Help? If it uses a command language, how do you find the list of commands? The program might display this information only when it starts. However it does it, you should not have to look in a manual to find the answers to questions like these.
Assuming printed documentation is readily available
Can you use the program after losing your manual? An experienced user should not have to rely on printed documentation.
Undocumented features
If most features or commands are documented onscreen, all should be. Skipping only a few causes much confusion. Similarly, if the program describes "special case" behavior for many commands, it should document them all.
States that appear impossible to exit
How do you cancel a command or back up in a deep menu tree? Programs should allow you to escape from undesired states. Failure to tell you how to escape is almost as bad as not providing an escape path.
No cursor
People rely on the cursor. It points to the place on the screen where they should focus attention. It can also show that the computer is still active and "listening." Every interactive program should show the cursor and display a salient message when turning the cursor off.
Failure to acknowledge input
An interactive program should acknowledge every keystroke by echoing it immediately on the screen. A few exceptions are not errors:
- When choosing a menu item, you are not confused if your keystroke is not echoed, as long as the next screen appears immediately and the words in the screen's title are identical to those of the menu choice.
- If the program ignores erroneous commands or keystrokes, it should not echo them.
- The program should honor your choice if you can tell it not to echo input.
- When you input your password or security code, the program should not echo it on the screen.
Failure to show activity during long delays
When a program is doing a long task (two seconds), it must show you that it's still working, it's not in an infinite loop, and you don't have to reset the computer.
Failure to advise when a change will take effect
A program may execute a command much sooner or later than you expect. For example, it may continue to display erased data until you exit. If it's unclear when the program will do something, customers will perceive it as having bugs and will make many errors.
Failure to check for the same document being opened more than once
Program that allows user to open multiple documents must check for the same document being opened more than once. Otherwise, the user will not be able to keep track of the changes made to the documents since they all have the same name. For example, the file My_Doc is open, if the user attempts to open My_Doc again, there must be the way for users to identify the first My_Doc versus the second one. A typical method for keeping track is to append a number after the file name such as My_Doc:1 and My_Doc:2 for the first and second file respectively. An alternative method is not to allow the same file to be opened twice.
Wrong, misleading, or confusing information
Every error trains you to mistrust everything else displayed by the program. Subtle errors that lead readers to make false generalizations, such as missing qualifications and bad analogies, annoy testers more than clear factual errors because it's harder to get them fixed.
Simple factual errors
After a program changes, updating screen displays is a low priority task. The result is that much onscreen information becomes obsolete. Whenever the program changes visibly, check every message that might say anything about that aspect of the program.
Spelling errors
Programers dont woory much abuot they're speling misteaks but customers do. Get them fixed.
Inaccurate simplifications
In the desire to keep a feature description as simple as possible, the author of a message may cover only the simplest aspects of the feature's behavior, omitting important qualifications. When she tries to eliminate jargon, she may paraphrase technical terminology inaccurately. Look for these errors. As the tester, you may be the only technically knowledgeable person who carefully reviews the screens.
Invalid metaphors
Metaphors make the computer system seem similar to something you already know. They are good if they help you predict the behavior of the computer and bad if they lead you to incorrect predictions. For example, the trash can icon can be a bad metaphor (Heckel, 1984). If you can take a discarded file out of the trash can, the metaphor is correct. If the file is gone forever once moved to trash, a paper shredder is a better icon.
Confusing feature names
A command named SAVE shouldn't erase a file; nor should it sort one. If a command name has a standard meaning, in the computer community or in the English language, the command must be compatible with its name.
More than one name for the same feature
The program shouldn't refer to the same feature by different names. Customers will waste much time trying to figure out the difference between shadow and drop shadow when the programmer uses both to mean the same thing.
Information overload
Some documents and help screens swamp you with technical detail, to the point of hiding the information or confusing the answers you're looking for. If you think these details are useful, ask whether they are more appropriately located in an appendix in the manual.
Some (not all) detailed discussions of program functioning are disguised apologies from the programmer or complaints from a writer over a poor design. Do users need this information? Also, is there a solution to the problem the programmer is claiming to be insoluble?
When are data saved?
Suppose you enter information that the program will save. Does it save data as you type, when you exit, when you ask for a save, every few minutes, when? You should always be able to find this out. If you get confused answers, look for bugs immediately. Two modules probably make different assumptions about when the same data will be saved. You can probably get one of them to claim that outdated data are up to date. You may find one module erasing or overwriting data that another has just saved.
Poor external modularity
External modularity refers to how modular the product appears from the outside. How easily can you understand any one piece of it? Poor external modularity increases learning time and scares away new users. Information should be presented as independently as possible. The less you need to know in order to do any particular task the better.
Help text and error messages
Help text and error messages are often considered minor pieces of the product. They may be written by junior programmers or writers. Updating them may be given low priority.
You ask for help or run into error handlers when confused or in trouble.You may be upset or impatient. You will not suffer bad messages gladly. The product will build credibility with you as you use it. If some messages mislead you, the rest may as well not be there.
Inappropriate reading level
People don't read as well at computer terminals (see Schneiderman, 1987). When given their choice of reading levels for onscreen tutorials, experimental subjects preferred Grade 5 (Roehmer and Chapanis, 1982). People reading help or error messages may be distressed: these messages should never be more complicated than tutorial text. Messages should be phrased simply, in short, active voice sentences, using few technical terms even if the readers are computer-experienced.
Verbosity
Messages must be short and simple. Harried readers are infuriated by chatty technobabble. When some users need much more information than others, it's common to give access to further information by a menu. Let people choose where and how much further they want to investigate (Houghton, 1984).
Inappropriate emotional tone
People feel bad enough when they make an error or have to ask for help. They don't need their noses rubbed in it. Look for messages that might make some people feel bad. Exclamation marks can be interpreted as scolding, so they should not be in error messages. Violent words like "abort," "crash," and "kill" may be frightening or distasteful. Even the word "error" is suspect: many "errors" wouldn't be if the program (or the programmer) were more intelligent. As a final note, many actions may be invalid or unexpected at a computer, but few are truly "illegal."
Factual errors
Help and error messages often give incorrect examples of how to do something "correctly." Some are outdated. Others were never right. Every message should be (re)checked in one of the last testing cycles.
Context errors
Context-sensitive help and error handlers check what you've been doing. They base their messages (recommendations, menu lists, etc.) on that context. This is excellent when it works, but the message makes no sense when they get the context wrong.
Failure to identify the source of an error
At a minimum, an error message should say what's wrong, and unless it's immediately obvious, repeat or point to the erroneous input (data, program line, whatever). A good error message will also say why something is wrong and what to do about it.
Hex dumps are not error messages
An error message that merely says Error 010 or dumps one or more lines of hexadecimal (octal, even decimal) data is acceptable only if the cost of printing a message is outrageous, the computer doesn't have enough time to print a real message before it crashes, or the programmer is the only person who will ever read it.
Forbidding a resource without saying why
If a program tries to use a printer, modem, more memory, or other resource, and can't, the error message should not only announce the failure but should also say why. You need this information because you'll respond differently to Printer already in use versus Printer not connected.
Reporting non-errors
Error messages should only be triggered by error states. You will ignore all error messages if most are normal-case debug messages or reports of events that are rare but not necessarily due to a bug.
Display bugs
Display bugs are visible. If you see many, you are less likely to buy or trust the program. Unfortunately, display bugs are often considered minor and not even investigated. This is risky. They may be symptoms of more serious underlying errors.
Two cursors
It is a nuisance if the programmer forgets to erase the old cursor when he jumps to another part of the screen. Worse, a second cursor might reflect confusion in the code about which area of the screen is active. The program may misinterpret inputs even it echoes them correctly. If the cursor misbehaves during a test, save and examine any data that you enter.
Disappearing cursor
The cursor usually disappears because the programmer displayed a character on top of it or moved it and forgot to redisplay it. However, a program's pointer to the cursor can be corrupted. If it is, then when it points to a memory location used for data or program storage rather than to screen memory, you won't see the cursor. The program will overwrite information in memory whenever it tries to display the cursor.
Cursor displayed in the wrong place
The program shows the cursor in one place but echoes inputs, etc., in another. This is annoying because it leads you to focus on the wrong part of the screen. A slightly misplaced cursor may warn that the program will truncate entered character strings or pad them with garbage. As with dual cursors, entered text may be echoed correctly but saved incorrectly.
Cursor moves out of data entry area
The cursor should never move out of data entry areas. This is usually a coding error but some programmers deliberately let you move the cursor anywhere on the screen, then beep and display an error message that says you can't enter anything here. This is a design error.
Writing to the wrong screen segment
The cursor is in the right location, but data are displayed in the wrong place on the screen.
Failure to clear part of the screen
A message is displayed for a few seconds, then only partially erased. Or your response to a previous question is left onscreen. It is confusing and annoying to have to type over prompts or irrelevant responses in order to enter something new.
Failure to highlight part of the screen
If a program usually highlights a particular class of items, such as prompts or all text in the active window, it must always do so.
Failure to clear highlighting
This is common when attributes of screen positions are stored separately from displayed text.
The programmer removes highlighted text but forgets to clear highlighting from that area of
the screen. The error is most confusing when the program highlights with double intensity or
boldfacing (as opposed to inverse video, for example). The blank screen area looks fine. The
problem only becomes evident when new text is displayed: it is always highlighted.
Wrong or partial string displayed
The displayed message might be garbage text, a segment of a longer message, or a complete message that should appear some other time. Any of these might reflect errors in the program logic, the values of the pointers used to find message text, or the stored copies of the text. They might indicate minor problems or severe ones.
Messages displayed for too long or not long enough
Many messages are displayed for a fixed time, then erased automatically. The message should be onscreen long enough to be noticed and then read. Unimportant messages can be cleared sooner than critical ones. Short messages can go sooner than long ones. Messages that come up frequently and are easily recognized can be displayed for less time than rare ones.
Be suspicious when the same message sometimes displays briefly and sometimes lasts longer. This may reflect unanticipated race conditions. Try to figure out how to obtain which delays, then get this looked into carefully by the programmer.
Display layout
The screen should look organized. It should not be cluttered. Different classes of objects should be displayed separately, in predictable areas. There are many guidelines for display layout, but they boil down to this: it should be easy to find what you want on the screen.
Poor aesthetics in the screen layout
The screen may be unbalanced, rows or columns may not be aligned, or it might just look "bad." Use your good taste. If you're not confident, get a second opinion. If it looks badly laid out to you, something is probably wrong even if you can't articulate the problem yet.
Menu layout errors
Schneiderman (1987) and Smith and Mosier (1984) cover this area well. Their discussions run to many more pages than are available here. Here are a few points that we want to emphasize:
- Similar or conceptually related menu choices should be grouped. Groups should be clearly separated.
- The action required to select a menu item should be obvious or should be stated onscreen.
- Menu selections should generally be independent. To achieve a single result, the customer shouldn't have to make two or more different selections on different menus.
- Selecting a menu item by typing its first letter is usually better than selecting it by a number. However, all items have to start with different letters for this to work well. Watch out for assignments of odd names to menu items.
Dialog box layout errors
For further reference in this area, we recommend IBM's SAA Advanced Interface Design Guide (1989), and SAA Basic Interface Design Guide (1989), and Apple's Human Interface Guidelines (1987).
- Dialog boxes should operate consistently. For example, they should use consistent capitalization, spelling, and text justification rules.Dialog box titles should occupy a consistent place and match the name of the command used to call up the dialog. The same shortcut keys should work from dialog to dialogó shouldn't cancel out of some dialogs (no changes made) but complete (all changes accepted) others.
- Controls in the dialog box must be arranged logically. Group related controls together, and separate groups by using proper spacing.
- Selection and entry fields should be aligned vertically and horizontally so users can navigate the cursor movement in a straight-line pattern.
- Watch out for interdependencies across dialogs. It is confusing when a selection in one dialog box determines which options are available in another box.
Obscured instructions
You should always know where to look to find out what to do next. If the screen is at all crowded, an area should be reserved for commands and messages. Once you understand this convention, you'll know where to focus attention. It may also be good to blaze critical information across the center of the screen, no matter what used to be there.
Misuse of flash
Flashing pictures or text are noticeable; lots of flashing is confusing and intimidating. You should be able to tell immediately why an object is flashing. Excessive or ambiguous flashing is an eyesore, not a good alert.
Misuse of color
Too much color can be distracting, make the text harder to read, and increase eye strain. Color shouldn't make the screen look busy or distracting. Programs like word processors, spreadsheets, and databases should use highlighting colors sparingly. Most text should be one color. You should also complain if the program's combinations of colors looks ugly.
Heavy reliance on color
Programs limit their audience severely if they use color as the only differentiator between items. What happens to a colorblind person or someone with a monochrome monitor? A few applications may not be worth running in monochrome, but many others (including drawing programs and many games) don't need color.
Inconsistent with the style of the environment
If the style associated with a computer offers certain consistencies and conveniences, you'll notice their absence from any one program. Even if the programmer thinks he can replace them with "better" ones, many people will resent having to learn a new series of conventions. For example, if the operating system and most applications are mouse and icon-based, an application that requires typed command words will feel inappropriate. New programs should also follow the lead when most others display error messages in a certain way, on a certain place onscreen.
Cannot get rid of onscreen information
It's great (often essential) to have a menu of command choices available on part of the screen. However, once you become proficient with the program, the menu is a waste of screen space. You should be able to issue a command to get rid of it, and another to call it back as needed.
COMMAND STRUCTURE AND ENTRY
This section deals with the way the program organizes commands and presents them to you, and with how you enter those commands. Schneiderman (1987) considers how to choose among the different command entry styles. There are many choices. This section assumes that the programmer's choice of style was reasonable. It deals only with flaws in the implementation.
Inconsistencies
Increasing the number of always-true rules shortens learning time and documentation and makes the program more professional-looking. Inconsistencies are so common because it takes planning and agony to choose a rule of operation that can always be followed. It is so tempting to do things differently now and again. Each minor inconsistency seems insignificant, but together they quickly make an otherwise well conceived product hard to use. It is good testing practice to flag all inconsistencies, no matter how minor.
"Optimizations"
Programmers deliberately introduce inconsistencies to optimize a program. Optimizations are tempting since they tailor the program to your most likely present need. But each new inconsistency brings complexity with it. Make the programmer aware of the tradeoff in each case. Is saving a keystroke or two worth the increase in learning time or the decrease in trust? Usually not.
Inconsistent syntax
Syntactic details should be easily learned. You should be able to stop thinking about them. Syntax of all commands should be consistent throughout the program. Syntax includes such things as:
- the order in which you specify source and destination locations (copy from source to destination or copy to destination from source)
- the type of separators used (spaces, commas, semicolons, slashes, etc.)
- the location of operators (infix (A+B), prefix (+AB), postfix (AB+)).
Inconsistent command entry style
You can select a command by pointing to it, pressing a function key, or typing its name, abbreviation, or number. A program should use one command style. If the program offers alternative styles for doing the same task, it should offer the same alternatives everywhere. If the program must switch styles across different parts of the program, it must make it clear when to use which.
Inconsistent abbreviations
Without clear-cut abbreviation rules, abbreviations can't be easily remembered. Abbreviating delete to del but list to ls and grep to grep makes no sense. Each choice is fine individually, but the collection is an ill-conceived mess of special cases.
Inconsistent termination rule
Fill-in-the-blanks forms only allow so much room for a command name or data item. Suppose an entry can be eight characters long. When you enter seven characters or less, you have to say you're done by pressing or some other terminator (, , , etc.). If you enter an eight character name, some programs act on it without waiting for the terminator. This is confusing.
The program should require terminators for multi-key entries. People rarely remember commands (or data) in terms of how many letters they are. They will habitually enter the eighth character then press . If the program has already supplied its own , the extra one typed by the user is an annoying input "error."
Inconsistent command options
If an option makes sense for two commands, it should be available with both (or neither), should have the same name, and should be invoked in the same sequence in both cases.
Similarly named commands
It is easier to confuse two different commands if their names are similar.
Inconsistent capitalization
If command entry is case sensitive, first letters of all commands should all be capitalized or none should be. First letters of embedded words in commands should always or never be capitalized.
Inconsistent menu position
It's hard to keep the same command in the same position on different menus if it occurs in many submenus, but with work and care the programmer can frequently achieve this.
Inconsistent function key usage
The meaning of function keys should remain constant across the program. Reversals (sometimes saves data and deletes, other times deletes and saves) are unacceptable.
Inconsistent error handling rules
When the program detects an error, it may announce it or attempt to correct it. After handling the error, the program may stop, restart, or return to its last state. The error handler may change data on disk or save new information. Error handlers can vary a great deal. The behavior of any one program's should be completely predictable.
Inconsistent editing rules
The same keys and commands should be available to change any datum as you entered it or examined it later.
Inconsistent data saving rules
The program should save data in the same way everywhere, with the same timing and scope. It shouldn't sometimes save data as each field is entered, but other times save at the end of a record, a group of records, or just before exit.
Time-wasters
Programs that seem designed to waste your time infuriate people.
Garden paths
A program leads you down the garden path if you must make choice after choice to get to a desired command, only finding at the end that it's not there, wasn't implemented, or can't be used unless you do something else (down a different path) first. Look for these problems in complex menu trees.
Choices that can't be taken
There is no excuse for including choices in a menu that cannot be made. How can you view, save or erase the data if there are no data? How can you print the document if there is no printer? How dare the programmer say, Press for Help, then when you press it say, Sorry, Help Not Available at this Level (whatever that means)?
Are you really, really sure?
Programs should ask you to confirm critically destructive commands. You should have to tell the computer twice to reformat a data-filled disk. The program should not pester you for confirmation of every little deletion. You become annoyed; you'll learn to answer Yes automatically, even in the critical cases when you should think first.
Obscurely or idiosyncratically named commands
Command names should be informative. You should not have to constantly look up the definition of a command name in the manual, until you eventually memorize it. There is no excuse for names like grep, finger, and time hog in products released to the public.
Menus
Menus should be simple, but they become complex when there are poor icons or command names and when choices hide under nonobvious topic headings. The more commands a menu covers, the more complex it will be no matter how well planned it is. But without planning, complex menus can become disasters.
Excessively complex menu hierarchy
If you have to wade through menu after menu before finally reaching the command you want, you'll probably want to use another program. Programmers who create deep menu trees cite the design rule that says no menu should have more than seven choices. This may be best for novices. Experienced users prefer many more choices per menu level, make fewer errors and respond more quickly, so long as the choices are well organized, neatly formatted, and not ridiculously crowded or abbreviated. Paap & Roske-Hostrand (1986) and MacGregor, Lee, & Lam (1986) provide some interesting discussion.
Inadequate menu navigation options
In even a modestly deep menu structure, you must be able to move back to the previous menu, move to the top of the menu structure, and leave the program at any time. If there are hundreds of topics, you should also be able to jump directly to any topic by entering its name or number.
Too many paths to the same place
The program needs reorganization if many commands reappear in many menus. It can be handy to have a command repeated in different places, but there are limits. You should worry about the program's internal structure and reliability if it feels like you can get anywhere from anywhere.
You can't get there from here
Some programs lock you out of one set of commands once you've taken a different path. You have to restart to regain access to them. This is usually unnecessary.
Related commands relegated to unrelated menus
Grouping commands or topics in a complex menu is not easy. It is easy to overlook what should be an obvious relationship between two items, and to arbitrarily assign them to separate menus. When you report these, explain the relationship between the two items, and suggest which menu both belong in.
Unrelated commands tossed under the same menu
Some commands are dumped under a totally unrelated heading because someone thought it would take too much work to put them where they belong, which may involve adding a new higher level heading and reorganizing.
Command lines
It is harder to type the command name without a hint than to recognize it in a menu, but experienced users prefer command line entry when there are many commands and many options. The menu system feels too bulky to them. Anything that makes it easier to remember the command names and options correctly is good. Anything that makes errors more likely is bad.
Forced distinction between uppercase and lowercase
Some programs won't recognize a correctly spelled command name if it isn't capitalized "correctly." This is more often a nuisance than a feature.
Reversed parameters
The most common example is the distinction between source and destination files. Does COPY FILE1 FILE2 mean copy from FILE1 to FILE2 or from FILE2 to FILE1? The order doesn't matter (people can get used to anything) as long as it stays consistent across all commands that use a source and a destination file. Application programs must follow the operating system's ordering conventions.
Full command names not allowed
Abbreviations are fine, but you should always be able to type delete, not just del. The full name of a command is much more reliably remembered than the abbreviation, especially if there are no consistent abbreviating rules.
Abbreviations not allowed
You should be able to enter del instead of having to type delete in full. Enough systems don't allow abbreviations that we can't class their absence a design error, but, implemented properly, it sure is a nice feature.
Demands complex input on one line
Some programs require complicated command specifications (Do X for all cases in which A or B and C is true unless D is false). You will make many mistakes if you have to specify compound logical operators as part of a one-line command. Fill-in-the-blank choices, sequential prompting, and query by example, are all more appropriate than command line entry with compound logical scope definition.
No batch input
You should be able to type and correct a list of commands using an editor, then tell the computer to treat this list as if you had typed each command freshly at the keyboard.
Can't edit commands
You should be able to backspace while typing a command. If you try to execute an incorrectly typed command, you should be able to call it back, change the erroneous piece, and re-execute it.
Inappropriate use of the keyboard
If a computer comes with a standard keyboard, with labeled function keys that have standard meanings, new programs should meet that standard.
Failure to use cursor, edit, or function keys
It doesn't matter if a program was ported from some other machine that doesn't have these keys. Users of this machine will expect their keys to work.
Non-standard use of cursor and edit keys
The keys should work the way they usually work on this machine, not the way they usually work on some other machine, and definitely not in some totally new way.
Non-standard use of function keys
If most other programs use as the Help key, defining it as Delete-File-And-Exit in this program is crazy or vicious.
Failure to filter invalid keys
The program should trap and discard invalid characters, such as letters if it adds numbers. It should not echo or acknowledge them. Ignoring them is less distracting than error messages.
Failure to indicate keyboard state changes
Lights on the keyboard or messages on the screen should tell you when Caps Lock and other such state changing features are on.
Failure to scan for function or control keys
You should be able to tell the computer to quit what it's doing (, for example). The program should also always recognize any other system-specific keys, such as , that programs on this machine usually recognize quickly.
MISSING COMMANDS
This section discusses commands or features that some programs don't, but should, include.
State transitions
Most programs move from state to state. The program is in one state before you choose a menu item or issue a command. It moves into another state in response to your choice. Programmers usually test their code well enough to confirm that you can reach any state that you should be able to reach. They don't always let you change your mind, once you've chosen a state.
Can't do nothing and leave
You should be able to tell an interactive program that you made your last choice by mistake, and go back to its previous state.
Can't quit mid-program
You should be able to quit while using a program without adversely affecting stored data. You should be able to stop editing or sorting a file, and revert to the version that was on disk when you started.
Can't stop mid-command
It should be easy to tell the program to stop executing a command. It shouldn't be hard to return to your starting point, to make a correction or choose a different command.
Can't pause
Some programs limit the time you have to enter data. When the time is up, the program changes state. It might display help text or accept a displayed "default" value, or it may log you off. Although time limits can be useful, people do get interrupted. You should be able to tell it that you are taking a break, and when you get back you'll want it in the same state it's in now.
Disaster prevention
System failures and user errors happen. Programs should minimize the consequences of them.
No backup facility
It should be easy to make an extra copy of a file. If you're changing a file, the computer should keep a copy of the original (or make it easy for you to tell it to keep it) so you have a known good version to return to if your changes go awry.
No undo
Undo lets you retract a command, typically any command, or a group of them. Undelete is a restricted case of undo that lets you recover data deleted in error. Undo is desirable. Undelete is essential.
No "Are you sure?"
If you issue a command that will wipe out a lot of work, or that wipes out less but is easy to issue in error, the program should stop you and ask whether you want the command executed.
No incremental saves
When entering large amounts of text or data, you should be able to tell the program to save your work at regular intervals. This ensures that most of your work will have been saved if the power fails. Several programs automatically save your work in progress at regular intervals. This is an excellent feature, so long as the customer who is bothered by the delay during saving can turn it off.
Error handling by the user
People can catch their own errors and recognize from experience that they are prone to others. They should be able to fix their work and, as much as possible, build in their own error checks.
No user-specifiable filters
When designing data entry forms, spreadsheet templates, etc., you should be able to specify, for each field, what types of data are valid and what the program should ignore or reject. As examples, you might have the program reject anything that isn't a digit, a letter, a number within a certain range, a valid date, or an entry that matches an item in a list stored on disk.
Awkward error correction
It should be easy to fix a mistake. You should never have to stop and restart a program just to return to a data entry screen where you made an error. You should always be able to back up the cursor to a field on the same screen in which you entered, or could have entered, data. When entering a list of numbers, you should be able to correct one without redoing the rest.
Can't include comments
When designing data entry forms, spreadsheet templates, expert systems, anything in which you are, in effect, writing a program, you should be able to enter notes for future reference and debugging.
Can't display relationships between variables
Variables in entry forms, spreadsheet templates, etc., are related. It should be easy to examine the dependence of any variable on the values of others.
Miscellaneous nuisances
Inadequate privacy or security
How much security is needed for a program or its data varies with the application and the market. On multi-user systems you should be able to hide your files so no one else can see them, and encrypt them so no one else, not even the system administrator can read them. You should also be able to lock files so no one else can change (or delete) them. Beizer (1984) discusses security in more detail.
Obsession with security
The security controls of a program should be as unobtrusive as possible. If you are working at your own personal computer at home, you should be able to stop a program from pestering you for passwords.
Can't hide menus
Many programs display a menu at the top, bottom, or side of the screen. They use the rest of the screen for data entry and manipulation. The menus are memory aids. Once you know all the commands you need, you should be able to remove the menus and use the full screen for entry and editing.
Doesn't support standard O/S features
For example, if the operating system uses subdirectories, program commands should be able to reference files in other subdirectories. If the O/S defines "wildcard" characters (such as * to match any group of characters), the program should recognize them.
Doesn't allow long names
Years ago, when memory was scarce and compilers were sluggish, it was necessary to limit the length of file and variable names to six or eight characters. We're past those days. Meaningful names are among the best possible forms of documentation. They should be allowed.
PROGRAM RIGIDITY
Some programs are very flexible. You can change minor aspects of their functioning easily. You can do tasks in any order you want. Other programs are utterly inflexible. Rigidity isn't always bad. The fewer choices and the more structured the task, the more easily (usually) you'll learn the program. And you won't be confused by aspects of a program's operation that can't be changed without affecting the others. On the other hand, different people will like different aspects of the program and dislike others. If you change these to suit your taste, you'll like the program more.
User tailorability
You should be able to change minor and arbitrary aspects of the program's user interface with a minimum of fuss and bother.
Can't turn off the noise
Many programs beep when you make errors and provide a loud key click that sounds every time you touch the keyboard. Auditory feedback is useful but in shared work areas, computer noises can be annoying. There must be a way to turn them off.
Can't turn off case sensitivity
A system that can distinguish between uppercase and lowercase should allow you to tell it to ignore cases.
Can't tailor to hardware at hand
Some programs are locked to input/output devices that have specific, limited capabilities. People who upgrade their equipment either can't use these programs or can't take advantage of the new devices' features. Experienced users should be able to tailor a program to the hardware. You should be able to change control codes sent to a printer and copy a program onto any mass storage device. It should not be impossible to use a mouse with any interactive program.
Can't change device initialization
An application program should either be able to send user-defined initializers or it should leave well enough alone. Suppose you want to send control codes to a printer to switch to condensed characters. If the program that prints the data doesn't let you initialize the printer, you have to change the printer mode from the device, then run the program. Some programs, however, defeat your printer setup by always sending the printer their own, inflexible, set of control codes. This is a design error.
Can't turn off automatic saves
Some programs protect you against power failures by automatically saving entered data to disk periodically. In principle this is great but in practice the pauses while it saves the data can be disruptive. Also, the program assumes that you always want to save your data. This assumption might not be true. You should be able to turn this off.
Can't slow down (speed up) scrolling
You should be able to slow down the screen display rate so you can read text as it scrolls by.
Can't do what you did last time
Can't find out what you did last time
You should be able to re-issue a command, examine it, or edit it.
Failure to execute a customization command
If the program lets you change how it interacts with you, your changes should take effect immediately. If a restart is unavoidable, the program should say so. You should not have to wonder why a command wasn't executed.
Failure to save customization commands
You should not only be able to tell the computer to turn off its beeps and clicks now, but should also be able to tell it to turn them off and keep them off forever.
Side-effects of feature changes
Changing how one feature operates should not affect another. When there are side effects, they should be well documented when you change the feature setting, in the manual and onscreen.
Infinite tailorability
You can change virtually all aspects of some programs. This flexibility can be good, but you have to step back from the program to figure out how it should work. To make the decisions intelligently you have to develop an expert user's view of the program itself along with learning the command language.
Programs this flexible usually have horrid user interfaces. The developers spent their energy making the program adjustable, and didn't bother making the uncustomized product any good. Since everyone will change it, they reason, its initial command set doesn't mean anything anyway. Such a program is terrible for novices and occasional users. It isn't worth their time (sometimes weeks or months) to figure out how to tune the program to their needs, but without the tuning, the program is only marginally usable.
The user interface of customizable products should be fully usable without modification. You should subject it to the same rigorous criticism applied to less flexible ones. Many people will use the uncustomized version for a long time.
Who's in control
Some programs are high-handed. Their error and help messages are condescending. Their style is unforgiving, you can't abort commands or change data after entering them. None of this is acceptable. Programs should make it easier and more pleasant for you to get a task done as quickly as possible. They should not second-guess you, force a style on you, or waste your time.
Unnecessary imposition of a conceptual style
Some programs demand that you enter data in a certain order, that you complete each task before moving to the next, that you make decisions before looking at their potential consequences. Examples:
- When designing a data entry form, why must you specify a field's name, type, width, or calculation order, before drawing it onscreen? As you see how the different fields look together, won't you change some fields, move them around, even get rid of a few? You may have to enter field specifications before using the form, but subject to that restriction, you should decide when to fill in the details.
- When describing tasks to a project management system, why must you list all tasks first, all available people second, then completely map the work assigned to one individual before entering any data for the next? Since you're probably trying to figure out what to assign to whom, won't you want to change these data as you see their consequences?
A surprising number of limits exist because some programmer decided that people should organize their work in a certain way. For "their own good" he won't let them deviate from this "optimal" approach. He is typically wrong.
Novice-friendly, experienced-hostile
Programs optimized for novices break tasks into many small, easily understood steps. This can be good for the newcomer, but anyone experienced with the system will be frustrated if they can't get around it.
Artificial intelligence and automated stupidity
In the names of "artificial intelligence" and "convenience" some programs guess what you want next and execute those guesses as if they were user-issued commands. This is fine unless you don't want them done. Similarly, the program that automatically corrects errors is great until it "corrects" correct data. People make enough of their own mistakes without having to put up with ones made by a program that's trying to second-guess them. Better than automatic execution, especially of something that takes noticeable time or changes data, the program should give you a choice. You should be able to set it to wait until you type Y (yes) before executing its suggestions. If you say No the program should abandon its suggestion and ask for new input.
Superfluous or redundant information required
Some programs ask for information they'll never use, or that they'll only use to display onscreen once, or ask you to re-enter data you've already entered, not to check it against the old copy, just to get it again. This is a surprisingly common waste of time.
Unnecessary repetition of steps
Some programs make you re-enter the works if you make one mistake in a long sequence of command steps or data. Others force you to re-enter or confirm any command that might be in error. To do something "unusual" you may have to confirm every step. Repetitions or confirmations that are not essential are a waste of your time.
Unnecessary limits
Why restrict a database to so many fields or records, a spreadsheet cell to digits only, a project manager to so many tasks, a word processor to so many characters? Limits that aren't essential for performance or reliability shouldn't be limits.
PERFORMANCE
Many experienced users consider performance the most important aspect of usability: with a fast program, they feel more able to concentrate and more in control. Errors are less important because they can be dealt with quickly. With few exceptions, reviewed by Schneiderman (1987), the faster the better.
Performance has different definitions, such as:
- Program Speed: how quickly the program does standard tasks. For example, how quickly does a word processor move to the end of the file?
- User Throughput: how quickly you can do standard tasks with the program. These are larger scale tasks. For example, how long does it take to enter and print a letter?
- Perceived Performance: How quick does the program seem to you?
Program speed is a big factor no matter how you define performance, but a fast program with a poorly designed user interface will seem much slower than it should.
Slow program
Many design and code errors can slow a program. The program might do unnecessary work, such as initializing an area of memory that will be overwritten before being read. It might repeat work unnecessarily, such as doing something inside a loop that could be done outside of it. Design decisions also slow the program, often more than the obvious errors. Whatever the reason for the program being slow, if it is, it's a problem. Delays as short as a quarter of a second can break your concentration, and substantially increase your time to finish a task.
Slow echoing
The program should display inputs immediately. If you notice a lag between the time you type a letter and the time you see it, the program is too slow. You will be much more likely to make mistakes. Fast feedback is essential for any input event, including moving mice, trackballs, and light pens.
How to reduce user throughput
A lightning quick program will be molasses for getting things done if it slows the person working with it. This includes:
- anything that makes user errors more likely.
- slow error recovery, such as making you re-enter everything if you make a mistake when entering a long series of numbers or a complicated command.
- anything that gets you so confused that you have to ask for help or look in the manual.
- making you type too much to do too little: no abbreviations, breaking a task into tiny subtasks, requiring confirmation of everything, and so on.
Other sections of the Appendix describe specific errors along these lines. One tactic for applying pressure to fix user interface errors involves a comparative test of user throughput. Compare the product under development with a few competitors. If people take longer to do things with your program, and if much of the delay can be attributed to user interface design errors, these errors take on a new significance.
Poor responsiveness
A responsive program doesn't force you to wait before issuing your next command. It
constantly scans for keyboard (or other) input, acknowledges commands quickly, and assigns
them high priority. For example, type a few lines of text while your word processor is
reformatting the screen. It should stop formatting, echo the input, format the display of these
lines as you enter them, and execute your editing commands. It should keep the area of the
screen near the cursor up to date. The rest is lower priority since you aren't working with it at
this instant. The program can update the rest of the display when you stop typing.
Responsive programs feel faster.
No type-ahead
A program that allows type-ahead lets you keep typing while it goes about other business. It remembers what you typed and displays and executes it later. You should not have to wait to enter the next command.
No warning that an operation will take a long time
The program should tell you if it needs more than a few seconds to do something. You should be able to cancel the command. For long jobs, it should tell you how long so you can use the time rather than waste it waiting.
No progress reports
For long tasks or delays, it is very desirable to indicate how much has been done and how much longer the machine will be tied up (Myers, 1985).
Problems with time-outs
Some programs limit the time you have to enter data. When the time's up, the program changes state. Except for arcade games, you should not have to race against a short time-out to stop the program from executing an undesired command.
Time-outs can also be too long. For example, you might have a short interval before a program does something reasonable but time consuming. If you respond during the interval, you shortcut the task (e.g., a menu isn't displayed unnecessarily). Program speed will suffer if this isn't kept short.
Time-out intervals can simultaneously be too long for people who wait them out and too short for others who try to enter data during them.
Program pesters you
BEEP! Are you sure?
BEEP! Your disk is now 85% full. Please swap disks shortly.
BEEP! Are you really sure?
BEEP! You haven't saved your text for the last hour.
BEEP! Your disk is now 86% full. Please swap disks shortly.
BEEP! Please re-enter your security code.
BEEP! You haven't entered anything for 10 minutes. Please log off.
BEEP! Your disk is now 86% full. Please swap disks shortly.
BEEP! 14 messages still unanswered.
Reminders, alerts, and queries can be useful but if they are frequent they can be annoying.
Do you really want help and graphics at 300 baud?
On a slow terminal, help text, long menus, and pretty pictures are usually irritating. You should be able to use a terse command language instead. Similarly, programs that format output in a beautiful but time-consuming way on a printer should have a "fast and ugly" mode for drafts.
OUTPUT
Program output should be as complete as desired and intelligible to the human or program that must read it. It should include whatever you want, in whatever format you want. It should go to whatever device you want it to go. These requirements are real, but too general to (usually) achieve. You'll probably add many other annoyances and limitations to the following list.
Can't output certain data
You should be able to print any information you enter, including technical entries like formulas in spreadsheets and field definitions. If it's important enough to enter, it's important enough for you to be able to print and desk check it.
Can't redirect output
You should be able to redirect output. In particular, you should be able to send a long "printout" to the disk and print this disk file later. This gives you the opportunity to polish the output file with a word processor or print it with a program that can print files more quickly or as a background task.
The program should not stop you from sending output to unexpected devices, such as plotters, laser printers, and cassette tapes.
Format incompatible with a follow-up process
If one program is supposed to be able to save data in a format that a second can understand, you must test that it does. This means buying or borrowing a copy of the second program, saving the data with the first, reading it with the second, and looking at what the second program tells you it got. This test is all too often forgotten, especially when the second program was not made by the company that developed the first.
Must output too little or too much
You should be able to modify reports to present only the information you need. Having to dig through pages of printouts that contain just a few lines of useful information is nearly as bad as not getting the information.
Can't control output layout
You should be able to emphasize information by changing fonts, boldfacing, underlining, etc. You should also be able to control the spacing of information; that is, you should be able to group some sections of information and keep others separate. At a minimum, the program should be able to print reports to a disk file, in a format suitable for touchups by a word processor.
Absurd printed level of precision
It is silly to say that 4.2 + 3.9 is 8.100000 or that the product of 4.234 and 3.987 is 16.880958. In final printouts, the program should round results to the precision of the original data, unless you tell it to do otherwise.
Can't control labeling of tables or figures
You should be able to change the typeface, wording, and position of any caption, heading, or other text included in a table, graph or chart. Can't control scaling of graphs Graphing programs should provide default vertical and horizontal scales, but you should be able to override the defaults.
|