Include Termios H

Posted onby

Let’s try and read keypresses from the user. (The lines you need to add arehighlighted and marked with arrows.)

Windows does terminal I/O with a very different model from any.nix system. As a result, there really is no direct equivalent to the termios.h header and its friends. You want to read at MSDN about the Windows Communications Resources. Some things to learn more about include.

Step 3
  1. The problem is that in there are no such functions, and I need to include.h for being able to use them. Problem: If I include both headers ( and ) the compiler will scream about functions and structure re-declaration, and he's right.
  2. The problem is that in termios.h there are no such functions, and I need to include for being able to use them. Problem: If I include both headers (termios.h and termios.h) the compiler will scream about functions and structure re-declaration, and he's right.
  3. Termios.h - define values for termios SYNOPSIS #include DESCRIPTION The termios.h header contains the definitions used by the terminal I/O interfaces (see the XBD specification, General Terminal Interface for the structures and names defined). The termios Structure The following data types are defined through typedef: cct.

read() and STDIN_FILENO come from <unistd.h>. We are asking read() toread 1 byte from the standard input into the variable c, and to keep doingit until there are no more bytes to read. read() returns the number of bytesthat it read, and will return 0 when it reaches the end of a file.

When you run ./kilo, your terminal gets hooked up to the standard input, andso your keyboard input gets read into the c variable. However, by defaultyour terminal starts in canonical mode, also called cooked mode. Inthis mode, keyboard input is only sent to your program when the user pressesEnter. This is useful for many programs: it lets the user type in aline of text, use Backspace to fix errors until they get their inputexactly the way they want it, and finally press Enter to send it tothe program. But it does not work well for programs with more complex userinterfaces, like text editors. We want to process each keypress as it comes in,so we can respond to it immediately.

What we want is raw mode. Unfortunately, there is no simple switch you canflip to set the terminal to raw mode. Raw mode is achieved by turning off agreat many flags in the terminal, which we will do gradually over the course ofthis chapter.

To exit the above program, press Ctrl-D to tell read() that it’sreached the end of file. Or you can always press Ctrl-C to signalthe process to terminate immediately.

Press q to quit?

To demonstrate how canonical mode works, we’ll have the program exit when itreads a q keypress from the user. (Lines you need to change arehighlighted and marked the same way as lines you need to add.)

Step 4

To quit this program, you will have to type a line of text that includes a qin it, and then press enter. The program will quickly read the line of text onecharacter at a time until it reads the q, at which point the while loopwill stop and the program will exit. Any characters after the q will be leftunread on the input queue, and you may see that input being fed into your shellafter your program exits.

Turn off echoing

We can set a terminal’s attributes by (1) using tcgetattr() to read thecurrent attributes into a struct, (2) modifying the struct by hand, and(3) passing the modified struct to tcsetattr() to write the new terminalattributes back out. Let’s try turning off the ECHO feature this way.

Step 5

struct termios, tcgetattr(), tcsetattr(), ECHO, and TCSAFLUSH allcome from <termios.h>.

The ECHO feature causes each key you type to be printed to the terminal, soyou can see what you’re typing. This is useful in canonical mode, but reallygets in the way when we are trying to carefully render a user interface in rawmode. So we turn it off. This program does the same thing as the one in theprevious step, it just doesn’t print what you are typing. You may be familiarwith this mode if you’ve ever had to type a password at the terminal, whenusing sudo for example.

After the program quits, depending on your shell, you may find your terminal isstill not echoing what you type. Don’t worry, it will still listen to what youtype. Just press Ctrl-C to start a fresh line of input to yourshell, and type in reset and press Enter. This resets yourterminal back to normal in most cases. Failing that, you can always restartyour terminal emulator. We’ll fix this whole problem in the next step.

Terminal attributes can be read into a termios struct by tcgetattr(). Aftermodifying them, you can then apply them to the terminal using tcsetattr().The TCSAFLUSH argument specifies when to apply the change: in this case, itwaits for all pending output to be written to the terminal, and also discardsany input that hasn’t been read.

The c_lflag field is for “local flags”. A comment in macOS’s <termios.h>describes it as a “dumping ground for other state”. So perhaps it should bethought of as “miscellaneous flags”. The other flag fields are c_iflag (inputflags), c_oflag (output flags), and c_cflag (control flags), all of whichwe will have to modify to enable raw mode.

ECHO is a bitflag, defined as00000000000000000000000000001000 in binary. We use the bitwise-NOT operator(~) on this value to get 11111111111111111111111111110111. We thenbitwise-AND this value with the flags field, which forces the fourth bit in theflags field to become 0, and causes every other bit to retain its currentvalue. Flipping bits like this is common in C.

Disable raw mode at exit

Let’s be nice to the user and restore their terminal’s original attributes whenour program exits. We’ll save a copy of the termios struct in its originalstate, and use tcsetattr() to apply it to the terminal when the programexits.

Include
Step 6

atexit() comes from <stdlib.h>. We use it to register ourdisableRawMode() function to be called automatically when the program exits,whether it exits by returning from main(), or by calling the exit()function. This way we can ensure we’ll leave the terminal attributes the waywe found them when our program exits.

We store the original terminal attributes in a global variable, orig_termios.We assign the orig_termios struct to the raw struct in order to make a copyof it before we start making our changes.

You may notice that leftover input is no longer fed into your shell after theprogram quits. This is because of the TCSAFLUSH option being passed totcsetattr() when the program exits. As described earlier, it discards anyunread input before applying the changes to the terminal. (Note: This doesn’thappen in Cygwin for some reason, but it won’t matter once we are reading inputone byte at a time.)

Turn off canonical mode

There is an ICANON flag that allows us to turn off canonical mode. This meanswe will finally be reading input byte-by-byte, instead of line-by-line.

Step 7

ICANON comes from <termios.h>. Input flags (the ones in the c_iflagfield) generally start with I like ICANON does. However, ICANON is not aninput flag, it’s a “local” flag in the c_lflag field. So that’s confusing.

Now the program will quit as soon as you press q.

Display keypresses

To get a better idea of how input in raw mode works, let’s print out each bytethat we read(). We’ll print each character’s numeric ASCII value, as well asthe character it represents if it is a printable character.

Step 8

iscntrl() comes from <ctype.h>, and printf() comes from <stdio.h>.

iscntrl() tests whether a character is a control character. Controlcharacters are nonprintable characters that we don’t want to print to thescreen. ASCII codes 0–31 are all control characters, and 127 is also acontrol character. ASCII codes 32–126 are all printable. (Check out theASCII table to see all of the characters.)

printf() can print multiple representations of a byte. %d tells it toformat the byte as a decimal number (its ASCII code), and %c tells it towrite out the byte directly, as a character.

This is a very useful program. It shows us how various keypresses translateinto the bytes we read. Most ordinary keys translate directly into thecharacters they represent. But try seeing what happens when you press the arrowkeys, or Escape, or Page Up, or Page Down, orHome, or End, or Backspace, orDelete, or Enter. Try key combinations withCtrl, like Ctrl-A, Ctrl-B, etc.

You’ll notice a few interesting things:

  • Arrow keys, Page Up, Page Down, Home, andEnd all input 3 or 4 bytes to the terminal: 27, '[', and thenone or two other characters. This is known as an escape sequence. Allescape sequences start with a 27 byte. Pressing Escape sends asingle 27 byte as input.
  • Backspace is byte 127. Delete is a 4-byte escapesequence.
  • Enter is byte 10, which is a newline character, also known as'n'.
  • Ctrl-A is 1, Ctrl-B is 2, Ctrl-C is…oh, that terminates the program, right. But the Ctrl keycombinations that do work seem to map the letters A–Z to the codes1–26.

By the way, if you happen to press Ctrl-S, you may find your programseems to be frozen. What you’ve done is you’ve asked your program to stopsending you output. PressCtrl-Q to tell it to resume sending you output.

Also, if you press Ctrl-Z (or maybe Ctrl-Y), your programwill be suspended to the background. Run the fg command to bring it back tothe foreground. (It may quit immediately after you do that, as a result ofread() returning -1 to indicate that an error occurred. This happens onmacOS, while Linux seems to be able to resume the read() call properly.)

Turn off Ctrl-C and Ctrl-Z signals

By default, Ctrl-C sends a SIGINT signal to the current processwhich causes it to terminate, and Ctrl-Z sends a SIGTSTP signal tothe current process which causes it to suspend. Let’s turn off the sending ofboth of these signals.

Step 9

ISIG comes from <termios.h>. Like ICANON, it starts with I but isn’t aninput flag.

Now Ctrl-C can be read as a 3 byte and Ctrl-Z can beread as a 26 byte.

This also disables Ctrl-Y on macOS, which is like Ctrl-Zexcept it waits for the program to read input before suspending it.

Disable Ctrl-S and Ctrl-Q

By default, Ctrl-S and Ctrl-Q are used forsoftware flow control.Ctrl-S stops data from being transmitted to the terminal until youpress Ctrl-Q. This originates in the days when you might want topause the transmission of data to let a device like a printer catch up. Let’sjust turn off that feature.

Step 10

IXON comes from <termios.h>. The I stands for “input flag” (which it is,unlike the other I flags we’ve seen so far) and XON comes from the names ofthe two control characters that Ctrl-S and Ctrl-Qproduce: XOFF to pause transmission and XON to resume transmission.

Now Ctrl-S can be read as a 19 byte and Ctrl-Q can beread as a 17 byte.

Disable Ctrl-V

On some systems, when you type Ctrl-V, the terminal waits for you totype another character and then sends that character literally. For example,before we disabled Ctrl-C, you might’ve been able to typeCtrl-V and then Ctrl-C to input a 3 byte. We can turnoff this feature using the IEXTEN flag.

Turning off IEXTEN also fixes Ctrl-O in macOS, whose terminaldriver is otherwise set to discard that control character.

Step 11

IEXTEN comes from <termios.h>. It is another flag that starts with I butbelongs in the c_lflag field.

Ctrl-V can now be read as a 22 byte, and Ctrl-O as a15 byte.

Fix Ctrl-M

If you run the program now and go through the whole alphabet while holding downCtrl, you should see that we have every letter except M.Ctrl-M is weird: it’s being read as 10, when we expect it to beread as 13, since it is the 13th letter of the alphabet, andCtrl-J already produces a 10. What else produces 10? TheEnter key does.

It turns out that the terminal is helpfully translating any carriage returns(13, 'r') inputted by the user into newlines (10, 'n'). Let’s turnoff this feature.

Step 12

ICRNL comes from <termios.h>. The I stands for “input flag”, CR standsfor “carriage return”, and NL stands for “new line”.

Now Ctrl-M is read as a 13 (carriage return), and theEnter key is also read as a 13.

Turn off all output processing

It turns out that the terminal does a similar translation on the output side.It translates each newline ('n') we print into a carriage return followed bya newline ('rn'). The terminal requires both of these characters in orderto start a new line of text. The carriage return moves the cursor back to thebeginning of the current line, and the newline moves the cursor down a line,scrolling the screen if necessary. (These two distinct operations originated inthe days of typewriters andteletypes.)

We will turn off all output processing features by turning off the OPOSTflag. In practice, the 'n' to 'rn' translation is likely the only outputprocessing feature turned on by default.

Step 13

OPOST comes from <termios.h>. O means it’s an output flag, and I assumePOST stands for “post-processing of output”.

If you run the program now, you’ll see that the newline characters we’reprinting are only moving the cursor down, and not to the left side of thescreen. To fix that, let’s add carriage returns to our printf() statements.

Step 14

From now on, we’ll have to write out the full 'rn' whenever we wantto start a new line.

Miscellaneous flags

Let’s turn off a few more flags.

Step 15

Include Termios H Auto Repair

BRKINT, INPCK, ISTRIP, and CS8 all come from <termios.h>.

This step probably won’t have any observable effect for you, because theseflags are either already turned off, or they don’t really apply to modernterminal emulators. But at one time or another, switching them off wasconsidered (by someone) to be part of enabling “raw mode”, so we carry on thetradition (of whoever that someone was) in our program.

As far as I can tell:

  • When BRKINT is turned on, abreak condition willcause a SIGINT signal to be sent to the program, like pressing Ctrl-C.
  • INPCK enables parity checking, which doesn’t seem to apply to modernterminal emulators.
  • ISTRIP causes the 8th bit of each input byte to be stripped, meaning itwill set it to 0. This is probably already turned off.
  • CS8 is not a flag, it is a bit mask with multiple bits, which we set usingthe bitwise-OR () operator unlike all the flags we are turning off. Itsets the character size (CS) to 8 bits per byte. On my system, it’s alreadyset that way.

A timeout for read()

Currently, read() will wait indefinitely for input from the keyboard beforeit returns. What if we want to do something like animate something on thescreen while waiting for user input? We can set a timeout, so that read()returns if it doesn’t get any input for a certain amount of time.

Step 16

VMIN and VTIME come from <termios.h>. They are indexes into the c_ccfield, which stands for “control characters”, an array of bytes that controlvarious terminal settings.

The VMIN value sets the minimum number of bytes of input needed beforeread() can return. We set it to 0 so that read() returns as soon as thereis any input to be read. The VTIME value sets the maximum amount of time towait before read() returns. It is in tenths of a second, so we set it to 1/10of a second, or 100 milliseconds. If read() times out, it will return 0,which makes sense because its usual return value is the number of bytes read.

When you run the program, you can see how often read() times out. If youdon’t supply any input, read() returns without setting the c variable,which retains its 0 value and so you see 0s getting printed out. If youtype really fast, you can see that read() returns right away after eachkeypress, so it’s not like you can only read one keypress every tenth of asecond.

If you’re using Bash on Windows, you may see that read() still blocks forinput. It doesn’t seem to care about the VTIME value. Fortunately, this won’tmake too big a difference in our text editor, as we’ll be basically blockingfor input anyways.

Error handling

enableRawMode() now gets us fully into raw mode. It’s time to clean up thecode by adding some error handling.

First, we’ll add a die() function that prints an error message and exits theprogram.

Step 17

Termios Icanon

perror() comes from <stdio.h>, and exit() comes from <stdlib.h>.

Most C library functions that fail will set the global errno variable toindicate what the error was. perror() looks at the global errno variableand prints a descriptive error message for it. It also prints the string givento it before it prints the error message, which is meant to provide contextabout what part of your code caused the error.

Include Termios H C

After printing out the error message, we exit the program with an exit statusof 1, which indicates failure (as would any non-zero value).

Let’s check each of our library calls for failure, and call die() when theyfail.

Step 18

errno and EAGAIN come from <errno.h>.

tcsetattr(), tcgetattr(), and read() all return -1 on failure, and setthe errno value to indicate the error.

In Cygwin, when read() times out it returns -1 with an errno of EAGAIN,instead of just returning 0 like it’s supposed to. To make it work in Cygwin,we won’t treat EAGAIN as an error.

An easy way to make tcgetattr() fail is to give your program a text file or apipe as the standard input instead of your terminal. To give it a file asstandard input, run ./kilo <kilo.c. To give it a pipe, runecho test ./kilo. Both should result in the same error from tcgetattr(),something like Inappropriate ioctl for device.

Sections

#include Termios.h

That just about concludes this chapter on entering raw mode. The last thingwe’ll do now is split our code into sections. This will allow these diffs to beshorter, as each section that isn’t changed in a diff will be folded into asingle line.

#include Termios.h Windows

Step 19

In the next chapter, we’ll do some more low-levelterminal input/output handling, and use that to draw to the screen and allowthe user to move the cursor around.