ThePOSIX terminal interface is the generalized abstraction, comprising both anapplication programming interface for programs, and a set of behavioural expectations for users of aterminal, as defined by thePOSIX standard and theSingle Unix Specification. It is a historical development from the terminal interfaces of BSD version 4 andSeventh Edition Unix.
A multiplicity of I/O devices are regarded as "terminals" in Unix systems.[1][2] These include:
Unlike its mainframe and minicomputer[citation needed] contemporaries, the original Unix system was developed solely fordumb terminals, and that remains the case today.[6] A terminal is a character-oriented device, comprising streams of characters received from and sent to the device.[6][7] Although the streams of characters are structured, incorporatingcontrol characters,escape codes, and special characters, the I/O protocol is not structured as would be the I/O protocol ofsmart, orintelligent, terminals. There are no field format specifications. There's no block transmission of entire screens (input forms) of input data.
By contrast mainframes and minicomputers in closed architectures commonly useBlock-oriented terminals.
The "capabilities" of a terminal comprise various dumb terminal features that are above and beyond what is available from a pure teletypewriter, which programs can make use of. They (mainly) comprise escape codes that can be sent to or received from the terminal. The escape codes sent to the terminal perform various functions that a CRT terminal (or software terminal emulator) is capable of that a teletypewriter is not, such as moving the terminal's cursor to positions on the screen, clearing and scrolling all or parts of the screen, turning on and off attached printer devices, programmable function keys, changing display colours and attributes (such asreverse video), and setting display title strings. The escape codes received from the terminal signify things such asfunction key,arrow key, and other specialkeystrokes (home key,end key,help key,PgUp key,PgDn key,insert key,delete key, and so forth).[8][9]
These capabilities are encoded in databases that are configured by a system administrator and accessed from programs via theterminfo library (which supersedes the oldertermcap library), upon which in turn are built libraries such as thecurses andncurses libraries. Application programs use the terminal capabilities to providetextual user interfaces with windows, dialogue boxes, buttons, labels, input fields, menus, and so forth.[10][11]
TERM et al.The particular set of capabilities for the terminal that a (terminal-aware) program's input and output uses is obtained from the database rather than hardwired into programs and libraries, and is controlled by theTERMenvironment variable (and, optionally for the termcap and terminfo libraries, theTERMCAP andTERMINFO environment variables, respectively).[10] This variable is set by whateverterminal monitor program spawns the programs that then use that terminal for its input and output, or sometimes explicitly. For example:
TERM environment variable according to a system database (variouslyinittab or the configuration files for thettymon orlaunchd programs) defining what local terminals are attached to what serial ports and what terminal types are provided by local virtual terminals or the local system console.TERM environment variable immediately after login to the correct type. (More usually, the terminal type set by the getty program for the dial-up line, that the system administrator has determined to be used most often by dial-up users with remote terminals, matches the one used by the dial-up user and that user has no need to override the terminal type.)TERM environment variable to the same terminal type as the SSH client.[12]TERM environment variable to specify the type of terminal that it is emulating. Emulated terminals often do not exactly match real terminal hardware, and terminal emulators have type names dedicated for their use. The xterm program (by default) setsxterm as the terminal type, for example.[13] TheGNU Screen program setsscreen as the terminal type.Terminals provide job control facilities. Interactively, the user at the terminal can send control characters that suspend the currently running job, reverting to the interactive job control shell that spawned the job, and can run commands that place jobs in the "background" or that switch another, background, job into the foreground (unsuspending it if necessary).[14][15]
Strictly speaking, in Unixes a terminal device comprises the underlyingtty device driver, responsible for the physical control of the device hardware via I/O instructions and handling device interrupt requests for character input and output, and theline discipline. A line discipline is independent of the actual device hardware, and the same line discipline can be used for aterminal concentrator device responsible for multiple controlling terminals as for a pseudoterminal. In fact, the line discipline (or, in the case of BSD, AIX, and other systems, linedisciplines) are the same across all terminal devices. It is the line discipline that is responsible for local echo, line editing, processing of input modes, processing of output modes, and character mapping. All these things are independent of the actual hardware, dealing as they do in the simple abstractions provided by tty device drivers: transmit a character, receive a character, set various hardware states.[16][17]
InSeventh Edition Unix,BSD systems and derivatives includingmacOS, andLinux, each terminal device can be switched amongst multiple line disciplines.[18] In the AT&TSTREAMS system, line disciplines are STREAMS modules that may be pushed onto and popped off a STREAMS I/O stack.[19]
The POSIX terminal interface is derived from the terminal interfaces of various Unix systems.
The terminal interface provided by Unix 32V and Seventh Edition Unix, and also presented by BSD version 4 as theold terminal driver, was a simple one, largely geared towards teletypewriters as terminals. Input was entered a line at a time, with the terminal driver in the operating system (and not the terminals themselves) providing simple line editing capabilities. A buffer was maintained by the kernel in which editing took place. Applications reading terminal input would receive the contents of the buffer only when thereturn key was pressed on the terminal to end line editing. The@ key sent from the terminal to the system would erase ("kill") the entire current contents of the editing buffer, and would be normally displayed as an '@' symbol followed by a newline sequence to move the print position to a fresh blank line. The# key sent from the terminal to the system would erase the last character from the end of the editing buffer, and would be normally displayed as an '#' symbol, which users would have to recognize as denoting a "rubout" of the preceding character (teletypewriters not being physically capable of erasing characters once they have been printed on the paper).[20][21][22][23][18]
From a programming point of view, a terminal device had transmit and receivebaud rates, "erase" and "kill" characters (that performed line editing, as explained), "interrupt" and "quit" characters (generatingsignals to all of the processes for which the terminal was a controlling terminal), "start" and "stop" characters (used formodem flow control), an "end of file" character (acting like a carriage return except discarded from the buffer by theread() system call and therefore potentially causing a zero-length result to be returned) and variousbasic mode flags determining whetherlocal echo was emulated by the kernel's terminal driver, whether modem flow control was enabled, the lengths of various output delays, mapping for the carriage return character, and the three input modes.[24]
The three input modes were:
In line mode the line discipline performs all line editing functions and recognizes the "interrupt" and "quit" control characters and transforms them into signals sent to processes. Applications programs reading from the terminal receive entire lines, after line editing has been completed by the user pressing return.[21][25]
cbreak mode is one of two character-at-a-time modes. (Stephen R. Bourne jokingly referred to it (Bourne 1983, p. 288) as a "half-cooked" and therefore "rare" mode.) The line discipline performs no line editing, and the control sequences for line editing functions are treated as normal character input. Applications programs reading from the terminal receive characters immediately, as soon as they are available in theinput queue to be read. However, the "interrupt" and "quit" control characters, as well as modem flow control characters, are still handled specially and stripped from the input stream.[26][27]
The programmatic interface for querying and modifying all of these modes and control characters was theioctl() system call. (This replaced thestty() andgtty() system calls of Sixth Edition Unix.)[29][30] Although the "erase" and "kill" characters were modifiable from their defaults of# and@, for many years they were the pre-set defaults in the terminal device drivers, and on many Unix systems, which only altered terminal device settings as part of the login process, in system login scripts that ranafter the user had entered username and password, any mistakes at the login and password prompts had to be corrected using the historical editing key characters inherited from teletypewriter terminals.[23]
With the BSD Unixes camejob control, and anew terminal driver with extended capabilities.[18] These extensions comprised additional (again programmatically modifiable) special characters:
SUB andEM) caused the generation of a newSIGTSTP signal to processes in the terminal's controlling process group.[27]ETB,SYN, andDC2) performed additional line editing functions. "word erase" erased the last word at the end of the line editing buffer. "literal next" allowed any special character to be entered into the line editing buffer (a function available, somewhat inconveniently, in Seventh Edition Unix via the backslash character). "reprint" caused the line discipline to reprint the current contents of the line editing buffer on a new line (useful for when another, background, process had generated output that had intermingled with line editing).[27]The programmatic interface for querying and modifying all of these extra modes and control characters was still theioctl() system call, which its creators (Leffler et al. 1989, p. 262) described as a "rather cluttered interface". All of the original Seventh Edition Unix functionality was retained, and the new functionality was added via additionalioctl() operation codes, resulting in a programmatic interface that had clearly grown, and that presented some duplication of functionality.[31]
System III introduced a new programming interface that combined Seventh Edition's separateioctl() operations to get and set flags and to get and set control characters into calls that used atermio structure to hold both flags and control characters and that could get them in a single operation and set them in another single operation. It also split some of the flags from the Seventh Edition interface into multiple separate flags, and added some additional capabilities, although it did not support job control or the cooked-mode enhancements of 4BSD.[32] For example, it replaced the "cooked", "cbreak", and "raw" modes of Seventh Edition with different abstractions. The recognition of signal-generating characters is independent of input mode, and there are only the two input modes: canonical and non-canonical. (This allows a terminal input mode not present in Seventh Edition and BSD: canonical mode with signal generation disabled.)
System III's successors, includingSystem V, used the same interface.
One of the major problems that the POSIX standard addressed with its definition of a general terminal interface was the plethora of programmatic interfaces. Although by the time of the standard the behaviour of terminals was fairly uniform from system to system, most Unixes having adopted the notions of line disciplines and the BSD job control capabilities, the programmatic interface to terminals via theioctl() system call was a mess. Different Unixes supplied differentioctl() operations, with different (symbolic) names, and different flags. Portable source code had to contain a significant amount of conditional compilation to accommodate the differences across software platforms, even though they were all notionally Unix.[33]
The POSIX standard replaces theioctl() system entirely, with a set of library functions (which, of course, may beimplemented under the covers via platform-specificioctl() operations) with standardized names and parameters. Thetermio data structure of System V Unix was used as a template for the POSIXtermios data structure, whose fields were largely unchanged except that they now used alias data types for specifying the fields, allowing them to be easily ported across multiple processor architectures by implementors, rather than explicitly requiring theunsigned short andchar data types of the C and C++ programming languages (which might be inconvenient sizes on some processor architectures).[33][34]
POSIX also introduced support for job control, with thetermios structure containing suspend and delayed-suspend characters in addition to the control characters supported by System III and System V. It did not add any of the cooked-mode extensions from BSD, althoughSunOS 4.x,System V Release 4 systems includingSolaris,HP-UX,AIX, newer BSDs,macOS, andLinux have implemented them as extensions totermios.
Eachprocess in the system has either a singlecontrolling terminal, or no controlling terminal at all. A process inherits its controlling terminal from its parent, and the only operations upon a process are acquiring a controlling terminal, by a process that has no controlling terminal, and relinquishing it, by a process that has a controlling terminal.[33]
No portable way of acquiring a controlling terminal is defined, the method being implementation defined. The standard defines theO_NOCTTY flag for theopen() system call, which is the way ofpreventing what is otherwise the conventional way of acquiring a controlling terminal (a process with no controlling terminalopen()s a terminal device file that isn't already the controlling terminal for some other process, without specifying theO_NOCTTY flag[35]) but leaves its conventional semantics optional.
Each process also is a member of a process group. Each terminal device records a process group that is termed itsforeground process group. The process groups control terminal access and signal delivery. Signals generated at the terminal are sent to all processes that are members of the terminal's foreground process group.read() andwrite() I/O operations on a terminal by a process that is not a member of the terminal's foreground process group will and may optionally (respectively) cause signals (SIGTTIN andSIGTTOU respectively) to be sent to the invoking process. Various terminal-mode-altering library functions have the same behaviour aswrite(), except that they always generate the signals, even if that functionality is turned off forwrite() itself.[36][37]
termios data structureThe data structure used by all of the terminal library calls is thetermios structure,[38] whose C and C++ programming language definition is as follows:[34]
structtermios{tcflag_tc_iflag;// Input modestcflag_tc_oflag;// Output modestcflag_tc_cflag;// Control modestcflag_tc_lflag;// Local modescc_tc_cc[NCCS];// Control characters};
The order of the fields within thetermios structure is not defined, and implementations are allowed to add non-standard fields.[34] Indeed, implementations have to add non-standard fields for recording input and output baud rates. These are recorded in the structure, in an implementation-defined form, and accessed via accessor functions, rather than by direct manipulation of the field values, as is the case for the standardized structure fields.[39]
The data type aliasestcflag_t andcc_t, as well as the symbolic constantNCCS and symbolic constants for the various mode flags, control character names, and baud rates, are all defined in a standard headertermios.h. (This is not to be confused with the similarly named headertermio.h from System III and System V, which defines a similartermio structure and a lot of similarly named symbolic constants. This interface is specific to System III and System V, and code that uses it will not necessarily be portable to other systems.)[40]
The structure's fields are (in summary, for details see the main article[clarification needed]):
c_iflagc_oflagc_cflagc_lflagSIGTTOU signal by thewrite() system call[39]The library functions are (in summary, for details see the main article[clarification needed]):
tcgetattr()termios structure[43]tcsetattr()termios structure, optionally waiting for queued output to drain and flushing queued input[43]cfgetispeed()termios structure[44]cfgetospeed()termios structure[44]cfsetispeed()termios structure[44]cfsetospeed()termios structure[44]tcsendbreak()tcdrain()tcflush()tcflow()tcgetpgrp()tcsetpgrp()| Field | meaning | Retrieved byread() | Notes |
|---|---|---|---|
c_cc[VEOF] | end of file | No | Only processed by canonical mode line editing |
c_cc[VEOL] | end of line | Yes | Only processed by canonical mode line editing |
c_cc[VERASE] | "erase" | No | Only processed by canonical mode line editing |
c_cc[VKILL] | "kill" | No | Only processed by canonical mode line editing |
c_cc[VINTR] | "interrupt" | No | Signal-generation character independent of input mode |
c_cc[VQUIT] | "quit" | No | Signal-generation character independent of input mode |
c_cc[VSUSP] | "suspend" | No | Signal-generation character independent of input mode |
c_cc[VSTOP] | "stop" | No | Modem flow control character independent of input mode |
c_cc[VSTART] | "start" | No | Modem flow control character independent of input mode |
Thec_cc[] array member of thetermios data structure specifies all of the (programmatically modifiable) special characters. The indexes into the array are symbolic constants, one for each special character type, as in the table at right. (Two further entries in the array are relevant to non-canonical mode input processing and are discussed below.)[43]
Non-programmatically modifiable special characters arelinefeed (ASCIILF) andcarriage return (ASCIICR).[47]
Input processing determines the behaviour of theread() system call on a terminal device and the line editing and signal-generation characteristics of the line discipline. Unlike the case of Seventh Edition Unix and BSD version 4, and like the case of System III and System V, line editing operates in one of just two modes: canonical mode and non-canonical mode. The basic difference between them is when, from the point of view of the blocking/non-blocking requirements of theread() system call (specified with theO_NONBLOCK flag on thefile descriptor viaopen() orfcntl()), data "are available for reading".[48]
In canonical mode, data are accumulated in a line editing buffer, and do not become "available for reading" until line editing has been terminated by the user (at the terminal) sending aline delimiter character. Line delimiter characters are special characters, and they areend of file,end of line, andlinefeed (ASCIILF). The former two are settable programmatically, whilst the latter is fixed. The latter two are included in the line editing buffer, whilst the former one is not.[49]
More strictly, zero or more lines are accumulated in the line editing buffer, separated by line delimiters (which may or may not be discarded onceread() comes around to reading them), and line editing operates upon the part of the line editing buffer that follows the last (if any) line delimiter in the buffer. So, for example, the "erase" character (whatever that has been programmed to be) will erase the last character in the line buffer only up to (but not including) a preceding line delimiter.[49]
In non-canonical mode, data are accumulated in a buffer (which may or may not be the line editing buffer — some implementations having separate "processed input" and "raw input" queues) and become "available for reading" according to the values of two input control parameters, thec_cc[MIN] andc_cc[TIME] members of thetermios data structure. Both are unsigned quantities (becausecc_t is required to be an alias for an unsigned type). The former specifies a minimum number of characters, and the latter specifies a timeout in tenths of a second.[50] There are four possibilities:
c_cc[TIME] andc_cc[MIN] are both zeroread() returns immediately with whatever data are in the buffer (potentially returning zero if there are zero data available).[51]c_cc[TIME] is non-zero andc_cc[MIN] is zeroread() system call, or if a single character is received. In other words,read() waits for a maximum specified total time, and may return zero data, and returns any data as soon as they are received.[51]c_cc[TIME] is zero andc_cc[MIN] is non-zeroread() waits for a minimum amount of data (which may be larger than what the caller is prepared to read in the system call), will not return zero data, and may wait indefinitely.[51]c_cc[TIME] andc_cc[MIN] are both non-zeroread() waits for a minimum amount of data (which may be larger than what the caller is prepared to read in the system call), will not return zero data, may wait indefinitely, but won't wait longer than the specified timeout if at least one character is in the buffer to be read.[51]Output processing is largely unchanged from its System III/System V roots. Output mode control flags determine various options: