terminfo − terminal capability data base




Terminfo is a data base describing terminals, used by screen-oriented programs such as nvi(1), rogue(1) and libraries such as ncurses(3NCURSES). Terminfo describes terminals by giving a set of capabilities which they have, by specifying how to perform screen operations, and by specifying padding requirements and initialization sequences. This describes ncurses version 6.0 (patch 20161126).

Entries in terminfo consist of a sequence of ’,’ separated fields (embedded commas may be escaped with a backslash or notated as \054). White space after the ’,’ separator is ignored. The first entry for each terminal gives the names which are known for the terminal, separated by ’|’ characters. The first name given is the most common abbreviation for the terminal, the last name given should be a long name fully identifying the terminal, and all others are understood as synonyms for the terminal name. All names but the last should be in lower case and contain no blanks; the last name may well contain upper case and blanks for readability.

Lines beginning with a ’#’ in the first column are treated as comments. While comment lines are legal at any point, the output of captoinfo and infotocap (aliases for tic) will move comments so they occur only between entries.

Newlines and leading tabs may be used for formatting entries for readability. These are removed from parsed entries. The infocmp −f option relies on this to format if-then-else expressions: the result can be read by tic.

Terminal names (except for the last, verbose entry) should be chosen using the following conventions. The particular piece of hardware making up the terminal should have a root name, thus “hp2621”. This name should not contain hyphens. Modes that the hardware can be in, or user preferences, should be indicated by appending a hyphen and a mode suffix. Thus, a vt100 in 132 column mode would be vt100−w. The following suffixes should be used where possible:

For more on terminal naming conventions, see the term(7) manual page.

Predefined Capabilities
The following is a complete table of the capabilities included in a terminfo description block and available to terminfo-using code. In each line of the table,

The variable is the name by which the programmer (at the terminfo level) accesses the capability.

The capname is the short name used in the text of the database, and is used by a person updating the database. Whenever possible, capnames are chosen to be the same as or similar to the ANSI X3.64-1979 standard (now superseded by ECMA-48, which uses identical or very similar names). Semantics are also intended to match those of the specification.

The termcap code is the old termcap capability name (some capabilities are new, and have names which termcap did not originate).

Capability names have no hard length limit, but an informal limit of 5 characters has been adopted to keep them short and to allow the tabs in the source file Caps to line up nicely.

Finally, the description field attempts to convey the semantics of the capability. You may find some codes in the description field:


indicates that padding may be specified


in the description field indicates that the string is passed through tparm with parms as given (#i).


indicates that padding may vary in proportion to the number of lines affected


indicates the ith parameter.

These are the boolean capabilities:

These are the numeric capabilities:

The following numeric capabilities are present in the SVr4.0 term structure, but are not yet documented in the man page. They came in with SVr4’s printer support.

These are the string capabilities:

The following string capabilities are present in the SVr4.0 term structure, but were originally not documented in the man page.

The XSI Curses standard added these hardcopy capabilities. They were used in some post-4.1 versions of System V curses, e.g., Solaris 2.5 and IRIX 6.x. Except for YI, the ncurses termcap names for them are invented. According to the XSI Curses standard, they have no termcap names. If your compiled terminfo entries use these, they may not be binary-compatible with System V terminfo entries after SVr4.1; beware!

User-Defined Capabilities
The preceding section listed the predefined capabilities. They deal with some special features for terminals no longer (or possibly never) produced. Occasionally there are special features of newer terminals which are awkward or impossible to represent by reusing the predefined capabilities.

ncurses addresses this limitation by allowing user-defined capabilities. The tic and infocmp programs provide the −x option for this purpose. When −x is set, tic treats unknown capabilities as user-defined. That is, if tic encounters a capability name which it does not recognize, it infers its type (boolean, number or string) from the syntax and makes an extended table entry for that capability. The use_extended_names function makes this information conditionally available to applications. The ncurses library provides the data leaving most of the behavior to applications:

User-defined capability strings whose name begins with “k” are treated as function keys.

The types (boolean, number, string) determined by tic can be inferred by successful calls on tigetflag, etc.

If the capability name happens to be two characters, the capability is also available through the termcap interface.

While termcap is said to be extensible because it does not use a predefined set of capabilities, in practice it has been limited to the capabilities defined by terminfo implementations. As a rule, user-defined capabilities intended for use by termcap applications should be limited to booleans and numbers to avoid running past the 1023 byte limit assumed by termcap implementations and their applications. In particular, providing extended sets of function keys (past the 60 numbered keys and the handful of special named keys) is best done using the longer names available using terminfo.

A Sample Entry
The following entry, describing an ANSI-standard terminal, is representative of what a terminfo entry for a modern terminal typically looks like.

ansi|ansi/pc-term compatible with color,
        am, mc5i, mir, msgr,
        colors#8, cols#80, it#8, lines#24, ncv#3, pairs#64,
        bel=^G, blink=\E[5m, bold=\E[1m, cbt=\E[Z, clear=\E[H\E[J,
        cr=^M, cub=\E[%p1%dD, cub1=\E[D, cud=\E[%p1%dB, cud1=\E[B,
        cuf=\E[%p1%dC, cuf1=\E[C, cup=\E[%i%p1%d;%p2%dH,
        cuu=\E[%p1%dA, cuu1=\E[A, dch=\E[%p1%dP, dch1=\E[P,
        dl=\E[%p1%dM, dl1=\E[M, ech=\E[%p1%dX, ed=\E[J, el=\E[K,
        el1=\E[1K, home=\E[H, hpa=\E[%i%p1%dG, ht=\E[I, hts=\EH,
        ich=\E[%p1%d@, il=\E[%p1%dL, il1=\E[L, ind=^J,
        indn=\E[%p1%dS, invis=\E[8m, kbs=^H, kcbt=\E[Z, kcub1=\E[D,
        kcud1=\E[B, kcuf1=\E[C, kcuu1=\E[A, khome=\E[H, kich1=\E[L,
        mc4=\E[4i, mc5=\E[5i, nel=\r\E[S, op=\E[39;49m,
        rep=%p1%c\E[%p2%{1}%-%db, rev=\E[7m, rin=\E[%p1%dT,
        rmacs=\E[10m, rmpch=\E[10m, rmso=\E[m, rmul=\E[m,
        s0ds=\E(B, s1ds=\E)B, s2ds=\E*B, s3ds=\E+B,
        setab=\E[4%p1%dm, setaf=\E[3%p1%dm,
        sgr0=\E[0;10m, smacs=\E[11m, smpch=\E[11m, smso=\E[7m,
        smul=\E[4m, tbc=\E[3g, u6=\E[%i%d;%dR, u7=\E[6n,
        u8=\E[?%[;0123456789]c, u9=\E[c, vpa=\E[%i%p1%dd,

Entries may continue onto multiple lines by placing white space at the beginning of each line except the first. Comments may be included on lines beginning with “#”. Capabilities in terminfo are of three types:

Boolean capabilities which indicate that the terminal has some particular feature,

numeric capabilities giving the size of the terminal or the size of particular delays, and

string capabilities, which give a sequence which can be used to perform particular terminal operations.

Types of Capabilities
All capabilities have names. For instance, the fact that ANSI-standard terminals have automatic margins (i.e., an automatic return and line-feed when the end of a line is reached) is indicated by the capability am. Hence the description of ansi includes am. Numeric capabilities are followed by the character “#” and then a positive value. Thus cols, which indicates the number of columns the terminal has, gives the value “80” for ansi. Values for numeric capabilities may be specified in decimal, octal or hexadecimal, using the C programming language conventions (e.g., 255, 0377 and 0xff or 0xFF).

Finally, string valued capabilities, such as el (clear to end of line sequence) are given by the two-character code, an “=”, and then a string ending at the next following “,”.

A number of escape sequences are provided in the string valued capabilities for easy encoding of characters there. Both \E and \e map to an ESCAPE character, ^x maps to a control-x for any appropriate x, and the sequences \n \l \r \t \b \f \s give a newline, line-feed, return, tab, backspace, form-feed, and space. Other escapes include

\^ for ^,

\\ for \,

\, for comma,

\: for :,

and \0 for null.

\0 will produce \200, which does not terminate a string but behaves as a null character on most terminals, providing CS7 is specified. See stty(1).

The reason for this quirk is to maintain binary compatibility of the compiled terminfo files with other implementations, e.g., the SVr4 systems, which document this. Compiled terminfo files use null-terminated strings, with no lengths. Modifying this would require a new binary format, which would not work with other implementations.

Finally, characters may be given as three octal digits after a \.

A delay in milliseconds may appear anywhere in a string capability, enclosed in $<..> brackets, as in el=\EK$<5>, and padding characters are supplied by tputs to provide this delay.

The delay must be a number with at most one decimal place of precision; it may be followed by suffixes “*” or “/” or both.

A “*” indicates that the padding required is proportional to the number of lines affected by the operation, and the amount given is the per-affected-unit padding required. (In the case of insert character, the factor is still the number of lines affected.)

Normally, padding is advisory if the device has the xon capability; it is used for cost computation but does not trigger delays.

A “/” suffix indicates that the padding is mandatory and forces a delay of the given number of milliseconds even on devices for which xon is present to indicate flow control.

Sometimes individual capabilities must be commented out. To do this, put a period before the capability name. For example, see the second ind in the example above.

Fetching Compiled Descriptions
The ncurses library searches for terminal descriptions in several places. It uses only the first description found. The library has a compiled-in list of places to search which can be overridden by environment variables. Before starting to search, ncurses eliminates duplicates in its search list.

If the environment variable TERMINFO is set, it is interpreted as the pathname of a directory containing the compiled description you are working on. Only that directory is searched.

If TERMINFO is not set, ncurses will instead look in the directory $HOME/.terminfo for a compiled description.

Next, if the environment variable TERMINFO_DIRS is set, ncurses will interpret the contents of that variable as a list of colon-separated directories (or database files) to be searched.

An empty directory name (i.e., if the variable begins or ends with a colon, or contains adjacent colons) is interpreted as the system location /etc/terminfo.

Finally, ncurses searches these compiled-in locations:

a list of directories (no default value), and

the system terminfo directory, /etc/terminfo (the compiled-in default).

Preparing Descriptions
We now outline how to prepare descriptions of terminals. The most effective way to prepare a terminal description is by imitating the description of a similar terminal in terminfo and to build up a description gradually, using partial descriptions with vi or some other screen-oriented program to check that they are correct. Be aware that a very unusual terminal may expose deficiencies in the ability of the terminfo file to describe it or bugs in the screen-handling code of the test program.

To get the padding for insert line right (if the terminal manufacturer did not document it) a severe test is to edit a large file at 9600 baud, delete 16 or so lines from the middle of the screen, then hit the “u” key several times quickly. If the terminal messes up, more padding is usually needed. A similar test can be used for insert character.

Basic Capabilities
The number of columns on each line for the terminal is given by the cols numeric capability. If the terminal is a CRT , then the number of lines on the screen is given by the lines capability. If the terminal wraps around to the beginning of the next line when it reaches the right margin, then it should have the am capability. If the terminal can clear its screen, leaving the cursor in the home position, then this is given by the clear string capability. If the terminal overstrikes (rather than clearing a position when a character is struck over) then it should have the os capability. If the terminal is a printing terminal, with no soft copy unit, give it both hc and os. (os applies to storage scope terminals, such as TEKTRONIX 4010 series, as well as hard copy and APL terminals.) If there is a code to move the cursor to the left edge of the current row, give this as cr. (Normally this will be carriage return, control M.) If there is a code to produce an audible signal (bell, beep, etc) give this as bel.

If there is a code to move the cursor one position to the left (such as backspace) that capability should be given as cub1. Similarly, codes to move to the right, up, and down should be given as cuf1, cuu1, and cud1. These local cursor motions should not alter the text they pass over, for example, you would not normally use “cuf1= ” because the space would erase the character moved over.

A very important point here is that the local cursor motions encoded in terminfo are undefined at the left and top edges of a CRT terminal. Programs should never attempt to backspace around the left edge, unless bw is given, and never attempt to go up locally off the top. In order to scroll text up, a program will go to the bottom left corner of the screen and send the ind (index) string.

To scroll text down, a program goes to the top left corner of the screen and sends the ri (reverse index) string. The strings ind and ri are undefined when not on their respective corners of the screen.

Parameterized versions of the scrolling sequences are indn and rin which have the same semantics as ind and ri except that they take one parameter, and scroll that many lines. They are also undefined except at the appropriate edge of the screen.

The am capability tells whether the cursor sticks at the right edge of the screen when text is output, but this does not necessarily apply to a cuf1 from the last column. The only local motion which is defined from the left edge is if bw is given, then a cub1 from the left edge will move to the right edge of the previous row. If bw is not given, the effect is undefined. This is useful for drawing a box around the edge of the screen, for example. If the terminal has switch selectable automatic margins, the terminfo file usually assumes that this is on; i.e., am. If the terminal has a command which moves to the first column of the next line, that command can be given as nel (newline). It does not matter if the command clears the remainder of the current line, so if the terminal has no cr and lf it may still be possible to craft a working nel out of one or both of them.

These capabilities suffice to describe hard-copy and “glass-tty” terminals. Thus the model 33 teletype is described as

33|tty33|tty|model 33 teletype,
        bel=^G, cols#72, cr=^M, cud1=^J, hc, ind=^J, os,

while the Lear Siegler ADM-3 is described as

adm3|3|lsi adm3,
        am, bel=^G, clear=^Z, cols#80, cr=^M, cub1=^H, cud1=^J,
        ind=^J, lines#24,

Parameterized Strings
Cursor addressing and other strings requiring parameters in the terminal are described by a parameterized string capability, with printf-like escapes such as %x in it. For example, to address the cursor, the cup capability is given, using two parameters: the row and column to address to. (Rows and columns are numbered from zero and refer to the physical screen visible to the user, not to any unseen memory.) If the terminal has memory relative cursor addressing, that can be indicated by mrcup.

The parameter mechanism uses a stack and special % codes to manipulate it. Typically a sequence will push one of the parameters onto the stack and then print it in some format. Print (e.g., "%d") is a special case. Other operations, including "%t" pop their operand from the stack. It is noted that more complex operations are often necessary, e.g., in the sgr string.

The % encodings have the following meanings:


outputs “%”


as in printf, flags are [−+#] and space. Use a “:” to allow the next character to be a “−” flag, avoiding interpreting "%−" as an operator.


print pop() like %c in printf


print pop() like %s in printf


push i’th parameter


set dynamic variable [a−z] to pop()


get dynamic variable [a−z] and push it


set static variable [a−z] to pop()


get static variable [a−z] and push it

The terms "static" and "dynamic" are misleading. Historically, these are simply two different sets of variables, whose values are not reset between calls to tparm. However, that fact is not documented in other implementations. Relying on it will adversely impact portability to other implementations.


char constant c


integer constant nn


push strlen(pop)

%+, %−, %*, %/, %m

arithmetic (%m is mod): push(pop() op pop())

%&, %|, %^

bit operations (AND, OR and exclusive-OR): push(pop() op pop())

%=, %>, %<

logical operations: push(pop() op pop())

%A, %O

logical AND and OR operations (for conditionals)

%!, %~

unary operations (logical and bit complement): push(op pop())


add 1 to first two parameters (for ANSI terminals)

%? expr %t thenpart %e elsepart %;

This forms an if-then-else. The %e elsepart is optional. Usually the %? expr part pushes a value onto the stack, and %t pops it from the stack, testing if it is nonzero (true). If it is zero (false), control passes to the %e (else) part.

It is possible to form else-if’s a la Algol 68:
c1 %t b1 %e c2 %t b2 %e c3 %t b3 %e c4 %t b4 %e %;

where ci are conditions, bi are bodies.

Use the −f option of tic or infocmp to see the structure of if-then-else’s. Some strings, e.g., sgr can be very complicated when written on one line. The −f option splits the string into lines with the parts indented.

Binary operations are in postfix form with the operands in the usual order. That is, to get x−5 one would use "%gx%{5}%-". %P and %g variables are persistent across escape-string evaluations.

Consider the HP2645, which, to get to row 3 and column 12, needs to be sent \E&a12c03Y padded for 6 milliseconds. Note that the order of the rows and columns is inverted here, and that the row and column are printed as two digits. Thus its cup capability is “cup=6\E&%p2%2dc%p1%2dY”.

The Microterm ACT-IV needs the current row and column sent preceded by a ^T, with the row and column simply encoded in binary, “cup=^T%p1%c%p2%c”. Terminals which use “%c” need to be able to backspace the cursor (cub1), and to move the cursor up one line on the screen (cuu1). This is necessary because it is not always safe to transmit \n ^D and \r, as the system may change or discard them. (The library routines dealing with terminfo set tty modes so that tabs are never expanded, so \t is safe to send. This turns out to be essential for the Ann Arbor 4080.)

A final example is the LSI ADM -3a, which uses row and column offset by a blank character, thus “cup=\E=%p1%’ ’%+%c%p2%’ ’%+%c”. After sending “\E=”, this pushes the first parameter, pushes the ASCII value for a space (32), adds them (pushing the sum on the stack in place of the two previous values) and outputs that value as a character. Then the same is done for the second parameter. More complex arithmetic is possible using the stack.

Cursor Motions
If the terminal has a fast way to home the cursor (to very upper left corner of screen) then this can be given as home; similarly a fast way of getting to the lower left-hand corner can be given as ll; this may involve going up with cuu1 from the home position, but a program should never do this itself (unless ll does) because it can make no assumption about the effect of moving up from the home position. Note that the home position is the same as addressing to (0,0): to the top left corner of the screen, not of memory. (Thus, the \EH sequence on HP terminals cannot be used for home.)

If the terminal has row or column absolute cursor addressing, these can be given as single parameter capabilities hpa (horizontal position absolute) and vpa (vertical position absolute). Sometimes these are shorter than the more general two parameter sequence (as with the hp2645) and can be used in preference to cup. If there are parameterized local motions (e.g., move n spaces to the right) these can be given as cud, cub, cuf, and cuu with a single parameter indicating how many spaces to move. These are primarily useful if the terminal does not have cup, such as the TEKTRONIX 4025.

If the terminal needs to be in a special mode when running a program that uses these capabilities, the codes to enter and exit this mode can be given as smcup and rmcup. This arises, for example, from terminals like the Concept with more than one page of memory. If the terminal has only memory relative cursor addressing and not screen relative cursor addressing, a one screen-sized window must be fixed into the terminal for cursor addressing to work properly. This is also used for the TEKTRONIX 4025, where smcup sets the command character to be the one used by terminfo. If the smcup sequence will not restore the screen after an rmcup sequence is output (to the state prior to outputting rmcup), specify nrrmc.

Area Clears
If the terminal can clear from the current position to the end of the line, leaving the cursor where it is, this should be given as el. If the terminal can clear from the beginning of the line to the current position inclusive, leaving the cursor where it is, this should be given as el1. If the terminal can clear from the current position to the end of the display, then this should be given as ed. Ed is only defined from the first column of a line. (Thus, it can be simulated by a request to delete a large number of lines, if a true ed is not available.)

Insert/delete line and vertical motions
If the terminal can open a new blank line before the line where the cursor is, this should be given as il1; this is done only from the first position of a line. The cursor must then appear on the newly blank line. If the terminal can delete the line which the cursor is on, then this should be given as dl1; this is done only from the first position on the line to be deleted. Versions of il1 and dl1 which take a single parameter and insert or delete that many lines can be given as il and dl.

If the terminal has a settable scrolling region (like the vt100) the command to set this can be described with the csr capability, which takes two parameters: the top and bottom lines of the scrolling region. The cursor position is, alas, undefined after using this command.

It is possible to get the effect of insert or delete line using csr on a properly chosen region; the sc and rc (save and restore cursor) commands may be useful for ensuring that your synthesized insert/delete string does not move the cursor. (Note that the ncurses(3NCURSES) library does this synthesis automatically, so you need not compose insert/delete strings for an entry with csr).

Yet another way to construct insert and delete might be to use a combination of index with the memory-lock feature found on some terminals (like the HP−700/90 series, which however also has insert/delete).

Inserting lines at the top or bottom of the screen can also be done using ri or ind on many terminals without a true insert/delete line, and is often faster even on terminals with those features.

The boolean non_dest_scroll_region should be set if each scrolling window is effectively a view port on a screen-sized canvas. To test for this capability, create a scrolling region in the middle of the screen, write something to the bottom line, move the cursor to the top of the region, and do ri followed by dl1 or ind. If the data scrolled off the bottom of the region by the ri re-appears, then scrolling is non-destructive. System V and XSI Curses expect that ind, ri, indn, and rin will simulate destructive scrolling; their documentation cautions you not to define csr unless this is true. This curses implementation is more liberal and will do explicit erases after scrolling if ndsrc is defined.

If the terminal has the ability to define a window as part of memory, which all commands affect, it should be given as the parameterized string wind. The four parameters are the starting and ending lines in memory and the starting and ending columns in memory, in that order.

If the terminal can retain display memory above, then the da capability should be given; if display memory can be retained below, then db should be given. These indicate that deleting a line or scrolling may bring non-blank lines up from below or that scrolling back with ri may bring down non-blank lines.

Insert/Delete Character
There are two basic kinds of intelligent terminals with respect to insert/delete character which can be described using terminfo. The most common insert/delete character operations affect only the characters on the current line and shift characters off the end of the line rigidly. Other terminals, such as the Concept 100 and the Perkin Elmer Owl, make a distinction between typed and untyped blanks on the screen, shifting upon an insert or delete only to an untyped blank on the screen which is either eliminated, or expanded to two untyped blanks.

You can determine the kind of terminal you have by clearing the screen and then typing text separated by cursor motions. Type “abc    def” using local cursor motions (not spaces) between the “abc” and the “def”. Then position the cursor before the “abc” and put the terminal in insert mode. If typing characters causes the rest of the line to shift rigidly and characters to fall off the end, then your terminal does not distinguish between blanks and untyped positions. If the “abc” shifts over to the “def” which then move together around the end of the current line and onto the next as you insert, you have the second type of terminal, and should give the capability in, which stands for “insert null”.

While these are two logically separate attributes (one line versus multi-line insert mode, and special treatment of untyped spaces) we have seen no terminals whose insert mode cannot be described with the single attribute.

Terminfo can describe both terminals which have an insert mode, and terminals which send a simple sequence to open a blank position on the current line. Give as smir the sequence to get into insert mode. Give as rmir the sequence to leave insert mode. Now give as ich1 any sequence needed to be sent just before sending the character to be inserted. Most terminals with a true insert mode will not give ich1; terminals which send a sequence to open a screen position should give it here.

If your terminal has both, insert mode is usually preferable to ich1. Technically, you should not give both unless the terminal actually requires both to be used in combination. Accordingly, some non-curses applications get confused if both are present; the symptom is doubled characters in an update using insert. This requirement is now rare; most ich sequences do not require previous smir, and most smir insert modes do not require ich1 before each character. Therefore, the new curses actually assumes this is the case and uses either rmir/smir or ich/ich1 as appropriate (but not both). If you have to write an entry to be used under new curses for a terminal old enough to need both, include the rmir/smir sequences in ich1.

If post insert padding is needed, give this as a number of milliseconds in ip (a string option). Any other sequence which may need to be sent after an insert of a single character may also be given in ip. If your terminal needs both to be placed into an “insert mode” and a special code to precede each inserted character, then both smir/rmir and ich1 can be given, and both will be used. The ich capability, with one parameter, n, will repeat the effects of ich1 n times.

If padding is necessary between characters typed while not in insert mode, give this as a number of milliseconds padding in rmp.

It is occasionally necessary to move around while in insert mode to delete characters on the same line (e.g., if there is a tab after the insertion position). If your terminal allows motion while in insert mode you can give the capability mir to speed up inserting in this case. Omitting mir will affect only speed. Some terminals (notably Datamedia’s) must not have mir because of the way their insert mode works.

Finally, you can specify dch1 to delete a single character, dch with one parameter, n, to delete n characters, and delete mode by giving smdc and rmdc to enter and exit delete mode (any mode the terminal needs to be placed in for dch1 to work).

A command to erase n characters (equivalent to outputting n blanks without moving the cursor) can be given as ech with one parameter.

Highlighting, Underlining, and Visible Bells
If your terminal has one or more kinds of display attributes, these can be represented in a number of different ways. You should choose one display form as standout mode, representing a good, high contrast, easy-on-the-eyes, format for highlighting error messages and other attention getters. (If you have a choice, reverse video plus half-bright is good, or reverse video alone.) The sequences to enter and exit standout mode are given as smso and rmso, respectively. If the code to change into or out of standout mode leaves one or even two blank spaces on the screen, as the TVI 912 and Teleray 1061 do, then xmc should be given to tell how many spaces are left.

Codes to begin underlining and end underlining can be given as smul and rmul respectively. If the terminal has a code to underline the current character and move the cursor one space to the right, such as the Microterm Mime, this can be given as uc.

Other capabilities to enter various highlighting modes include blink (blinking) bold (bold or extra bright) dim (dim or half-bright) invis (blanking or invisible text) prot (protected) rev (reverse video) sgr0 (turn off all attribute modes) smacs (enter alternate character set mode) and rmacs (exit alternate character set mode). Turning on any of these modes singly may or may not turn off other modes.

If there is a sequence to set arbitrary combinations of modes, this should be given as sgr (set attributes), taking 9 parameters. Each parameter is either 0 or nonzero, as the corresponding attribute is on or off. The 9 parameters are, in order: standout, underline, reverse, blink, dim, bold, blank, protect, alternate character set. Not all modes need be supported by sgr, only those for which corresponding separate attribute commands exist.

For example, the DEC vt220 supports most of the modes:

We begin each escape sequence by turning off any existing modes, since there is no quick way to determine whether they are active. Standout is set up to be the combination of reverse and bold. The vt220 terminal has a protect mode, though it is not commonly used in sgr because it protects characters on the screen from the host’s erasures. The altcharset mode also is different in that it is either ^O or ^N, depending on whether it is off or on. If all modes are turned on, the resulting sequence is \E[0;1;4;5;7;8m^N.

Some sequences are common to different modes. For example, ;7 is output when either p1 or p3 is true, that is, if either standout or reverse modes are turned on.

Writing out the above sequences, along with their dependencies yields

Putting this all together into the sgr sequence gives:


Remember that if you specify sgr, you must also specify sgr0. Also, some implementations rely on sgr being given if sgr0 is, Not all terminfo entries necessarily have an sgr string, however. Many terminfo entries are derived from termcap entries which have no sgr string. The only drawback to adding an sgr string is that termcap also assumes that sgr0 does not exit alternate character set mode.

Terminals with the “magic cookie” glitch (xmc) deposit special “cookies” when they receive mode-setting sequences, which affect the display algorithm rather than having extra bits for each character. Some terminals, such as the HP 2621, automatically leave standout mode when they move to a new line or the cursor is addressed. Programs using standout mode should exit standout mode before moving the cursor or sending a newline, unless the msgr capability, asserting that it is safe to move in standout mode, is present.

If the terminal has a way of flashing the screen to indicate an error quietly (a bell replacement) then this can be given as flash; it must not move the cursor.

If the cursor needs to be made more visible than normal when it is not on the bottom line (to make, for example, a non-blinking underline into an easier to find block or blinking underline) give this sequence as cvvis. If there is a way to make the cursor completely invisible, give that as civis. The capability cnorm should be given which undoes the effects of both of these modes.

If your terminal correctly generates underlined characters (with no special codes needed) even though it does not overstrike, then you should give the capability ul. If a character overstriking another leaves both characters on the screen, specify the capability os. If overstrikes are erasable with a blank, then this should be indicated by giving eo.

Keypad and Function Keys
If the terminal has a keypad that transmits codes when the keys are pressed, this information can be given. Note that it is not possible to handle terminals where the keypad only works in local (this applies, for example, to the unshifted HP 2621 keys). If the keypad can be set to transmit or not transmit, give these codes as smkx and rmkx. Otherwise the keypad is assumed to always transmit.

The codes sent by the left arrow, right arrow, up arrow, down arrow, and home keys can be given as kcub1, kcuf1, kcuu1, kcud1, and khome respectively. If there are function keys such as f0, f1, ..., f10, the codes they send can be given as kf0, kf1, ..., kf10. If these keys have labels other than the default f0 through f10, the labels can be given as lf0, lf1, ..., lf10.

The codes transmitted by certain other special keys can be given:

kll (home down),

kbs (backspace),

ktbc (clear all tabs),

kctab (clear the tab stop in this column),

kclr (clear screen or erase key),

kdch1 (delete character),

kdl1 (delete line),

krmir (exit insert mode),

kel (clear to end of line),

ked (clear to end of screen),

kich1 (insert character or enter insert mode),

kil1 (insert line),

knp (next page),

kpp (previous page),

kind (scroll forward/down),

kri (scroll backward/up),

khts (set a tab stop in this column).

In addition, if the keypad has a 3 by 3 array of keys including the four arrow keys, the other five keys can be given as ka1, ka3, kb2, kc1, and kc3. These keys are useful when the effects of a 3 by 3 directional pad are needed.

Strings to program function keys can be given as pfkey, pfloc, and pfx. A string to program screen labels should be specified as pln. Each of these strings takes two parameters: the function key number to program (from 0 to 10) and the string to program it with. Function key numbers out of this range may program undefined keys in a terminal dependent manner. The difference between the capabilities is that pfkey causes pressing the given key to be the same as the user typing the given string; pfloc causes the string to be executed by the terminal in local; and pfx causes the string to be transmitted to the computer.

The capabilities nlab, lw and lh define the number of programmable screen labels and their width and height. If there are commands to turn the labels on and off, give them in smln and rmln. smln is normally output after one or more pln sequences to make sure that the change becomes visible.

Tabs and Initialization
If the terminal has hardware tabs, the command to advance to the next tab stop can be given as ht (usually control I). A “back-tab” command which moves leftward to the preceding tab stop can be given as cbt. By convention, if the teletype modes indicate that tabs are being expanded by the computer rather than being sent to the terminal, programs should not use ht or cbt even if they are present, since the user may not have the tab stops properly set. If the terminal has hardware tabs which are initially set every n spaces when the terminal is powered up, the numeric parameter it is given, showing the number of spaces the tabs are set to. This is normally used by the tset command to determine whether to set the mode for hardware tab expansion, and whether to set the tab stops. If the terminal has tab stops that can be saved in non-volatile memory, the terminfo description can assume that they are properly set.

Other capabilities include is1, is2, and is3, initialization strings for the terminal, iprog, the path name of a program to be run to initialize the terminal, and if, the name of a file containing long initialization strings. These strings are expected to set the terminal into modes consistent with the rest of the terminfo description. They are normally sent to the terminal, by the init option of the tput program, each time the user logs in. They will be printed in the following order:

run the program



is1 is2

set the margins using

mgc, smgl and smgr

set tabs using

tbc and hts

print the file


and finally

output is3.

Most initialization is done with is2. Special terminal modes can be set up without duplicating strings by putting the common sequences in is2 and special cases in is1 and is3.

A set of sequences that does a harder reset from a totally unknown state can be given as rs1, rs2, rf and rs3, analogous to is1 , is2 , if and is3 respectively. These strings are output by the reset program, which is used when the terminal gets into a wedged state. Commands are normally placed in rs1, rs2 rs3 and rf only if they produce annoying effects on the screen and are not necessary when logging in. For example, the command to set the vt100 into 80-column mode would normally be part of is2, but it causes an annoying glitch of the screen and is not normally needed since the terminal is usually already in 80 column mode.

The reset program writes strings including iprog, etc., in the same order as the init program, using rs1, etc., instead of is1, etc. If any of rs1, rs2, rs3, or rf reset capability strings are missing, the reset program falls back upon the corresponding initialization capability string.

If there are commands to set and clear tab stops, they can be given as tbc (clear all tab stops) and hts (set a tab stop in the current column of every row). If a more complex sequence is needed to set the tabs than can be described by this, the sequence can be placed in is2 or if.

Delays and Padding
Many older and slower terminals do not support either XON/XOFF or DTR handshaking, including hard copy terminals and some very archaic CRTs (including, for example, DEC VT100s). These may require padding characters after certain cursor motions and screen changes.

If the terminal uses xon/xoff handshaking for flow control (that is, it automatically emits ^S back to the host when its input buffers are close to full), set xon. This capability suppresses the emission of padding. You can also set it for memory-mapped console devices effectively that do not have a speed limit. Padding information should still be included so that routines can make better decisions about relative costs, but actual pad characters will not be transmitted.

If pb (padding baud rate) is given, padding is suppressed at baud rates below the value of pb. If the entry has no padding baud rate, then whether padding is emitted or not is completely controlled by xon.

If the terminal requires other than a null (zero) character as a pad, then this can be given as pad. Only the first character of the pad string is used.

Status Lines
Some terminals have an extra “status line” which is not normally used by software (and thus not counted in the terminal’s lines capability).

The simplest case is a status line which is cursor-addressable but not part of the main scrolling region on the screen; the Heathkit H19 has a status line of this kind, as would a 24-line VT100 with a 23-line scrolling region set up on initialization. This situation is indicated by the hs capability.

Some terminals with status lines need special sequences to access the status line. These may be expressed as a string with single parameter tsl which takes the cursor to a given zero-origin column on the status line. The capability fsl must return to the main-screen cursor positions before the last tsl. You may need to embed the string values of sc (save cursor) and rc (restore cursor) in tsl and fsl to accomplish this.

The status line is normally assumed to be the same width as the width of the terminal. If this is untrue, you can specify it with the numeric capability wsl.

A command to erase or blank the status line may be specified as dsl.

The boolean capability eslok specifies that escape sequences, tabs, etc., work ordinarily in the status line.

The ncurses implementation does not yet use any of these capabilities. They are documented here in case they ever become important.

Line Graphics
Many terminals have alternate character sets useful for forms-drawing. Terminfo and curses build in support for the drawing characters supported by the VT100, with some characters from the AT&T 4410v1 added. This alternate character set may be specified by the acsc capability.

The best way to define a new device’s graphics set is to add a column to a copy of this table for your terminal, giving the character which (when emitted between smacs/rmacs switches) will be rendered as the corresponding graphic. Then read off the VT100/your terminal character pairs right to left in sequence; these become the ACSC string.

Color Handling
Most color terminals are either “Tektronix-like” or “HP-like”. Tektronix-like terminals have a predefined set of N colors (where N usually 8), and can set character-cell foreground and background characters independently, mixing them into N * N color-pairs. On HP-like terminals, the use must set each color pair up separately (foreground and background are not independently settable). Up to M color-pairs may be set up from 2*M different colors. ANSI-compatible terminals are Tektronix-like.

Some basic color capabilities are independent of the color method. The numeric capabilities colors and pairs specify the maximum numbers of colors and color-pairs that can be displayed simultaneously. The op (original pair) string resets foreground and background colors to their default values for the terminal. The oc string resets all colors or color-pairs to their default values for the terminal. Some terminals (including many PC terminal emulators) erase screen areas with the current background color rather than the power-up default background; these should have the boolean capability bce.

To change the current foreground or background color on a Tektronix-type terminal, use setaf (set ANSI foreground) and setab (set ANSI background) or setf (set foreground) and setb (set background). These take one parameter, the color number. The SVr4 documentation describes only setaf/setab; the XPG4 draft says that "If the terminal supports ANSI escape sequences to set background and foreground, they should be coded as setaf and setab, respectively. If the terminal supports other escape sequences to set background and foreground, they should be coded as setf and setb, respectively. The vidputs function and the refresh functions use setaf and setab if they are defined."

The setaf/setab and setf/setb capabilities take a single numeric argument each. Argument values 0-7 of setaf/setab are portably defined as follows (the middle column is the symbolic #define available in the header for the curses or ncurses libraries). The terminal hardware is free to map these as it likes, but the RGB values indicate normal locations in color space.


The argument values of setf/setb historically correspond to a different mapping, i.e.,


It is important to not confuse the two sets of color capabilities; otherwise red/blue will be interchanged on the display.

On an HP-like terminal, use scp with a color-pair number parameter to set which color pair is current.

On a Tektronix-like terminal, the capability ccc may be present to indicate that colors can be modified. If so, the initc capability will take a color number (0 to colors − 1)and three more parameters which describe the color. These three parameters default to being interpreted as RGB (Red, Green, Blue) values. If the boolean capability hls is present, they are instead as HLS (Hue, Lightness, Saturation) indices. The ranges are terminal-dependent.

On an HP-like terminal, initp may give a capability for changing a color-pair value. It will take seven parameters; a color-pair number (0 to max_pairs − 1), and two triples describing first background and then foreground colors. These parameters must be (Red, Green, Blue) or (Hue, Lightness, Saturation) depending on hls.

On some color terminals, colors collide with highlights. You can register these collisions with the ncv capability. This is a bit-mask of attributes not to be used when colors are enabled. The correspondence with the attributes understood by curses is as follows:

For example, on many IBM PC consoles, the underline attribute collides with the foreground color blue and is not available in color mode. These should have an ncv capability of 2.

SVr4 curses does nothing with ncv, ncurses recognizes it and optimizes the output in favor of colors.

If the terminal requires other than a null (zero) character as a pad, then this can be given as pad. Only the first character of the pad string is used. If the terminal does not have a pad character, specify npc. Note that ncurses implements the termcap-compatible PC variable; though the application may set this value to something other than a null, ncurses will test npc first and use napms if the terminal has no pad character.

If the terminal can move up or down half a line, this can be indicated with hu (half-line up) and hd (half-line down). This is primarily useful for superscripts and subscripts on hard-copy terminals. If a hard-copy terminal can eject to the next page (form feed), give this as ff (usually control L).

If there is a command to repeat a given character a given number of times (to save time transmitting a large number of identical characters) this can be indicated with the parameterized string rep. The first parameter is the character to be repeated and the second is the number of times to repeat it. Thus, tparm(repeat_char, ’x’, 10) is the same as “xxxxxxxxxx”.

If the terminal has a settable command character, such as the TEKTRONIX 4025, this can be indicated with cmdch. A prototype command character is chosen which is used in all capabilities. This character is given in the cmdch capability to identify it. The following convention is supported on some UNIX systems: The environment is to be searched for a CC variable, and if found, all occurrences of the prototype character are replaced with the character in the environment variable.

Terminal descriptions that do not represent a specific kind of known terminal, such as switch, dialup, patch, and network, should include the gn (generic) capability so that programs can complain that they do not know how to talk to the terminal. (This capability does not apply to virtual terminal descriptions for which the escape sequences are known.)

If the terminal has a “meta key” which acts as a shift key, setting the 8th bit of any character transmitted, this fact can be indicated with km. Otherwise, software will assume that the 8th bit is parity and it will usually be cleared. If strings exist to turn this “meta mode” on and off, they can be given as smm and rmm.

If the terminal has more lines of memory than will fit on the screen at once, the number of lines of memory can be indicated with lm. A value of lm#0 indicates that the number of lines is not fixed, but that there is still more memory than fits on the screen.

If the terminal is one of those supported by the UNIX virtual terminal protocol, the terminal number can be given as vt.

Media copy strings which control an auxiliary printer connected to the terminal can be given as mc0: print the contents of the screen, mc4: turn off the printer, and mc5: turn on the printer. When the printer is on, all text sent to the terminal will be sent to the printer. It is undefined whether the text is also displayed on the terminal screen when the printer is on. A variation mc5p takes one parameter, and leaves the printer on for as many characters as the value of the parameter, then turns the printer off. The parameter should not exceed 255. All text, including mc4, is transparently passed to the printer while an mc5p is in effect.

Glitches and Braindamage
Hazeltine terminals, which do not allow “~” characters to be displayed should indicate hz.

Terminals which ignore a line-feed immediately after an am wrap, such as the Concept and vt100, should indicate xenl.

If el is required to get rid of standout (instead of merely writing normal text on top of it), xhp should be given.

Teleray terminals, where tabs turn all characters moved over to blanks, should indicate xt (destructive tabs). Note: the variable indicating this is now “dest_tabs_magic_smso”; in older versions, it was teleray_glitch. This glitch is also taken to mean that it is not possible to position the cursor on top of a “magic cookie”, that to erase standout mode it is instead necessary to use delete and insert line. The ncurses implementation ignores this glitch.

The Beehive Superbee, which is unable to correctly transmit the escape or control C characters, has xsb, indicating that the f1 key is used for escape and f2 for control C. (Only certain Superbees have this problem, depending on the ROM.) Note that in older terminfo versions, this capability was called “beehive_glitch”; it is now “no_esc_ctl_c”.

Other specific terminal problems may be corrected by adding more capabilities of the form xx.

Similar Terminals
If there are two very similar terminals, one (the variant) can be defined as being just like the other (the base) with certain exceptions. In the definition of the variant, the string capability use can be given with the name of the base terminal. The capabilities given before use override those in the base type named by use. If there are multiple use capabilities, they are merged in reverse order. That is, the rightmost use reference is processed first, then the one to its left, and so forth. Capabilities given explicitly in the entry override those brought in by use references.

A capability can be canceled by placing xx@ to the left of the use reference that imports it, where xx is the capability. For example, the entry

2621−nl, smkx@, rmkx@, use=2621,

defines a 2621−nl that does not have the smkx or rmkx capabilities, and hence does not turn on the function key labels when in visual mode. This is useful for different modes for a terminal, or for different user preferences.

Pitfalls of Long Entries
Long terminfo entries are unlikely to be a problem; to date, no entry has even approached terminfo’s 4096-byte string-table maximum. Unfortunately, the termcap translations are much more strictly limited (to 1023 bytes), thus termcap translations of long terminfo entries can cause problems.

The man pages for 4.3BSD and older versions of tgetent instruct the user to allocate a 1024-byte buffer for the termcap entry. The entry gets null-terminated by the termcap library, so that makes the maximum safe length for a termcap entry 1k−1 (1023) bytes. Depending on what the application and the termcap library being used does, and where in the termcap file the terminal type that tgetent is searching for is, several bad things can happen.

Some termcap libraries print a warning message or exit if they find an entry that’s longer than 1023 bytes; others do not; others truncate the entries to 1023 bytes. Some application programs allocate more than the recommended 1K for the termcap entry; others do not.

Each termcap entry has two important sizes associated with it: before "tc" expansion, and after "tc" expansion. "tc" is the capability that tacks on another termcap entry to the end of the current one, to add on its capabilities. If a termcap entry does not use the "tc" capability, then of course the two lengths are the same.

The "before tc expansion" length is the most important one, because it affects more than just users of that particular terminal. This is the length of the entry as it exists in /etc/termcap, minus the backslash-newline pairs, which tgetent strips out while reading it. Some termcap libraries strip off the final newline, too (GNU termcap does not). Now suppose:

a termcap entry before expansion is more than 1023 bytes long,

and the application has only allocated a 1k buffer,

and the termcap library (like the one in BSD/OS 1.1 and GNU) reads the whole entry into the buffer, no matter what its length, to see if it is the entry it wants,

and tgetent is searching for a terminal type that either is the long entry, appears in the termcap file after the long entry, or does not appear in the file at all (so that tgetent has to search the whole termcap file).

Then tgetent will overwrite memory, perhaps its stack, and probably core dump the program. Programs like telnet are particularly vulnerable; modern telnets pass along values like the terminal type automatically. The results are almost as undesirable with a termcap library, like SunOS 4.1.3 and Ultrix 4.4, that prints warning messages when it reads an overly long termcap entry. If a termcap library truncates long entries, like OSF/1 3.0, it is immune to dying here but will return incorrect data for the terminal.

The "after tc expansion" length will have a similar effect to the above, but only for people who actually set TERM to that terminal type, since tgetent only does "tc" expansion once it is found the terminal type it was looking for, not while searching.

In summary, a termcap entry that is longer than 1023 bytes can cause, on various combinations of termcap libraries and applications, a core dump, warnings, or incorrect operation. If it is too long even before "tc" expansion, it will have this effect even for users of some other terminal types and users whose TERM variable does not have a termcap entry.

When in −C (translate to termcap) mode, the ncurses implementation of tic(1) issues warning messages when the pre-tc length of a termcap translation is too long. The −c (check) option also checks resolved (after tc expansion) lengths.

Binary Compatibility
It is not wise to count on portability of binary terminfo entries between commercial UNIX versions. The problem is that there are at least two versions of terminfo (under HP−UX and AIX) which diverged from System V terminfo after SVr1, and have added extension capabilities to the string table that (in the binary format) collide with System V and XSI Curses extensions.


Searching for terminal descriptions in $HOME/.terminfo and TERMINFO_DIRS is not supported by older implementations.

Some SVr4 curses implementations, and all previous to SVr4, do not interpret the %A and %O operators in parameter strings.

SVr4/XPG4 do not specify whether msgr licenses movement while in an alternate-character-set mode (such modes may, among other things, map CR and NL to characters that do not trigger local motions). The ncurses implementation ignores msgr in ALTCHARSET mode. This raises the possibility that an XPG4 implementation making the opposite interpretation may need terminfo entries made for ncurses to have msgr turned off.

The ncurses library handles insert-character and insert-character modes in a slightly non-standard way to get better update efficiency. See the Insert/Delete Character subsection above.

The parameter substitutions for set_clock and display_clock are not documented in SVr4 or the XSI Curses standard. They are deduced from the documentation for the AT&T 505 terminal.

Be careful assigning the kmous capability. The ncurses wants to interpret it as KEY_MOUSE, for use by terminals and emulators like xterm that can return mouse-tracking information in the keyboard-input stream.

X/Open Curses does not mention italics. Portable applications must assume that numeric capabilities are signed 16-bit values. This includes the no_color_video (ncv) capability. The 32768 mask value used for italics with ncv can be confused with an absent or cancelled ncv. If italics should work with colors, then the ncv value must be specified, even if it is zero.

Different commercial ports of terminfo and curses support different subsets of the XSI Curses standard and (in some cases) different extension sets. Here is a summary, accurate as of October 1995:

SVR4, Solaris, ncurses −− These support all SVr4 capabilities.

SGI −− Supports the SVr4 set, adds one undocumented extended string capability (set_pglen).

SVr1, Ultrix −− These support a restricted subset of terminfo capabilities. The booleans end with xon_xoff; the numerics with width_status_line; and the strings with prtr_non.

HP/UX −− Supports the SVr1 subset, plus the SVr[234] numerics num_labels, label_height, label_width, plus function keys 11 through 63, plus plab_norm, label_on, and label_off, plus some incompatible extensions in the string table.

AIX −− Supports the SVr1 subset, plus function keys 11 through 63, plus a number of incompatible string table extensions.

OSF −− Supports both the SVr4 set and the AIX extensions.



files containing terminal descriptions


tic(1), infocmp(1), ncurses(3NCURSES), printf(3), term(5). terminfo_variables(3NCURSES).


Zeyd M. Ben-Halim, Eric S. Raymond, Thomas E. Dickey. Based on pcurses by Pavel Curtis.