The compiled file is created with the tic program, and read by the routine setupterm(3X). The file is divided into six parts: the header, terminal names, boolean flags, numbers, strings, and string table.
The header section begins the file. This section contains six short integers in the format described below. These integers are
Short integers are stored in two 8-bit bytes. The first byte contains the least significant 8 bits of the value, and the second byte contains the most significant 8 bits. (Thus, the value represented is 256*second+first.) The value -1 is represented by the two bytes 0377, 0377; other negative values are illegal. This value generally means that the corresponding capability is missing from this terminal. Note that this format corresponds to the hardware of the VAX and PDP-11 (that is, little-endian machines). Machines where this does not correspond to the hardware must read the integers as two bytes and compute the little-endian value.
The terminal names section comes next. It contains the first line of the terminfo description, listing the various names for the terminal, separated by the ``|'' character. The section is terminated with an ASCII NUL character.
The boolean flags have one byte for each flag. This byte is either 0 or 1 as the flag is present or absent. The capabilities are in the same order as the file <term.h>.
Between the boolean section and the number section, a null byte will be inserted, if necessary, to ensure that the number section begins on an even byte (this is a relic of the PDP-11's word-addressed architecture, originally designed in to avoid IOT traps induced by addressing a word on an odd byte boundary). All short integers are aligned on a short word boundary.
The numbers section is similar to the flags section. Each capability takes up two bytes, and is stored as a little-endian short integer. If the value represented is -1, the capability is taken to be missing.
The strings section is also similar. Each capability is stored as a short integer, in the format above. A value of -1 means the capability is missing. Otherwise, the value is taken as an offset from the beginning of the string table. Special characters in ^X or \c notation are stored in their interpreted form, not the printing representation. Padding information $<nn> and parameter information %x are stored intact in uninterpreted form.
The final section is the string table. It contains all the values of string capabilities referenced in the string section. Each string is null terminated.
The ncurses libraries and applications support extended terminfo binary format, allowing users to define capabilities which are loaded at runtime. This extension is made possible by using the fact that the other implementations stop reading the terminfo data when they have reached the end of the size given in the header. ncurses checks the size, and if it exceeds that due to the predefined data, continues to parse according to its own scheme.
First, it reads the extended header (5 short integers):
The count- and size-values for the extended string table include the extended capability names as well as extended capability values.
Using the counts and sizes, ncurses allocates arrays and reads data for the extended capabilities in the same order as the header information.
The extended string table contains values for string capabilities. After the end of these values, it contains the names for each of the extended capabilities in order, e.g., booleans, then numbers and finally strings.
Applications which manipulate terminal data can use the definitions described in term_variables(3X) which associate the long capability names with members of a TERMTYPE structure.
On occasion, 16-bit signed integers are not large enough.
With ncurses 6.1, a new format was introduced by making a few changes
to the legacy format:
.IP • 4
a different magic number (octal 01036)
.IP • 4
changing the type for the number array from signed 16-bit integers
to signed 32-bit integers.
To maintain compatibility, the library presents the same data structures to direct users of the TERMTYPE structure as in previous formats. However, that cannot provide callers with the extended numbers. The library uses a similar but hidden data structure TERMTYPE2 to provide data for the terminfo functions.
Note that it is possible for setupterm to expect a different set of capabilities than are actually present in the file. Either the database may have been updated since setupterm has been recompiled (resulting in extra unrecognized entries in the file) or the program may have been recompiled more recently than the database was updated (resulting in missing entries). The routine setupterm must be prepared for both possibilities - this is why the numbers and sizes are included. Also, new capabilities must always be added at the end of the lists of boolean, number, and string capabilities.
X/Open Curses does not specify a format for the terminfo database. UNIX System V curses used a directory-tree of binary files, one per terminal description.
Despite the consistent use of little-endian for numbers and the otherwise self-describing format, it is not wise to count on portability of binary terminfo entries between commercial UNIX versions. The problem is that there are at least three versions of terminfo (under HP-UX, AIX, and OSF/1) 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. See terminfo(5) for detailed discussion of terminfo source compatibility issues.
This implementation is by default compatible with the binary terminfo format used by Solaris curses, except in a few less-used details where it was found that the latter did not match X/Open Curses. The format used by the other Unix versions can be matched by building ncurses with different configuration options.
The magic number in a binary terminfo file is the first 16-bits (two bytes). Besides making it more reliable for the library to check that a file is terminfo, utilities such as file also use that to tell what the file-format is. System V defined more than one magic number, with 0433, 0435 as screen-dumps (see scr_dump(5)). This implementation uses 01036 as a continuation of that sequence, but with a different high-order byte to avoid confusion.
Direct access to the TERMTYPE structure is provided for legacy applications. Portable applications should use the tigetflag and related functions described in curs_terminfo(3X) for reading terminal capabilities.
A small number of terminal descriptions use uppercase characters in their names. If the underlying filesystem ignores the difference between uppercase and lowercase, ncurses represents the ``first character'' of the terminal name used as the intermediate level of a directory tree in (two-character) hexadecimal form.
adm3a|lsi adm3a,
am,
cols#80, lines#24,
bel=^G, clear= 32$<1>, cr=^M, cub1=^H, cud1=^J,
cuf1=^L, cup=\E=%p1%{32}%+%c%p2%{32}%+%c, cuu1=^K,
home=^^, ind=^J,
.sp .in +4
and a hexadecimal dump of the compiled terminal description:
.sp .in +40000 1a 01 10 00 02 00 03 00 82 00 31 00 61 64 6d 33 ........ ..1.adm3
0010 61 7c 6c 73 69 20 61 64 6d 33 61 00 00 01 50 00 a|lsi ad m3a...P.
0020 ff ff 18 00 ff ff 00 00 02 00 ff ff ff ff 04 00 ........ ........
0030 ff ff ff ff ff ff ff ff 0a 00 25 00 27 00 ff ff ........ ..%.'...
0040 29 00 ff ff ff ff 2b 00 ff ff 2d 00 ff ff ff ff ).....+. ..-.....
0050 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0060 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0070 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0080 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0090 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
00a0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
00b0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
00c0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
00d0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
00e0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
00f0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0100 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0110 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ........ ........
0120 ff ff ff ff ff ff 2f 00 07 00 0d 00 1a 24 3c 31 ....../. .....$<1
0130 3e 00 1b 3d 25 70 31 25 7b 33 32 7d 25 2b 25 63 >..=%p1% {32}%+%c
0140 25 70 32 25 7b 33 32 7d 25 2b 25 63 00 0a 00 1e %p2%{32} %+%c....
0150 00 08 00 0c 00 0b 00 0a 00 ........ .
Eric S. Raymond
documented legacy terminfo format, e.g., from pcurses.