/* $OpenBSD: README,v 1.4 2003/06/03 02:56:23 millert Exp $ */ /*- * Copyright (c) 1990, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Edward Wang at The University of California, Berkeley. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)README 8.1 (Berkeley) 6/6/93 */ Compilation notes: Compiler options: BYTE_ORDER (used only in ww.h) It should already be defined in machine/endian.h. The code knows about BIG_ENDIAN, LITTLE_ENDIAN, and PDP_ENDIAN. It only cares about byte order in words, so PDP_ENDIAN is the same as LITTLE_ENDIAN. OLD_TTY If you don't have Posix termios, then define this. VMIN_BUG Even if you have Posix termios, define this if the MIN and TIME feature in noncanonical mode doesn't work correctly. Ok, there's another one, STR_DEBUG. It turns on consistency checks in the string allocator. It's been left on since performace doesn't seem to suffer. There's an abort() somewhere when an inconsistency is found. It hasn't happened in years. The file local.h contains locally tunable constants. The makefile used to be updated with mkmf; it has been changed at various times to use cpp -M and, currently, mkdep. The only library it needs is termcap. Window, as is, only runs on 4.3 (or later) machines. On 4.2 machines, at least these modifications must be done: delete uses of window size ioctls: TIOCGWINSZ, TIOCSWINSZ, struct winsize add to ww.h typedef int fd_set; #define FD_ZERO(s) (*(s) = 0) #define FD_SET(b, s) (*(s) |= 1 << (b)) #define FD_ISSET(b, s) (*(s) & 1 << (b)) add to ww.h #define sigmask(s) (1 << (s) - 1) A few notes about the internals: The window package. Windows are opened by calling wwopen(). Wwwrite() is the primitive for writing to windows. Wwputc(), wwputs(), and wwprintf() are also supported. Some of the outputs to windows are delayed. Wwupdate() updates the terminal to match the internal screen buffer. Wwspawn() spawns a child process on the other end of a window, with its environment tailored to the window. Visible windows are doubly linked in the order of their overlap. Wwadd() inserts a window into the list at a given place. Wwdelete() deletes it. Windows not in the list are not visible, though wwwrite() still works. Window was written before the days of X and Sunview, so some of the terminology is not standard. Most functions return -1 on error. Wwopen() returns the null pointer. An error number is saved in wwerrno. Wwerror() returns an error string based on wwerrno suitable for printing. The terminal drivers perform all output to the physical terminal, including special functions like character and line insertion and deletion. The window package keeps a list of known terminals. At initialization time, the terminal type is matched against the list to find the right terminal driver to use. The last driver, the generic driver, matches all terminals and uses the termcap database. The interface between the window package the terminal driver is the `tt' structure. It contains pointers to functions to perform special functions and terminal output, as well as flags about the characteristics of the terminal. Most of these ideas are borrowed from the Maryland window package, which in turn is based on Goslin's Emacs. The IO system is semi-synchronous. Terminal input is signal driven, and everything else is done synchronously with a single select(). It is roughly event-driven, though not in a clean way. Normally, in both conversation mode and command mode, window sleeps in a select() in wwiomux() waiting for data from the pseudo-terminals. At the same time, terminal input causes SIGIO which is caught by wwrint(). The select() returns when at least one of the pseudo-terminals becomes ready for reading. Wwrint() is the interrupt handler for tty input. It reads input into a linear buffer accessed through four pointers: +-------+--------------+----------------+ | empty | data | empty | +-------+--------------+----------------+ ^ ^ ^ ^ | | | | wwib wwibp wwibq wwibe Wwrint() appends characters at the end and increments wwibq (*wwibq++ = c), and characters are taken off the buffer at wwibp using the wwgetc() and wwpeekc() macros. As is the convention in C, wwibq and wwibe point to one position beyond the end. In addition, wwrint() will do a longjmp(wwjmpbuf) if wwsetjmp is true. This is used by wwiomux() to interrupt the select() which would otherwise resume after the interrupt. (Actually, I hear this is not true, but the longjmp feature is used to avoid a race condition as well. Anyway, it means I didn't have to depend on a feature in a daily-changing kernel, but that's another story.) The macro wwinterrupt() returns true if the input buffer is non-empty. Wwupdate(), wwwrite(), and wwiomux() check this condition and will return at the first convenient opportunity when it becomes true. In the case of wwwrite(), the flag ww_nointr in the window structure overrides this. This feature allows the user to interrupt lengthy outputs safely. The structure of the input buffer is designed to avoid race conditions without blocking interrupts. Actually, wwsetjmp and wwinterrupt() are part of a software interrupt scheme used by the two interrupt catchers wwrint() and wwchild(). Asserting the interrupt lets the synchronous parts of the program know that there's an interesting asynchronous condition (i.e., got a keyboard character, or a child process died) that they might want to process before anything else. The synchronous routines can check for this condition with wwinterrupt() or by arranging that a longjmp() be done. Wwiomux() copies pseudo-terminal output into their corresponding windows. Without anything to do, it blocks in a select(), waiting for read ready on pseudo-terminals. Reads are done into per-window buffers in the window structures. When there is at least one buffer non-empty, wwiomux() finds the top most of these windows and writes it using wwwrite(). Then the process is repeated. A non-blocking select() is done after a wwwrite() to pick up any output that may have come in during the write, which may take a long time. Specifically, we use this to stop output or flush buffer when a pseudo-terminal tells us to (we use pty packet mode). The select() blocks only when all of the windows' buffers are empty. A wwupdate() is done prior to this, which is the only time the screen is guaranteed to be completely up to date. Wwiomux() loops until wwinterrupt() becomes true. The top level routine for all this is mloop(). In conversation mode, it simply calls wwiomux(), which only returns when input is available. The input buffer is then written to the pseudo-terminal of the current window. If the escape character is found in the input, command mode is entered. Otherwise, the process is repeated. In command mode, control is transferred to docmd() which returns only when conversation mode is reentered. Docmd() and other command processing routines typically wait for input in a loop: while (wwpeekc() < 0) wwiomux(); When the loop terminates, wwgetc() is used to read the input buffer. Output to the physical terminal is handled by the lowest level routines of the window package, in the files ttoutput.c and tt.h. The standard IO package is not used, to get better control over buffering and to use non-blocking reads in wwrint(). The buffer size is set to approximately one second of output time, based on the baudrate. The result of all this complexity is faster response time, especially in output stopping and flushing. Wwwrite() checks wwinterrupt() after every line. It also calls wwupdate() for each line it writes. The output buffer is limited to one second of output time. Thus, there is usually only a delay of one to two lines plus one second after a ^C or ^S. Also, commands that produce lengthy output can be aborted without actually showing all of it on the terminal. (Try the '?' command followed by escape immediately.)