|
|
1.1 root 1: /*-
2: * Copyright (c) 1990 The Regents of the University of California.
3: * All rights reserved.
4: *
5: * This code is derived from software contributed to Berkeley by
6: * Edward Wang at The University of California, Berkeley.
7: *
8: * Redistribution and use in source and binary forms are permitted provided
9: * that: (1) source distributions retain this entire copyright notice and
10: * comment, and (2) distributions including binaries display the following
11: * acknowledgement: ``This product includes software developed by the
12: * University of California, Berkeley and its contributors'' in the
13: * documentation or other materials provided with the distribution and in
14: * all advertising materials mentioning features or use of this software.
15: * Neither the name of the University nor the names of its contributors may
16: * be used to endorse or promote products derived from this software without
17: * specific prior written permission.
18: * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED
19: * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
20: * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
21: *
22: * @(#)README 3.13 (Berkeley) 6/6/90
23: */
24:
25: Compilation notes:
26:
27: There is only one compiler option:
28:
29: BYTE_ORDER (used only in ww.h)
30: It should already be defined in machine/endian.h.
31: The code knows about BIG_ENDIAN, LITTLE_ENDIAN, and PDP_ENDIAN.
32: It only cares about byte order in words, so PDP_ENDIAN
33: is the same as LITTLE_ENDIAN.
34:
35: Ok, there's another one, STR_DEBUG. It turns on consistency checks
36: in the string allocator. It's been left on since performace doesn't
37: seem to suffer. There's an abort() somewhere when an inconsistency
38: is found. It hasn't happened in years.
39:
40: The file local.h contains locally tunable constants.
41:
42: The makefile used to be updated with mkmf; it has been changed
43: at various times to use cpp -M and, currently, mkdep. The only library
44: it needs is termcap.
45:
46: Window, as is, only runs on 4.3 machines.
47:
48: On 4.2 machines, at least these modifications must be done:
49:
50: delete uses of window size ioctls: TIOCGWINSZ, TIOCSWINSZ,
51: struct winsize
52: add to ww.h
53: typedef int fd_set;
54: #define FD_ZERO(s) (*(s) = 0)
55: #define FD_SET(b, s) (*(s) |= 1 << (b))
56: #define FD_ISSET(b, s) (*(s) & 1 << (b))
57: add to ww.h
58: #define sigmask(s) (1 << (s) - 1)
59:
60:
61: A few notes about the internals:
62:
63: The window package. Windows are opened by calling wwopen().
64: Wwwrite() is the primitive for writing to windows. Wwputc(), wwputs(),
65: and wwprintf() are also supported. Some of the outputs to windows are
66: delayed. Wwupdate() updates the terminal to match the internal screen
67: buffer. Wwspawn() spawns a child process on the other end of a window,
68: with its environment tailored to the window. Visible windows are
69: doubly linked in the order of their overlap. Wwadd() inserts a window
70: into the list at a given place. Wwdelete() deletes it. Windows not in
71: the list are not visible, though wwwrite() still works. Window was
72: written before the days of X and Sunview, so some of the terminology
73: is not standard.
74:
75: Most functions return -1 on error. Wwopen() returns the null
76: pointer. An error number is saved in wwerrno. Wwerror() returns an
77: error string based on wwerrno suitable for printing.
78:
79: The terminal drivers perform all output to the physical terminal,
80: including special functions like character and line insertion and
81: deletion. The window package keeps a list of known terminals. At
82: initialization time, the terminal type is matched against the list to
83: find the right terminal driver to use. The last driver, the generic
84: driver, matches all terminals and uses the termcap database. The
85: interface between the window package the terminal driver is the `tt'
86: structure. It contains pointers to functions to perform special
87: functions and terminal output, as well as flags about the
88: characteristics of the terminal. Most of these ideas are borrowed
89: from the Maryland window package, which in turn is based on Goslin's
90: Emacs.
91:
92: The IO system is semi-synchronous. Terminal input is signal
93: driven, and everything else is done synchronously with a single
94: select(). It is roughly event-driven, though not in a clean way.
95:
96: Normally, in both conversation mode and command mode, window
97: sleeps in a select() in wwiomux() waiting for data from the
98: pseudo-terminals. At the same time, terminal input causes SIGIO which
99: is caught by wwrint(). The select() returns when at least one of the
100: pseudo-terminals becomes ready for reading.
101:
102: Wwrint() is the interrupt handler for tty input. It reads input
103: into a linear buffer accessed through four pointers:
104:
105: +-------+--------------+----------------+
106: | empty | data | empty |
107: +-------+--------------+----------------+
108: ^ ^ ^ ^
109: | | | |
110: wwib wwibp wwibq wwibe
111:
112: Wwrint() appends characters at the end and increments wwibq (*wwibq++
113: = c), and characters are taken off the buffer at wwibp using the
114: wwgetc() and wwpeekc() macros. As is the convention in C, wwibq
115: and wwibe point to one position beyond the end. In addition,
116: wwrint() will do a longjmp(wwjmpbuf) if wwsetjmp is true. This is
117: used by wwiomux() to interrupt the select() which would otherwise
118: resume after the interrupt. (Actually, I hear this is not true,
119: but the longjmp feature is used to avoid a race condition as well.
120: Anyway, it means I didn't have to depend on a feature in a
121: daily-changing kernel, but that's another story.) The macro
122: wwinterrupt() returns true if the input buffer is non-empty.
123: Wwupdate(), wwwrite(), and wwiomux() check this condition and will
124: return at the first convenient opportunity when it becomes true.
125: In the case of wwwrite(), the flag ww_nointr in the window structure
126: overrides this. This feature allows the user to interrupt lengthy
127: outputs safely. The structure of the input buffer is designed to
128: avoid race conditions without blocking interrupts.
129:
130: Actually, wwsetjmp and wwinterrupt() are part of a software
131: interrupt scheme used by the two interrupt catchers wwrint() and
132: wwchild(). Asserting the interrupt lets the synchronous parts of
133: the program know that there's an interesting asynchronous condition
134: (i.e., got a keyboard character, or a child process died) that they
135: might want to process before anything else. The synchronous routines
136: can check for this condition with wwinterrupt() or by arranging
137: that a longjmp() be done.
138:
139: Wwiomux() copies pseudo-terminal output into their corresponding
140: windows. Without anything to do, it blocks in a select(), waiting for
141: read ready on pseudo-terminals. Reads are done into per-window buffers
142: in the window structures. When there is at least one buffer non-empty,
143: wwiomux() finds the top most of these windows and writes it using
144: wwwrite(). Then the process is repeated. A non-blocking select() is
145: done after a wwwrite() to pick up any output that may have come in
146: during the write, which may take a long time. Specifically, we use
147: this to stop output or flush buffer when a pseudo-terminal tells us to
148: (we use pty packet mode). The select() blocks only when all of the
149: windows' buffers are empty. A wwupdate() is done prior to this, which
150: is the only time the screen is guaranteed to be completely up to date.
151: Wwiomux() loops until wwinterrupt() becomes true.
152:
153: The top level routine for all this is mloop(). In conversation
154: mode, it simply calls wwiomux(), which only returns when input is
155: available. The input buffer is then written to the pseudo-terminal of
156: the current window. If the escape character is found in the input,
157: command mode is entered. Otherwise, the process is repeated. In
158: command mode, control is transferred to docmd() which returns only when
159: conversation mode is reentered. Docmd() and other command processing
160: routines typically wait for input in a loop:
161:
162: while (wwpeekc() < 0)
163: wwiomux();
164:
165: When the loop terminates, wwgetc() is used to read the input buffer.
166:
167: Output to the physical terminal is handled by the lowest level
168: routines of the window package, in the files ttoutput.c and tt.h. The
169: standard IO package is not used, to get better control over buffering
170: and to use non-blocking reads in wwrint(). The buffer size is set to
171: approximately one second of output time, based on the baudrate.
172:
173: The result of all this complexity is faster response time,
174: especially in output stopping and flushing. Wwwrite() checks
175: wwinterrupt() after every line. It also calls wwupdate() for each line
176: it writes. The output buffer is limited to one second of output time.
177: Thus, there is usually only a delay of one to two lines plus one second
178: after a ^C or ^S. Also, commands that produce lengthy output can be
179: aborted without actually showing all of it on the terminal. (Try the
180: '?' command followed by escape immediately.)
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.