1
2Comments on the IEEE P1003.2 Draft 12
3     Part 2: Shell and Utilities
4  Section 4.55: sed - Stream editor
5
6Diomidis Spinellis <dds@doc.ic.ac.uk>
7Keith Bostic <bostic@cs.berkeley.edu>
8
9In the following paragraphs, "wrong" usually means "inconsistent with
10historic practice", as most of the following comments refer to
11undocumented inconsistencies between the historical versions of sed and
12the POSIX 1003.2 standard.  All the comments are notes taken while
13implementing a POSIX-compatible version of sed, and should not be
14interpreted as official opinions or criticism towards the POSIX committee.
15All uses of "POSIX" refer to section 4.55, Draft 12 of POSIX 1003.2.
16
17 1.	32V and BSD derived implementations of sed strip the text
18	arguments of the a, c and i commands of their initial blanks,
19	i.e.
20
21	#!/bin/sed -f
22	a\
23		foo\
24		\  indent\
25		bar
26
27	produces:
28
29	foo
30	  indent
31	bar
32
33	POSIX does not specify this behavior as the System V versions of
34	sed do not do this stripping.  The argument against stripping is
35	that it is difficult to write sed scripts that have leading blanks
36	if they are stripped.  The argument for stripping is that it is
37	difficult to write readable sed scripts unless indentation is allowed
38	and ignored, and leading whitespace is obtainable by entering a
39	backslash in front of it.  This implementation follows the BSD
40	historic practice.
41
42 2.	Historical versions of sed required that the w flag be the last
43	flag to an s command as it takes an additional argument.  This
44	is obvious, but not specified in POSIX.
45
46 3.	Historical versions of sed required that whitespace follow a w
47	flag to an s command.  This is not specified in POSIX.  This
48	implementation permits whitespace but does not require it.
49
50 4.	Historical versions of sed permitted any number of whitespace
51	characters to follow the w command.  This is not specified in
52	POSIX.  This implementation permits whitespace but does not
53	require it.
54
55 5.	The rule for the l command differs from historic practice.  Table
56	2-15 includes the various ANSI C escape sequences, including \\
57	for backslash.  Some historical versions of sed displayed two
58	digit octal numbers, too, not three as specified by POSIX.  POSIX
59	is a cleanup, and is followed by this implementation.
60
61 6.	The POSIX specification for ! does not specify that for a single
62	command the command must not contain an address specification
63	whereas the command list can contain address specifications.  The
64	specification for ! implies that "3!/hello/p" works, and it never
65	has, historically.  Note,
66
67		3!{
68			/hello/p
69		}
70
71	does work.
72
73 7.	POSIX does not specify what happens with consecutive ! commands
74	(e.g. /foo/!!!p).  Historic implementations allow any number of
75	!'s without changing the behaviour.  (It seems logical that each
76	one might reverse the behaviour.)  This implementation follows
77	historic practice.
78
79 8.	Historic versions of sed permitted commands to be separated
80	by semi-colons, e.g. 'sed -ne '1p;2p;3q' printed the first
81	three lines of a file.  This is not specified by POSIX.
82	Note, the ; command separator is not allowed for the commands
83	a, c, i, w, r, :, b, t, # and at the end of a w flag in the s
84	command.  This implementation follows historic practice and
85	implements the ; separator.
86
87 9.	Historic versions of sed terminated the script if EOF was reached
88	during the execution of the 'n' command, i.e.:
89
90	sed -e '
91	n
92	i\
93	hello
94	' </dev/null
95
96	did not produce any output.  POSIX does not specify this behavior.
97	This implementation follows historic practice.
98
9910.	Deleted.
100
10111.	Historical implementations do not output the change text of a c
102	command in the case of an address range whose first line number
103	is greater than the second (e.g. 3,1).  POSIX requires that the
104	text be output.  Since the historic behavior doesn't seem to have
105	any particular purpose, this implementation follows the POSIX
106	behavior.
107
10812.	POSIX does not specify whether address ranges are checked and
109	reset if a command is not executed due to a jump.  The following
110	program will behave in different ways depending on whether the
111	'c' command is triggered at the third line, i.e. will the text
112	be output even though line 3 of the input will never logically
113	encounter that command.
114
115	2,4b
116	1,3c\
117		text
118
119	Historic implementations did not output the text in the above
120	example.  Therefore it was believed that a range whose second
121	address was never matched extended to the end of the input.
122	However, the current practice adopted by this implementation,
123	as well as by those from GNU and SUN, is as follows:  The text
124	from the 'c' command still isn't output because the second address
125	isn't actually matched; but the range is reset after all if its
126	second address is a line number.  In the above example, only the
127	first line of the input will be deleted.
128
12913.	Historical implementations allow an output suppressing #n at the
130	beginning of -e arguments as well as in a script file.  POSIX
131	does not specify this.  This implementation follows historical
132	practice.
133
13414.	POSIX does not explicitly specify how sed behaves if no script is
135	specified.  Since the sed Synopsis permits this form of the command,
136	and the language in the Description section states that the input
137	is output, it seems reasonable that it behave like the cat(1)
138	command.  Historic sed implementations behave differently for "ls |
139	sed", where they produce no output, and "ls | sed -e#", where they
140	behave like cat.  This implementation behaves like cat in both cases.
141
14215.	The POSIX requirement to open all w files at the beginning makes
143	sed behave nonintuitively when the w commands are preceded by
144	addresses or are within conditional blocks.  This implementation
145	follows historic practice and POSIX, by default, and provides the
146	-a option which opens the files only when they are needed.
147
14816.	POSIX does not specify how escape sequences other than \n and \D
149	(where D is the delimiter character) are to be treated.  This is
150	reasonable, however, it also doesn't state that the backslash is
151	to be discarded from the output regardless.  A strict reading of
152	POSIX would be that "echo xyz | sed s/./\a" would display "\ayz".
153	As historic sed implementations always discarded the backslash,
154	this implementation does as well.
155
15617.	POSIX specifies that an address can be "empty".  This implies
157	that constructs like ",d" or "1,d" and ",5d" are allowed.  This
158	is not true for historic implementations or this implementation
159	of sed.
160
16118.	The b t and : commands are documented in POSIX to ignore leading
162	white space, but no mention is made of trailing white space.
163	Historic implementations of sed assigned different locations to
164	the labels "x" and "x ".  This is not useful, and leads to subtle
165	programming errors, but it is historic practice and changing it
166	could theoretically break working scripts.  This implementation
167	follows historic practice.
168
16919.	Although POSIX specifies that reading from files that do not exist
170	from within the script must not terminate the script, it does not
171	specify what happens if a write command fails.  Historic practice
172	is to fail immediately if the file cannot be opened or written.
173	This implementation follows historic practice.
174
17520.	Historic practice is that the \n construct can be used for either
176	string1 or string2 of the y command.  This is not specified by
177	POSIX.  This implementation follows historic practice.
178
17921.	Deleted.
180
18122.	Historic implementations of sed ignore the RE delimiter characters
182	within character classes.  This is not specified in POSIX.  This
183	implementation follows historic practice.
184
18523.	Historic implementations handle empty RE's in a special way: the
186	empty RE is interpreted as if it were the last RE encountered,
187	whether in an address or elsewhere.  POSIX does not document this
188	behavior.  For example the command:
189
190		sed -e /abc/s//XXX/
191
192	substitutes XXX for the pattern abc.  The semantics of "the last
193	RE" can be defined in two different ways:
194
195	1. The last RE encountered when compiling (lexical/static scope).
196	2. The last RE encountered while running (dynamic scope).
197
198	While many historical implementations fail on programs depending
199	on scope differences, the SunOS version exhibited dynamic scope
200	behaviour.  This implementation does dynamic scoping, as this seems
201	the most useful and in order to remain consistent with historical
202	practice.
203