1# @(#)POSIX 8.1 (Berkeley) 6/6/93 2 3Comments on the IEEE P1003.2 Draft 12 4 Part 2: Shell and Utilities 5 Section 4.55: sed - Stream editor 6 7Diomidis Spinellis <dds@doc.ic.ac.uk> 8Keith Bostic <bostic@cs.berkeley.edu> 9 10In the following paragraphs, "wrong" usually means "inconsistent with 11historic practice", as most of the following comments refer to 12undocumented inconsistencies between the historical versions of sed and 13the POSIX 1003.2 standard. All the comments are notes taken while 14implementing a POSIX-compatible version of sed, and should not be 15interpreted as official opinions or criticism towards the POSIX committee. 16All uses of "POSIX" refer to section 4.55, Draft 12 of POSIX 1003.2. 17 18 1. 32V and BSD derived implementations of sed strip the text 19 arguments of the a, c and i commands of their initial blanks, 20 i.e. 21 22 #!/bin/sed -f 23 a\ 24 foo\ 25 \ indent\ 26 bar 27 28 produces: 29 30 foo 31 indent 32 bar 33 34 POSIX does not specify this behavior as the System V versions of 35 sed do not do this stripping. The argument against stripping is 36 that it is difficult to write sed scripts that have leading blanks 37 if they are stripped. The argument for stripping is that it is 38 difficult to write readable sed scripts unless indentation is allowed 39 and ignored, and leading whitespace is obtainable by entering a 40 backslash in front of it. This implementation follows the BSD 41 historic practice. 42 43 2. Historical versions of sed required that the w flag be the last 44 flag to an s command as it takes an additional argument. This 45 is obvious, but not specified in POSIX. 46 47 3. Historical versions of sed required that whitespace follow a w 48 flag to an s command. This is not specified in POSIX. This 49 implementation permits whitespace but does not require it. 50 51 4. Historical versions of sed permitted any number of whitespace 52 characters to follow the w command. This is not specified in 53 POSIX. This implementation permits whitespace but does not 54 require it. 55 56 5. The rule for the l command differs from historic practice. Table 57 2-15 includes the various ANSI C escape sequences, including \\ 58 for backslash. Some historical versions of sed displayed two 59 digit octal numbers, too, not three as specified by POSIX. POSIX 60 is a cleanup, and is followed by this implementation. 61 62 6. The POSIX specification for ! does not specify that for a single 63 command the command must not contain an address specification 64 whereas the command list can contain address specifications. The 65 specification for ! implies that "3!/hello/p" works, and it never 66 has, historically. Note, 67 68 3!{ 69 /hello/p 70 } 71 72 does work. 73 74 7. POSIX does not specify what happens with consecutive ! commands 75 (e.g. /foo/!!!p). Historic implementations allow any number of 76 !'s without changing the behaviour. (It seems logical that each 77 one might reverse the behaviour.) This implementation follows 78 historic practice. 79 80 8. Historic versions of sed permitted commands to be separated 81 by semi-colons, e.g. 'sed -ne '1p;2p;3q' printed the first 82 three lines of a file. This is not specified by POSIX. 83 Note, the ; command separator is not allowed for the commands 84 a, c, i, w, r, :, b, t, # and at the end of a w flag in the s 85 command. This implementation follows historic practice and 86 implements the ; separator. 87 88 9. Historic versions of sed terminated the script if EOF was reached 89 during the execution of the 'n' command, i.e.: 90 91 sed -e ' 92 n 93 i\ 94 hello 95 ' </dev/null 96 97 did not produce any output. POSIX does not specify this behavior. 98 This implementation follows historic practice. 99 10010. Deleted. 101 10211. Historical implementations do not output the change text of a c 103 command in the case of an address range whose first line number 104 is greater than the second (e.g. 3,1). POSIX requires that the 105 text be output. Since the historic behavior doesn't seem to have 106 any particular purpose, this implementation follows the POSIX 107 behavior. 108 10912. POSIX does not specify whether address ranges are checked and 110 reset if a command is not executed due to a jump. The following 111 program will behave in different ways depending on whether the 112 'c' command is triggered at the third line, i.e. will the text 113 be output even though line 3 of the input will never logically 114 encounter that command. 115 116 2,4b 117 1,3c\ 118 text 119 120 Historic implementations, and this implementation, do not output 121 the text in the above example. The general rule, therefore, 122 is that a range whose second address is never matched extends to 123 the end of the input. 124 12513. Historical implementations allow an output suppressing #n at the 126 beginning of -e arguments as well as in a script file. POSIX 127 does not specify this. This implementation follows historical 128 practice. 129 13014. POSIX does not explicitly specify how sed behaves if no script is 131 specified. Since the sed Synopsis permits this form of the command, 132 and the language in the Description section states that the input 133 is output, it seems reasonable that it behave like the cat(1) 134 command. Historic sed implementations behave differently for "ls | 135 sed", where they produce no output, and "ls | sed -e#", where they 136 behave like cat. This implementation behaves like cat in both cases. 137 13815. The POSIX requirement to open all w files at the beginning makes 139 sed behave nonintuitively when the w commands are preceded by 140 addresses or are within conditional blocks. This implementation 141 follows historic practice and POSIX, by default, and provides the 142 -a option which opens the files only when they are needed. 143 14416. POSIX does not specify how escape sequences other than \n and \D 145 (where D is the delimiter character) are to be treated. This is 146 reasonable, however, it also doesn't state that the backslash is 147 to be discarded from the output regardless. A strict reading of 148 POSIX would be that "echo xyz | sed s/./\a" would display "\ayz". 149 As historic sed implementations always discarded the backslash, 150 this implementation does as well. 151 15217. POSIX specifies that an address can be "empty". This implies 153 that constructs like ",d" or "1,d" and ",5d" are allowed. This 154 is not true for historic implementations or this implementation 155 of sed. 156 15718. The b t and : commands are documented in POSIX to ignore leading 158 white space, but no mention is made of trailing white space. 159 Historic implementations of sed assigned different locations to 160 the labels "x" and "x ". This is not useful, and leads to subtle 161 programming errors, but it is historic practice and changing it 162 could theoretically break working scripts. This implementation 163 follows historic practice. 164 16519. Although POSIX specifies that reading from files that do not exist 166 from within the script must not terminate the script, it does not 167 specify what happens if a write command fails. Historic practice 168 is to fail immediately if the file cannot be opened or written. 169 This implementation follows historic practice. 170 17120. Historic practice is that the \n construct can be used for either 172 string1 or string2 of the y command. This is not specified by 173 POSIX. This implementation follows historic practice. 174 17521. Deleted. 176 17722. Historic implementations of sed ignore the RE delimiter characters 178 within character classes. This is not specified in POSIX. This 179 implementation follows historic practice. 180 18123. Historic implementations handle empty RE's in a special way: the 182 empty RE is interpreted as if it were the last RE encountered, 183 whether in an address or elsewhere. POSIX does not document this 184 behavior. For example the command: 185 186 sed -e /abc/s//XXX/ 187 188 substitutes XXX for the pattern abc. The semantics of "the last 189 RE" can be defined in two different ways: 190 191 1. The last RE encountered when compiling (lexical/static scope). 192 2. The last RE encountered while running (dynamic scope). 193 194 While many historical implementations fail on programs depending 195 on scope differences, the SunOS version exhibited dynamic scope 196 behaviour. This implementation does dynamic scoping, as this seems 197 the most useful and in order to remain consistent with historical 198 practice. 199