This is an old revision of the document!

Obsolete and deprecated syntax

This (incomplete) page describes some syntax and commands considered obsolete by some measure. A through discussion of the rationale is beyond the scope of this page. See the portability page for a discussion on portability issues.

This first table lists syntax that is tolerated by Bash but has few if any legitimate uses. These features mostly exist for Bourne, csh, or some other backwards compatibility with obsolete shells, or were Bash-specific features considered failed experiments and deprecated or replaced with a better alternative. These should be irrelevant to most everyone except maybe code golfers. New scripts should never use them. None of the items on this list are specified by the most current version of POSIX, and some may be incompatible with POSIX.

Syntax Replacement Description
&>FILE and >&FILE >FILE 2>&1 This redirection syntax is short for >FILE 2>&1 and originates in the C Shell. The latter form is especially uncommon and should never be used, and the explicit form using separate redirections should be preferred over both. These shortcuts contribute to confusion about the copy descriptor because the syntax is unclear. They also introduce parsing ambiguity and conflict with POSIX. Shells without this feature treat cmd1 &>file cmd2 as: "background cmd1 and then execute cmd2 with its stdout redirected to file", which is the correct interpretation of this expression. See redirection
$[EXPRESSION] $((EXPRESSION)) This undocumented syntax is completely replaced by the POSIX-conforming arithmetic expansion $((EXPRESSION)). It is unimplemented almost everywhere except Bash and Zsh. See arithmetic expansion. Some discussion here:
COMMAND |& COMMAND COMMAND 2>&1 | COMMAND This is an alternate pipeline operator derived from Zsh. Officially, it is not considered deprecated by Bash, but I highly discourage it. It conflicts with the list operator used for coprocess creation in most Korn shells. It also has confusing behavior. The stdout is redirected first like an ordinary pipe, while the stderr is actually redirected last – after other redirects preceding the pipe operator. Overall, it's pointless syntax bloat. Use an explicit redirect instead.
function NAME() COMPOUND-CMD NAME() COMPOUND-CMD or function NAME { CMDS; } This is an amalgamation between the Korn and POSIX style function definitions - using both the function keyword and parentheses. It has no useful purpose and no historical basis or reason to exist. It is not specified by POSIX. It is accepted by Bash, mksh, zsh, and perhaps some other Korn shells, where it is treated as identical to the POSIX-style function. It is not accepted by AT&T ksh. It should never be used. See the next table for the function keyword. Bash doesn't have this feature documented as expressly deprecated.
for x; { …;} do, done, in, esac, etc. This undocumented syntax replaces the do and done reserved words with braces. Many Korn shells support various permutations on this syntax for certain compound commands like for, case, and while. Which ones and certain details like whether a newline or semicolon are required vary. Only for works in Bash. Needless to say, don't use it.

This table lists syntax that is specified by POSIX (unless otherwise specified below), but has been superseded by superior alternatives (either in POSIX, Bash, or both), or is highly discouraged for other reasons such as encouraging bad practices or dangerous code. Those that are specified by POSIX may be badly designed and unchangeable for historical reasons.

Syntax Replacement Description
Unquoted expansions, Word splitting, and Pathname expansion (globbing) Proper quoting, Ksh/Bash-style arrays, The "$@" expansion, The read builtin command Word-splitting is undoubtedly and by far the #1 source of bugs, security issues, and beginner mistakes in all of shell scripting. It is a misfeature reluctantly carried over into ksh (and consequently, POSIX) from the Bourne shell by David Korn due to complaints of broken scripts and changes in previously documented behavior. Most of the important expansions are performed at the same time from left to right. However, a few expansions, most notably word-splitting and globbing, and also in shells other than Bash, brace expansion, are performed on the results of previous expansions, by default, unless they are quoted. This means that the act of dereferencing a variable, depending on the value of the variable, can yield different results depending on possibly uncontrolled side-effects like the value of IFS, and the names of files in the current working directory. You can't get globbing without word-splitting, or vice versa (without set -f). You cannot store a command in a variable, and safely evaluate it by expanding it unquoted. If possible, always choose a shell with Korn shell arrays such as Bash. They are a vital but non-standard feature for writing clean, safe scripts. Well-written scripts don't use word-splitting at all. A significant proportion of the issues on the famous Pitfalls list fall under this category. See also:
`COMMANDS` $(COMMANDS) This is the older Bourne-compatible form of the command substitution. Both the `COMMANDS` and $(COMMANDS) syntaxes are specified by POSIX, but the latter is _greatly_ preferred, though the former is unfortunately still very prevalent in scripts. New-style command substitutions are widely implemented by every modern shell (and then some). The only reason for using backticks is for compatibility with a real Bourne shell (like Heirloom). Backtick command substitutions require special escaping when nested, and examples found in the wild are improperly quoted more often than not. See:
[ EXPRESSION ] and test EXPRESSION [[ EXPRESSION ]] test and [ are the Bourne/POSIX commands for evaluating test expressions (they are almost identical, and [ is somewhat more common). The expressions consist of regular arguments, unlike the Ksh/Bash [[ command. While the issue is analogous to let vs ((, the advantages of [[ vs [ are even more important because the arguments/expansions aren't just concatenated into one expression. With the classic [ command, the number of arguments is significant. If at all possible, use the conditional expression ("new test command") [[ EXPRESSION ]]. Unless there is a need for POSIX compatibility, there are only a few reasons to use [. [[ is one of the most portable and consistent non-POSIX ksh extensions available. See: The conditional expression and BashFAQ/031
set -e, set -o errexit, and the ERR trap proper control flow and error handling set -e causes untested non-zero exit statuses to be fatal. It is a debugging feature intended for use only during development and should not be used in production code, especially init scripts and other high-availability scripts. Do not be tempted to think of this as "error handling"; it's not, it's just a way to find the place you've forgotten to put error handling. Think of it as akin to "use strict" in Perl or "throws" in C++: tough love that makes you write better code. Many guides recommend avoiding it entirely because of the apparently-complex rules for when non-zero statuses cause the script to abort. Conversely, large software projects with experienced coders may recommend or even mandate its use. Because it provides no notification of the location of the error, it's more useful combined with set -x or the DEBUG trap and other Bash debug features, and both flags are normally better set on the command line rather than within the script itself. Most of this also applies to the ERR trap, though I've seen it used in a few places in shells that lack pipefail or PIPESTATUS. The ERR trap is not POSIX, while set -e is. failglob is another Bash feature that falls into this category (mainly useful for debugging). The set -e feature generates more questions and false bug reports on the Bash mailing list than all other features combined! Please do not rely upon set -e for logic in scripts. If you still refuse to take this advice, make sure you understand exactly how it works. For more details, see:
set -u or set -o nounset Proper control flow and error handling set -u causes attempts to expand unset variables or parameters as fatal errors. Like set -e, it bypasses control flow and exits immediately from the current shell environment. Like non-zero statuses, unset variables are a normal part of most non-trivial shell scripts. Living with set -u requires hacks like ${1+"$1"} for each expansion that might possibly be unset. Apparently some find it useful for debugging. See BashFAQ/083 for how to properly test for defined variables. Don't use set -u.
${var?msg} or ${var:?msg} Proper control flow and error handling Like set -u, this expansion throws a fatal error which immediately exits the current shell environment if the given parameter is unset (or unset or null). It prints the error message given to the right of the operator. If a value is expected and you'd like to create an assertion or throw errors, it is better to test for undefined variables using one of these techniques and handle the error manually, or call a die function. This expansion is defined by POSIX. It's better than set -u, because it's explicit, but not by much.

This table lists features that are discouraged by many "Bash-centric" experts and guides, but do have some legitimate uses if you know what you're doing, such as for those with specific portability requirements, or in order to make use of some subtle behavior. These are frequently (mis)used for no reason. Writing portable scripts that use non-POSIX features requires detailed knowledge to account for many (sometimes undocumented) differences across many shells. If you do happen to know what you're doing, don't be too surprised if you run across someone telling you not to use these. They'd be right 99% of the time.

Syntax Replacement Description
function NAME { CMDS; } NAME() COMPOUND-CMD This is the ksh form of function definition created to extend the Bourne and POSIX form with modified behaviors and additional features like local variables. The idea was for new-style functions to be analogous to regular builtins with their own environment and scope, while POSIX-style functions are more like special builtins. function is supported by almost every ksh-derived shell including Bash and Zsh, but isn't specified by POSIX. Bash treats all function styles the same, but this is unusual. function has some preferable characteristics in many ksh variants, making it more portable for scripts that use non-POSIX extensions by some measures. If you're going to use the function keyword, it implies that you're either targeting Ksh specifically, or that you have detailed knowledge of how to compensate for differences across shells. It should always be consistently used with typeset, never declare or local. Also in ksh93, the braces are not a command group, but a required part of the syntax (unlike Bash and others). See shell function definitions
typeset declare, local, export, readonly This is closely related to the above, and should often be used together. typeset exists primarily for ksh compatibility, but is marked as "deprecated" in Bash (though I don't entirely agree with this). This somewhat makes sense, because future compatibility can't be guaranteed, and any compatibility at all requires understanding the non-POSIX features of other shells and their differences. Using declare instead of typeset emphasizes your intention to be "bash-only", and definitely breaks everywhere else (except possibly zsh if you're lucky). The issue is further complicated by Dash and the Debian policy requirement for a local builtin, which is itself not entirely compatible with Bash and other shells.
let 'EXPR' ((EXPR)) or [ $((EXPR)) -ne 0 ] let is the "simple command" variant of arithmetic evaluation command, which takes regular arguments. Both let and ((expr)) were present in ksh88, and everything that supports one should support the other. Neither are POSIX. The compound variant is preferable because it doesn't take regular arguments for wordsplitting and globbing, which makes it safer and clearer. It is also usually faster, especially in Bash, where compound commands are typically significantly faster. Some of the (few) reasons for using let are detailed on the let page. See arithmetic evaluation compound command
eval Depends. Often code can be restructured to use better alternatives. eval is thrown in here for good measure, as sadly it is so often misused that any use of eval (even the rare clever one) is immediately dismissed as wrong by experts, and among the most immediate solutions abused by beginners. In reality, there are correct ways to use eval, and even cases in which it's necessary, even in sophisticated shells like Bash and Ksh. eval is unusual in that it is less frequently appropriate in more feature-rich shells than in more minimal shells like Dash, where it is used to compensate for more limitations. If you find yourself needing eval too frequently, it might be a sign that you're either better off using a different language entirely, or trying to borrow an idiom from some other paradigm that isn't well suited to the shell language. By the same token, there are some cases in which working too hard to avoid eval ends up adding a lot of complexity and sacrificing all portability. Basically, don't substitute a clever eval for something else that's a bit "too clever", just to avoid the eval, while still taking reasonable measures to avoid it where it is sensible to do so. See: The eval builtin command and BashFAQ/048 -- how to use eval safely
This website uses cookies for visitor traffic analysis. By using the website, you agree with storing the cookies on your computer.More information
You could leave a comment if you were logged in.
  • scripting/obsolete.1357503448.txt
  • Last modified: 2013/01/06 20:17
  • by ormaaj