Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
1 contributor

Users who have contributed to this file

1193 lines (864 sloc) 28.6 KB
RFC Author Status Area Comments Due Plan to implement
RFC0046
Robert Holt
Experimental-Accepted
Pipeline Chain Operators
2019-07-12 0000
true

Pipeline Chain Operators

Background

POSIX shells have what may be referred to as AND-OR lists, what the Open Group's Shell Command Language specification describes as:

a sequence of one or more pipelines separated by the operators && and ||

Semantically, these operators look at the exit code of the left-hand side pipeline and conditionally execute the right-hand side pipeline. && executes the right-hand side if the left-hand side "succeeded", and || executes the right-hand side if the left-hand side did not "succeed".

Some examples in bash:

echo 'Success' && echo 'Second success' # Executes both left-hand and right-hand command
echo 'Success' || echo 'Second success' # Executes only the left-hand command
false && echo 'Second success' # Executes only the left-hand command (false always "fails")
false || echo 'Second success' # Executes both left-hand and right-hand commands
false && echo foo || echo bar # Writes only "bar" to stdout (operators are left-associative)
true || echo foo && echo bar # Writes only "bar" to stdout

Similarly, cmd.exe also supports && and ||, which it terms conditional processing symbols. From the Command shell overview page:

When you run multiple commands with conditional processing symbols, the commands to the right of the conditional processing symbol act based upon the results of the command to the left of the conditional processing symbol.

...

&&: Use to run the command following && only if the command preceding the symbol is successful. Cmd.exe runs the first command, and then runs the second command only if the first command completed successfully.

...

||: Use to run the command following || only if the command preceding || fails. Cmd.exe runs the first command, and then runs the second command only if the first command did not complete successfully (receives an error code greater than zero).

Despite the wording here, && has a lower precedence than | in cmd.exe and can be used to sequence pipelines:

dir C:\ | sort && echo 'done'

Historically, these operators have been reserved for implementation in PowerShell for some time. Since PowerShell v2, including a && token in a PowerShell script results in the following parse error:

The token '&&' is not a valid statement separator in this version.

Despite the error message implying the intent to separate statements, the code to parse these operators occurs within the pipeline parsing logic, possibly implying the intent to implement them as command separators (within a pipeline).

Proposal outline

This RFC proposes:

  • The addition of && and || as pipeline chain operators to PowerShell.
  • That && and || may be used between PowerShell pipelines to conditionally sequence them.
  • That $? (PowerShell's execution success indicator) be used to determine the sequence from pipeline to pipeline.
  • That such sequences of pipelines using && and || be called pipeline chains.

Motivation

As a PowerShell user, I can chain commands and pipelines with && and || so that I have an ergonomic syntax to conditionally invoke side-effectful commands.

The chief motivation of pipeline chain operators is to make native commands (i.e. commands run by invoking an executable as a subprocess) simpler to use and sequence, as they are in other shells.

Often these commands perform some action, emit some informational output (and/or error output) and return an exit code.

The aim of chain operators is to make the action success as easy to process as the output, providing a convenient way to manipulate control flow around command outcome rather than output.

User Experience

Also see: test cases for implementation.

Pipeline chain operators are intended to behave as if pipelines were written as a sequence of statements conditioned on $?:

cmd1 && cmd2 || cmd3

should be the same as

cmd1
if ($?) { cmd2 }
if (-not $?) { cmd3 }

Native commands

In these examples:

  • echo is a native command that writes its argument as output and returns an exit code of 0
  • error is a native command that writes its argument as output and returns an exit code of 1

Simple successful command chain

echo 'Hello' && echo 'Again'
Hello
Again

Simple error after successful command

echo 'Hello' && error 'Bad'
Hello
Bad

Error followed by command in success case

error 'Bad' && echo 'Hello'
Bad

Error followed by command in failure case

error 'Bad' || echo 'Hello'
Bad
Hello

Command followed by command in failure case

echo 'Hello' || echo 'Again'
Hello

Error followed by error in failure case

error 'Bad' || error 'Very bad'
Bad
Very bad

Composite chain: 1st succeeds, 2nd is skipped, 3rd is run

echo 'Hi' || echo 'Message' && echo '2nd message'
Hi
2nd message

Composite chain: 1st fails, 2nd is run, 3rd is run

error 'Bad' || echo 'Message' && echo '2nd message'
Bad
Message
2nd message

Composite chain: 1st succeeds, 2nd fails, 3rd is run

echo 'Hi' && error 'Bad' || echo 'Message'
Hi
Bad
Message

Cmdlets and Functions

Cmdlets and functions work just like native commands, except they don't set $LASTEXITCODE and have other ways of expressing error conditions.

Here the same principle applies as with native commands; the statements proceed as if the next is wrapped in if ($?) { ... }.

Simple cmdlet chain: success then success

Write-Output "Hello" && Write-Output "Hello again"
Hello
Hello again

Simple cmdlet chain: success otherwise success

Write-Output "Hello" || Write-Output "Hello again"
Hello

Simple cmdlet chain: error then success

Write-Error "Bad" && Write-Output "Hello again"
Write-Error "Bad" : Bad
+ CategoryInfo          : NotSpecified: (:) [Write-Error], WriteErrorException
+ FullyQualifiedErrorId : Microsoft.PowerShell.Commands.WriteErrorException

Write-Error here emits a non-terminating error and so we proceed to evaluate $?.


Simple cmdlet chain: error otherwise success

Write-Error "Bad" || Write-Output "Hello again"
Write-Error "Bad" : Bad
+ CategoryInfo          : NotSpecified: (:) [Write-Error], WriteErrorException
+ FullyQualifiedErrorId : Microsoft.PowerShell.Commands.WriteErrorException

Hello again

Simple command chain: native success then cmdlet success

echo 'Hi' && Write-Output 'Hello'
Hi
Hello

Simple command chain: native error then cmdlet success

error 'Bad' && Write-Output 'Hello'
Bad

Simple command chain: cmdlet error otherwise native success

Write-Error 'Bad' || echo 'Message'
Bad
Message

Pipelines

Pipeline chains allow whole pipelines between chain operators. As above, when a pipeline ends other than with a terminating error, $? determines the chain logic.

The whole pipeline on the left-hand side of an operator will be evaluated before evaluating chain condition and then right-hand side.

Succeeding pipeline on the left hand side of a chain

1,2,3 | ForEach-Object { $_ + 1 } && Write-Output 'Hello'
2
3
4
Hello

Non-terminating error in pipeline

When some input fails while processing a pipeline, that sets $? for that command invocation and the pipeline chain proceeds accordingly.

function Test-NonTerminatingError
{
    [CmdletBinding()]
    param(
        [Parameter(ValueFromPipeline)]
        $Input
    )

    process
    {
        if ($Input -ne 2)
        {
            return $Input
        }

        # Write a non-terminating error when $Input is 2
        # Note that Write-Error will not set $? for the caller here

        $exception = [System.Exception]::new('Bad')
        $errorId = 'Bad'
        $errorCategory = 'InvalidData'
        $errorRecord = [System.Management.Automation.ErrorRecord]::new($exception, $errorId, $errorCategory, $null)

        $PSCmdlet.WriteError($errorRecord)
    }
}

1,2,3 | Test-NonTerminatingError && Write-Output 'Hello'
1
1,2,3 | ForEach-Object { if ($_ -eq 2) { Write-Error 'Bad' } else { $_ } } && Write-Output 'Hello' : Bad
+ CategoryInfo          : NotSpecified: (:) [Write-Error], WriteErrorException
+ FullyQualifiedErrorId : Microsoft.PowerShell.Commands.WriteErrorException

3

Non-terminating error with ||

1,2,3 | Test-NonTerminatingError || Write-Output 'Problem!'
1
1,2,3 | ForEach-Object { if ($_ -eq 2) { Write-Error 'Bad' } else { $_ } } && Write-Output 'Hello' : Bad
+ CategoryInfo          : NotSpecified: (:) [Write-Error], WriteErrorException
+ FullyQualifiedErrorId : Microsoft.PowerShell.Commands.WriteErrorException

3
Problem!

Pipeline-terminating error in a chain

function Test-PipelineTerminatingError
{
    [CmdletBinding()]
    param(
        [Parameter(ValueFromPipeline)]
        $Input
    )

    process
    {
        if ($Input -ne 2)
        {
            return $Input
        }

        # Write a non-terminating error when $Input is 2

        $exception = [System.Exception]::new('Bad')
        $errorId = 'Bad'
        $errorCategory = 'InvalidData'
        $errorRecord = [System.Management.Automation.ErrorRecord]::new($exception, $errorId, $errorCategory, $null)

        # Note the use of ThrowTerminatingError() rather than WriteError()
        $PSCmdlet.ThrowTerminatingError($errorRecord)
    }
}

1,2,3 | Test-PipelineTerminatingError && Write-Output 'Succeeded'
1
Test-PipelineTerminatingError : Bad
At line:1 char:9
+ 1,2,3 | Test-PipelineTerminatingError && Write-Output 'Succeeded'
+         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ CategoryInfo          : InvalidData: (:) [Test-PipelineTerminatingError], Exception
+ FullyQualifiedErrorId : Bad,Test-PipelineTerminatingError

Note that unlike with the non-terminating error, the pipeline does not proceed to process 3.

Pipeline termination is not chain termination

1,2,3 | Test-PipelineTerminatingError || Write-Output 'Failed'
1
Test-PipelineTerminatingError : Bad
At line:1 char:9
+ 1,2,3 | Test-PipelineTerminatingError || Write-Output 'Failed'
+         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ CategoryInfo          : InvalidData: (:) [Test-PipelineTerminatingError], Exception
+ FullyQualifiedErrorId : Bad,Test-PipelineTerminatingError

Failed

Here, the pipeline to the left of || is terminated, but the chain continues since || is used and $? is false, meaning Write-Output 'Failed' is executed.

Interaction with try/catch

If an error is caught from within a pipeline chain, the chain will be abandoned for the catch block.

try
{
    1,2,3 | Test-PipelineTerminatingError || Write-Output 'Failed'
}
catch
{
    Write-Output "Caught error"
}
1
Caught error

Script-terminating errors and error handling

Script-terminating errors supercede chain sequencing, just as they would in a semicolon-separated sequence of statements.

Uncaught errors will terminate the script.

function ThrowBad
{
    throw 'Bad'
}

ThrowBad || Write-Output 'Failed'
Bad
At line:3 char:5
+     throw 'Bad'                                                                                   +     ~~~~~~~~~~~
+ CategoryInfo          : OperationStopped: (Bad:String) [], RuntimeException
+ FullyQualifiedErrorId : Bad

When the error is caught, the flow of control abandons the chain for the catch as expected.

function ThrowBad
{
    throw 'Bad'
}

try
{
    ThrowBad || Write-Output 'Failed'
}
catch
{
    Write-Output $_.FullyQualifiedErrorId
}
Bad

If traps are set, they will continue or break the pipeline chain as configured.

trap
{
    Write-Output 'TRAP'
    break
}

function ThrowTerminating
{
    [CmdletBinding()]
    param()

    Write-Output 'In ThrowTerminating'
    $ex = [System.Exception]::new('Bad')
    $errId = 'Bad'
    $errCat = 'NotSpecified'
    $err = [System.Management.Automation.ErrorRecord]::new($ex, $errId, $errCat, $null)
    $PSCmdlet.ThrowTerminatingError($err)
}

ThrowTerminating && Write-Output 'Success'
In ThrowTerminating
TRAP
ThrowTerminating : Bad
At line:20 char:1
+ ThrowTerminating && Write-Output 'Success'
+ ~~~~~~~~~~~~~~~~
+ CategoryInfo          : NotSpecified: (:) [ThrowTerminating], Exception
+ FullyQualifiedErrorId : Bad,ThrowTerminating


trap
{
    Write-Output 'TRAP'
    continue
}

function ThrowTerminating
{
    [CmdletBinding()]
    param()

    Write-Output 'In ThrowTerminating'
    $ex = [System.Exception]::new('Bad')
    $errId = 'Bad'
    $errCat = 'NotSpecified'
    $err = [System.Management.Automation.ErrorRecord]::new($ex, $errId, $errCat, $null)
    $PSCmdlet.ThrowTerminatingError($err)
}

ThrowTerminating && Write-Output 'Success'
In ThrowTerminating
TRAP

trap
{
    Write-Output 'TRAP'
    continue
}

function ThrowTerminating
{
    [CmdletBinding()]
    param()

    Write-Output 'In ThrowTerminating'
    $ex = [System.Exception]::new('Bad')
    $errId = 'Bad'
    $errCat = 'NotSpecified'
    $err = [System.Management.Automation.ErrorRecord]::new($ex, $errId, $errCat, $null)
    $PSCmdlet.ThrowTerminatingError($err)
}

ThrowTerminating || Write-Output 'Continued'
In ThrowTerminating
TRAP
Continued

Assignment

Because pipeline chains are statements, they can be assigned from.

$x = Write-Output 'Hi' && Write-Output 'Hello'
$x
Hi
Hello

With pipeline chains, as with any other PowerShell statements, assignment either succeeds or does not; there is no concept of partial success.

try
{
    $x = Write-Output 'Hi' && throw 'Bad'
}
catch
{
    Write-Output "Error: $($_.FullyQualifiedErrorId)"
}
Write-Output "`$x: $x"
Error: Bad
$x:

(Compare this to $x = . { 'Hi'; throw 'Bad' }.)


Specification

Grammar

Pipeline chains will be implemented using the already reserved && and || operators. These operators will have the following grammar:

statement:
    | ... <existing statements except pipeline>   # Other statements
    | variable_expression "=" statement           # Assignment
    | pipeline_chain ["&"]                        # Pipeline chains

pipeline_chain:
    | pipeline
    | pipeline_chain "&&" [newlines] pipeline
    | pipeline_chain "||" [newlines] pipeline

Optional pipeline chaining

A pipeline chain is a chain of one or more pipelines and takes the place of a pipeline in the current PowerShell syntax.

In the degenerate case, a pipeline chain of a single pipeline will work exactly as pipelines currently do in PowerShell, so there is no breakage to the existing grammar.

Line continuation

After a pipeline chain operator, any newlines will be skipped in anticipation of the following pipeline.

For example, the following would be a single pipeline chain:

cmd1 &&
    cmd2 ||
    cmd3

If the end of file is reached after a pipeline chain operator, incomplete input will be reported so that integrating tools will know to keep prompting for input.

Left-associativity

Like in POSIX shells, pipeline chain operators will be left associative, meaning chain operators will group from left to right.

For example, given the following pipeline chain:

cmd1 && cmd2 || cmd3

This will be grouped as:

[[cmd1 && cmd2] || cmd3]

As a syntax tree, this would look like:

            ||
          /    \
        &&      cmd3
      /     \
   cmd1    cmd2

With the syntax tree deepening on the left as more operators are chained.

Semantically, this means that cmd1 && cmd2 would be evaluated first, and its result used to govern the evaluation of || cmd3.

Higher precedence than & and ;

Pipeline chain operators will have higher precedence than pipeline background operators (&) or statement separators (;).

For example:

cmd1 && cmd2 &

Will bind as:

[[cmd1 && cmd2] &]

Having the syntax tree:

           <statement>
          /           \
         &&            &
       /    \
     cmd1  cmd2

The consequence of this will be that an entire pipeline chain can be sent to a background job for evaluation, rather than individual pipelines within it.

Semantics

Pipeline "success"

The && and || operators proceed based on the "success" of the previous pipeline.

The marker of command success proposed is the $? automatic variable. This is proposed because:

  • It builds off an existing PowerShell concept
  • It applies to both native commands and cmdlets/functions
  • Diverging from this would create inconsistency
  • Changes to the behaviour of $? would be reasonably expected to change here

That is, there are the following direct equivalents:

cmd1 && cmd2

      |
      v

. { cmd1; if ($?) { cmd2 } }
cmd1 || cmd2

      |
      v

. { cmd1; if (-not $?) { cmd2 } }
cmd1 && cmd2 || cmd3

         |
         v

. { cmd1; if ($?) { cmd2 }; if (-not $?) { cmd3 } }
# Note that cmd1 failing runs cmd3

Also see the alternate proposals section.

Pipeline output

The output of a pipeline chain is the concatenation of all its pipelines, in the sequence of their output:

$x = 'a','b','c' | Write-Output && Write-Output 'd'
$x # 'a','b','c','d'
$x = 'a','b','c' | Write-Output && Write-Error 'Bad' || Write-Output 'd'
# Writes the error record 'Bad'
$x # 'a','b','c','d'

Commands that fail will have any pipeline output emitted before evaluation of the chain operator.

For example:

$x = cmd1 && cmd2

If cmd1 fails but emits output, $x will hold that value.

Error semantics

Errors will have the same semantics as the equivalent cmd1; if ($?) { cmd2 } syntax.

Non-terminating and pipeline-terminating errors will cause the immediate pipeline to continue, and the pipeline chain will evaluate as normal based on the value of $?.

Script-terminating errors will terminate the entire pipeline chain, unless a trap { continue } is used.

While output from a chain that later throws a script-terminating error will be written to the pipeline, it will not be assigned to a variable. This is consistent with existing assignment semantics in PowerShell.

Other notes

New Abstract Syntax Tree (AST) types

A single new AST leaf type is proposed, PipelineChainAst, which refers to a pipeline chain that can be used anywhere where a pipeline could be currently. This inherits from PipelineBaseAst.

ICustomAstVisitor2 and AstVisitor2 would be extended to deal with this new AST type, and .NET Core 3's new default interface implementation feature would be leveraged to ensure this does not break things as previous syntactic introductions have been forced to.

Statements may not be chained

PowerShell has a notable divergence from POSIX shells:

  • In POSIX shells, a pipeline is composed of statements
  • In PowerShell, a pipeline is a kind of statement

This means that bash allows the following:

if [ -n $VAR ]; then echo 'IF'; fi || echo 'AFTER-IF'

In fact, this can be evaluated in the background:

if [ -n $VAR ]; then echo 'IF'; fi || echo 'AFTER-IF' &

However, in PowerShell, pipelines are subordinate to statements. The proposed implementation of chain operators precludes use of && and || between statements like if or while, or constructions like $x = cmd1 && $y = cmd2 && $x + $y.

Also see the alternate proposals section.

Related Material

Alternate Proposals and Considerations

Command vs pipeline vs statement separation

An important syntactic and semantic question is what level && and || operate at (using brackets to denote syntactic groupings in $x = cmd1 | cmd2 && cmd3):

  • Between commands, where they occur within pipelines ($x = [[cmd1] | [cmd2 && cmd3]])
  • Between pipelines, where they occur within statements ($x = [[cmd1 | cmd2] && cmd3]])
  • Between statements ([$x = [cmd1 | cmd2]] && [cmd3])

In the POSIX shell, && and || separate pipelines, but pipelines encompass all statements. Statelines like if, for and case are compound commands where between keywords like if and fi, everything is considered a single command.

So for example the following is possible:

if [ -e ./file.txt ]; then echo 'File exists'; fi && echo 'File does not exist' | cat -

(This always prints File does not exist, since the if is always considered to succeed.)

In cmd.exe, the more template-driven approach means that if and for being commands treat && as part of the argument:

>for %i in (1 2 3) do echo %i && echo 'done'
>echo 1   && echo 'done'
1
'done'

>echo 2   && echo 'done'
2
'done'

>echo 3   && echo 'done'
3
'done'

In PowerShell, unlike in the POSIX shell, all pipelines are statements but not all statements are pipelines. This means we must choose between separating statements and separating pipelines.

Allowing && and || between statements in PowerShell might look like:

$x = cmd1 && $y = cmd2 && $x + $y

Other possibilities would include:

if ($condition) { cmd1 } && while ($anotherCondition) { cmd2 }
foreach ($v in 1..100)
{
    cmd1 $v && break
}

Background operator changes

In bash, background operators are lower precedence than chain separators (& has the same precedence as ;). In PowerShell, background operators are higher precedence.

Making chain operators apply to any statement would lead to the question of the precedence of the background operator.

The example:

cmd1 && cmd2 &

Could either be [[cmd1 && cmd2] &] or [cmd1 && [cmd2 &]].

Having chain operators apply to statements would mean the second case is the correct one, unless changes are made to the background operator precedence (which would also allow if ($condition) { cmd1 } &) for example).

The use of this is possibly less than the use of being able to background an entire pipeline chain; cmd1 & && cmd2 would be equivalent to cmd1 &; cmd2 since a backgrounded pipeline will never fail in the invoking context.

Such a change to the background operator precedence would cause ambiguity with uses like return cmd1 & and throw cmd1 &, where the background operator would currently apply to the pipeline under the keyword. To not break PowerShell's existing semantics, special behaviour would need to be defined for return $expr &.

Reasons against

  • Pipelines have an established concept of "success", whereas other statements generally do not.
  • Background operators become less useful with respect to chains unless their syntax is changed in a significant way.

Allowing control flow statements at the end of chains

An original addendum to the pipeline chain proposal was to allow adding control flow statements at the end of pipeline chains:

  • cmd1 && return 'Done'
  • cmd2 || throw 'Error'
  • cmd3 && break
  • cmd4 || continue
  • cmd5 || exit 1

This introduces complications:

  • A pipeline chain can be both over and under a return:

    cmd1 && return cmd2 && cmd3

    groups as

    cmd1 && [return [cmd2 && cmd3]]
    
  • A throw can do the same:

    cmd1 && throw 'a' && 'b'

    This is especially unhelpful since throw stringifies its given value, making a construction like the above much less useful than for return.

This also complicates the grammar and the AST, since:

  • We have to complexify logic about stamtents vs pipelines and control flow logic
  • More AST types may be needed to prevent bad AST constructions

By keeping control flow statements directly out of pipeline chains:

  • The grammar is simplified
  • We only need one AST type
  • There's no confusing embedding of chains over and under a return/throw/exit

Control flow statements can still be used by embedding them into a subexpression:

  • cmd1 && $(return 'Done')
  • cmd2 || $(throw 'Error')
  • cmd3 && $(break)
  • cmd4 || $(continue)
  • cmd5 || $(exit 1)

Reasons against

  • Control flow manipulation with native commands is currently not ergonomic
  • A construction like cmd || throw 'Failed' is very handy
  • Bad or confusing cases are corner cases, unlikely to be hit under normal usage

Different evaluations of command "success"

The current proposal is to simply use $? to determine chain continuation.

Alternatives include:

  • $LASTEXITCODE
  • Whether errors have been written
  • A new custom semantics

A problem is that exit code semantics assume that a non-zero exit code means command failure. POSIX shells also make this assumption, but the convention may not be as widespread on Windows platforms.

Reasons against

  • $LASTEXITCODE is specific to native commands and use with cmdlets and functions may lead to unexpected results. For example:

    cmd_that_fails
    Write-Output "SUCCESS" && Write-Output "ALSO SUCCESS"

    Will never write "ALSO SUCCESS" if $LASTEXITCODE is used.

  • Commands can fail without writing an error and some utilities will write to standard error without having failed (time being a good example on *nix).

  • Having chain operators use their own failure semantics would create more conceptual complexity and be inconsistent with the established $? concept.

  • Allowing configurability of the success determination would likely only make sense with a per-command configuration. This would already be achieved simply with the proposed $? semantics using a wrapper function.

Terminology

We should have a unified terminology to describe these operators in PowerShell for use with:

  • The experimental feature name
  • The about_ topic
  • Web searchability

This RFC proposes Pipeline chain operators. Other possibilities are:

  • AND-OR lists
  • Command chain operators
  • Command control operators
  • Bash control operators
  • Control operators
  • Short circuit operators
You can’t perform that action at this time.