Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Our Error Handling, Ourselves - time to fully understand and properly document PowerShell's error handling #1583

Open
mklement0 opened this issue Aug 15, 2017 · 42 comments
Open

Our Error Handling, Ourselves - time to fully understand and properly document PowerShell's error handling #1583

mklement0 opened this issue Aug 15, 2017 · 42 comments
Labels
area-error-handling Area - Error handling

Comments

@mklement0
Copy link
Contributor

mklement0 commented Aug 15, 2017 •
edited
Loading

The existing help topics that touch on error handling (about_Throw,
about_CommonParameters,
about_Preference_Variables, about_Trap, about_Try_Catch_Finally
):

  • have sown longstanding confusion due to conflating the two distinct types of terminating errors:

    • statement-terminating errors, as reported by cmdlets in certain non-recoverable situations (via the .ThrowTerminatingError() method) and by expressions in which a .NET exception / a PS runtime error occurs.
    • script-terminating errors (fatal errors), as either triggered by Throw or by escalating one of the other error types via error-action preference-variable / parameter value Stop.

  • have always contained the incorrect claim that the error-action preference / parameter values only pertain to non-terminating errors - which is true for the -ErrorAction parameter, but not the $ErrorActionPreference preference variable - see Clarify the intended behavior/documentation/terminology of terminating errors PowerShell/PowerShell#4292

It's time to:

Below is my understanding of how PowerShell's error handling actually works as of Windows PowerShell v5.1 / PowerShell Core v7.3.4, which can serve as a starting point for about_Error_Handling, along with links to issues to related problems.

Do tell me if and where I got things wrong.
The sheer complexity of PowerShell's current error handling is problematic, though I do realize that making changes in this area is a serious backward-compatibility concern.

  • Types of errors:

    • True PowerShell errors:

      • Non-terminating errors are issued by cmdlets or functions to signal failure with respect to specific inputs while continuing to process further (pipeline) input by default.

        • This allows potentially long-running commands to run to completion, despite partial failure. The errors reported can then be inspected via the error records collected in automatic variable $Error, allowing reprocessing of only the failed objects later (see below).

        • Note, however, that is possible for all input objects to cause nonterminating errors, amounting to complete failure overall.

        • Most cmdlet errors are non-terminating errors; e.g.: '/NoSuch', '/' | Get-Item reports an error for non-existent path /NoSuch, but continues processing with valid path /.

      • Terminating errors:

        • Important: In the context of remoting - whether explicit (e.g., via Invoke-Command) or implicit (e.g., via modules using implicit remoting) - terminating errors (of either kind) are converted to non-terminating ones.
        • Types of terminating errors:

          • Script-terminating (fatal) errors:

            • By default, they abort the entire enclosing script as well as any calling scripts. On the command line, a single statement or a list of statements submitted together can be thought of as an implicit script.
              • Only a try / catch or Trap statement can prevent hat.
            • The only way to directly trigger such an error is with the Throw keyword.
            • Note: Script-terminating errors are in effect fatal errors from the perspective of your code, if unhandled, - see Cmdlet.ThrowTerminatingError() is not affected by -ErrorAction PowerShell/PowerShell#14819 (comment) for a technical discussion.

          • Statement-terminating errors are the (statement-level) counterparts to non-terminating errors: they terminate the enclosing statement (pipeline or expression) and are issued to signal that a statement encountered a problem that doesn't allow it to meaningfully start or continue processing.

            • Important:
              • Statement-terminating errors truly only terminate the statement at hand. By default, the enclosing script continues to run.
              • The statement that is terminated is only the immediately enclosing statement; therefore, for instance, a statement-terminating error that occurs in the (-Process) script block of a ForEach-Object call does NOT terminate the pipeline as a whole - see below for an example.
            • Situations in which statement-terminating errors occur:
              • PowerShell's fundamental inability to even invoke a command generates a statement-terminating runtime error, namely:
                • A non-existent command:
                  nosuch -l # no such command exists
                • A cmdlet or function call with invalid syntax, such as incorrect parameter names or missing or mistyped parameter values:
                  Get-Item -Foo # -Foo is an invalid parameter name
                  Select-Object -First notanumber # -First expects an [int]
                  • Exception: If the mistyped value is bound via the pipeline, the resulting error is non-terminating see @alx9r's example.
                • Attempting to call another script that fails to parse (that is syntactically invalid).
              • While rare, cmdlet invocations that succeed syntactically can themselves generate statement-terminating errors, such as based on the contents of parameter values (not caught by parameter validation) or the state of the environment.
              • Strict-mode violations (such as trying to access a nonexistent variable when Set-StrictMode -Version 1 or higher is in effect).
              • Expressions can trigger statement-terminating errors too, namely:

                • via PowerShell expression runtime errors; e.g.:
                  1 / 0

                • via exceptions thrown by .NET method calls; e.g.:
                  [int]::Parse('foo')

                • via statement-terminating errors that occurs inside a (...) subexpression (but not inside $(...) or @(...), which are independent statement contexts); e.g.: (Get-Item -Foo) + 'hi' is a statement composed of a single expression that is terminated as a whole; by contrast, $(Get-Item -Foo) + 'hi' is a statement composed of an expression and an embedded statement, so only the $(...) part is terminated - + 'hi' is still executed.

                  • As an aside: Better names for operators $() and @() would therefore have been [array] substatement operators rather than [array] subexpression operators.

    • Failures signaled by external utilities (command-line / console applications such as findstr.exe on Windows and terminal-based utilities such as awk on Unix) via their exit codes are non-terminating - in fact, external utilities reporting nonzero exit codes are not errors in a PowerShell sense.

      • Note that the only standardized information that a utility's exit code carries is whether it succeeded overall (exit code 0) or failed overall (nonzero exit code), with no further distinction.
        • The exit code of the most recently executed external utility is stored in the automatic $LASTEXITCODE variable - see below.
      • Stderr output from external utilities is NOT considered error output - see below.

  • Default error actions, logging and success indicators:

    • Default actions when errors occur:

      • a non-terminating error:

      • a statement-terminating error:

        • terminates the current statement, but the script continues processing.
          • Important: The statement that is terminated is only the immediately enclosing statement; therefore, for instance, a statement-terminating error that occurs in the (-Process) script block of a ForEach-Object call does NOT terminate the pipeline as a whole - it only terminates the script-block invocation at hand.
            • Example: 1, 2 | ForEach-Object -Process { Get-Item -Foo } -End { 'pipeline ran to completion' }
              • The Get-Item -Foo calls cause statement-terminating errors, but the statements they terminate are the instances of the -Process script blocks, so pipeline processing continues, and pipeline ran to completion prints.
        • is output to PowerShell's error stream.
        • is logged in the automatic $Error collection.

      • a script-terminating error:

        • terminates the entire script
        • is output to PowerShell's error stream
        • is logged in the automatic $Error collection (which only matters if the session isn't terminated as a whole)

      • an external utility signaling failure by exit code / producing stderr output:

        • does not terminate the current statement (pipeline).
        • While stderr output is sent to PowerShell's error stream, it is not formatted like an error in the console, and by itself does not imply that an error occurred.
        • Stderr output is NOT logged in the automatic $Error collection

    • Default success indicators:

      • Automatic Boolean variable $? reflects whether the most recent statement, including calls to external utilities, experienced any error:

        • $True, if none, and $False, if at least one error occurred (because, in the case of non-terminating errors, multiple errors may have occurred).

        • In short: $? returning $False tells you only that some error occurred:

          • You cannot infer how many inputs experienced failure, which can range from 1 to all of them.
          • You cannot infer (from $? alone) whether a terminating error occurred or not.

        • Caveat: Because $? is set by every statement, its value is only meaningful immediately after the statement of interest.

        • $? is NOT set / set as expected in the following cases:

          • If a (by definition terminating) error was handled with a Try/Catch or Trap statement - unless the Catch handler is empty. (Non-terminating errors cannot be caught that way.)
            • With a Catch or Finally block present, the success of whatever statement appears last inside of them, if any, determines what $? is set to, with a (non-empty) Finally block taking precedence.
          • If the error is a non-terminating error and the command originating the error is enclosed in (...) - which turns the command into an expression - it is then the expression's own success that is reflected in $?:
            (Get-Item /NoSuch); $? yields $True, because the expression itself - whose sole purpose was to wrap the command - technically succeeded.
            (Get-Item -Foo); $? yields $False, because the statement-terminating error terminated the expression as a whole.

      • When calling external utilities, automatic variable $LASTEXITCODE complements $? by containing the specific exit code set by the most recently executed external utility.

        • $? is set to $True if the utility's exit code was 0, and to $False otherwise.

        • Note that while $? is only meaningful immediately after the statement of interest, $LASTEXITCODE remains relevant until the next external-utility call is made; either way, however, it is preferable to save the value in another variable if it needs to be inspected later.

      • Note: Currently, PowerShell lacks operators that allow chaining of commands based on whether they indicate success or not, such as the && and || control operators in Bash:

    • Default logging:

      • Errors are logged in memory (for the duration of the session) in the global automatic $Error variable in reverse chronological order (most recent error first; i.e., $Error[0] refers to the most recent error):

        • $Error is a collection of type [System.Collections.ArrayList] and the errors are stored as error records of type [System.Management.Automation.ErrorRecord] that wrap the underlying .NET exception, which all errors ultimately are (instances of [System.Exception] or a derived type).

        • To inspect $Error items, pipe them to Format-List -Force (direct output would result in the same format as when the error originally occurred); e.g.: $Error[0] | Format-List -Force

        • You can clear the collection anytime with $Error.Clear()

        • Errors are NOT logged in the following circumstances:

          • When non-terminating errors occur in a cmdlet / advanced function to which -ErrorAction Ignore was passed (see below).

          • What external utilities print to stderr is not considered error output, therefore it is NOT logged in $Error (see below).

  • Modifying the default error actions and logging behavior:

    • PowerShell commands and expressions:

      • Via common parameter -ErrorAction or preference variable $ErrorActionPreference:

        • Per the documentation as of this writing, specifying an error action should only affect non-terminating errors.

        • In the context of remoting, terminating errors (of either type) are converted to non-terminating ones.

          • Caveat: When invoking functions from a module that uses implicit remoting, neither -ErrorAction nor $ErrorActionPreference work as expected:

        • The supported action values are:

          • Continue ... non-terminating errors only: output errors and log them, but continue processing the current statement (non-terminating errors).
          • Stop ... non-terminating errors and statement-terminating errors via $ErrorActionPreference only: escalate the error to a script-terminating one.
          • SilentlyContinue ... like Continue, but silence error output, while still logging errors. Via $ErrorActionPreference only, also applies to both types of terminating errors (processing continues).
          • Ignore (-ErrorAction parameter only) ... non-terminating errors only: like SilentlyContinue, but without logging errors in $Error. Due to Ignore not being supported via $ErrorActionPreference, n/a to terminating errors.
          • Inquire ... prompt the user for the desired action, including the option to temporarily enter a nested session for debugging. Via $ErrorActionPreference only, also applies to both types of terminating errors.
          • Suspend (workflows only) ... automatically suspends a workflow job to allow for investigation.

        • Ad-hoc, when calling cmdlet/advanced functions, you can pass common parameter -ErrorAction to modify the behavior of non-terminating errors (only!).

          • As stated, the -ErrorAction parameter has no impact on terminating errors - in line with documented, but debatable behavior.

        • Scope-wide (including descendant scopes, unless overridden there), you can set preference variable $ErrorActionPreference, which sets the scope's default behavior for all occurrences of non-terminating behaviors, and - against documented behavior - also for terminating errors.

        • The -ErrorAction parameter takes precedence over the $ErrorActionPreference variable.

        • Non-terminating errors:

          • may alternatively be silenced with 2>$null, analogous to silencing stderr output from external utilities (see below). Terminating errors ignore 2>$null
          • may additionally be collected command-scoped in a user variable, via common parameter -ErrorVariable, unless -ErrorAction Ignore is also used. Note that use of -ErrorVariable does not affect the error output behavior; e.g., with error action Continue in effect, non-terminating errors still print to the console, whether or not you use -ErrorAction.

      • Catching terminating errors with Try / Catch or Trap statements:

        • Important: Only terminating errors (of either type) can be caught this way, and Try / Catch and Trap are effective irrespective of the current $ErrorActionPreference value (and any -ErrorAction common parameter, which fundamentally only applies to non-terminating errors).

        • Inside a Catch block:

          • Automatic variable $_ contains the [System.Management.Automation.ErrorRecord] instance representing the terminating error at hand.
          • You can use just Throw (without an argument) to re-throw the error at hand.

        • You may define multiple Catch blocks by applying optional filters based on the underlying .NET exception types - see Get-Help about_Try_Catch_Finally.

        • Error logging: Errors caught this way are still logged in the $Error collection, but there is debate around that - see Why do handled exceptions show in ErrorVariable? PowerShell/PowerShell#3768

    • External utilities:

      • Given that an external utility signaling failure by returning a nonzero exit code is not a true PowerShell error:

        • A failure signaled this way cannot be escalated to a script-terminating error via the $ErrorActionPreference variable.

        • Error messages (printed to stderr) are formatted like regular output and are NOT logged in the automatic $Error collection variable, nor do they affect how automatic variable $? is set (that is solely based the utility's exit code).

          • The rationale is that external utilities, unlike PowerShell commands, have only two output streams at their disposal: stdout for success (data) output, and stderr for everything else, and while error messages are printed to stderr, other non-data output (warnings, status information) is sent there too, so you cannot make the assumption that all stderr output represents errors.

      • However, you can capture stderr output for later inspection:

        • Using output redirection 2> allows you to suppress stderr output or send it to a file:
          whoami invalidarg 2>$null suppresses stderr output;
          whoami invalidarg 2>err.txt captures stderr output in file err.txt

          • Caveat: Due to a bug still present as of PowerShell Core v6.0.0-beta.5, when you use any 2> redirection (redirection of PowerShell's error stream):

        • There is currently no convenient mechanism for collections stderr lines in a variable, but, as a workaround, you can use redirection 2>&1 to merge stderr output into PowerShell's success output stream (interleaved with stdout output), which allows you to filter out the stderr messages later by type, because PowerShell wraps the stderr lines in [System.Management.Automation.ErrorRecord] instances; e.g.:

  • Reporting custom errors in functions and scripts:

    • Guidelines for when to use non-terminating errors vs. statement-terminating errors in the context of creating cmdlets are in MSDN topic Cmdlet Error Reporting; a pragmatic summary is in this Stack Overflow answer of mine.

    • Non-terminating errors:

    • Terminating errors:

      • Statement-terminating errors:

        • PowerShell has NO keyword- or cmdlet-based mechanism for generating a statement-terminating error.

        • The workaround - available in advanced functions only - is to use $PSCmdlet.ThrowTerminatingError():

          • Note: Despite the similarity in name with the Throw keyword (statement), this truly only generates a statement-terminating error.
          • See example below.

        • On a side note: Sometimes it is desirable to terminate upstream cmdlets, without terminating the pipeline as a whole, the way Select-Object -First <n> does, for instance. As of Windows PowerShell v5.1 / PowerShell Core v6.0.0-beta.5, there is no way to do this, but such a feature has been suggested here: Allow user code to stop a pipeline on demand / to terminate upstream cmdlets. PowerShell/PowerShell#3821

      • Script-terminating errors:

        • Use the Throw keyword.
          • Any object can be thrown, including none. In the context of a Catch handler (as part of a Try / Catch statement), not specifying an object causes the current exception to be re-thrown.
          • Unless the object thrown already is of type [System.Management.Automation.ErrorRecord], PowerShell automatically wraps the object in an instance of that type and stores the object thrown in that instance's .TargetObject property.

Example use of $PSCmdlet.WriteError() in an advanced function so as to create a non-terminating error (to work around the issue that Write-Error doesn't set$? to $False in the caller's context):

# PSv5+, using the static ::new() method to call the constructor.
# The specific error message and error category are sample values.
& { [CmdletBinding()] param()  # quick mock-up of an advanced function

  $PSCmdlet.WriteError([System.Management.Automation.ErrorRecord]::new(
      # The underlying .NET exception: if you pass a string, as here,
      #  PS creates a [System.Exception] instance.
      "Couldn't process this object.",
      $null, # error ID
      [System.Management.Automation.ErrorCategory]::InvalidData, # error category
      $null) # offending object
    )
}

# PSv4-, using New-Object:
& { [CmdletBinding()] param()  # quick mock-up of an advanced function

  $PSCmdlet.WriteError((
    New-Object System.Management.Automation.ErrorRecord "Couldn't process this object.",
      $null,
      ([System.Management.Automation.ErrorCategory]::InvalidData),
      $null
  ))

}

Example use of $PSCmdlet.ThrowTerminatingError() to create a statement-terminating error:

# PSv5+, using the static ::new() method to call the constructor.
# The specific error message and error category are sample values.
& { [CmdletBinding()] param()  # quick mock-up of an advanced function

  $PSCmdlet.ThrowTerminatingError(
    [System.Management.Automation.ErrorRecord]::new(
      # The underlying .NET exception: if you pass a string, as here,
      #  PS creates a [System.Exception] instance.
      "Something went wrong; cannot continue pipeline",
      $null, # error ID
      [System.Management.Automation.ErrorCategory]::InvalidData, # error category
      $null  # offending object
    )
  )

}

# PSv4-, using New-Object:
& { [CmdletBinding()] param()  # quick mock-up of an advanced function

  $PSCmdlet.ThrowTerminatingError((
    New-Object System.Management.Automation.ErrorRecord "Something went wrong; cannot continue pipeline",
      $null, # a custom error ID (string)
      ([System.Management.Automation.ErrorCategory]::InvalidData), # the PS error category
      $null # the target object (what object the error relates to)
  ))

}
This was referenced Aug 15, 2017
Closed
Closed
Closed
Open
@SteveL-MSFT
Copy link
Contributor

@juanpablojofre this would make a great about topic

This was referenced Aug 18, 2017
Open
Closed
Closed
Closed
@zjalexander zjalexander added the Pri2 Priority - Medium label Aug 30, 2017
@zjalexander zjalexander mentioned this issue Aug 30, 2017
Closed
5 tasks
This was referenced Sep 11, 2017
Closed
Closed
@zjalexander
Copy link
Contributor

PowerShell/PowerShell#3629

@sdwheeler sdwheeler assigned sdwheeler and unassigned juanpablojofre Nov 2, 2017
@mklement0 mklement0 mentioned this issue Dec 15, 2017
Closed
@nightroman
Copy link

A note about ThrowTerminatingError, either regression from v2 or some not clear design change.

In the following script, ThrowTerminatingError is not caught in v5 and v6. It is caught in v2.

[CmdletBinding()]
param()

$caught = 'not-caught'
try {
    # this is not caught in v5 (works in v2)
    $PSCmdlet.ThrowTerminatingError((New-Object System.Management.Automation.ErrorRecord ([Exception]'some-error'), $null, 0, $null))
}
catch {
    $caught = 'was-caught'
    throw
}
finally {
    $caught
    'in-finally'
}

Copied from https://github.com/nightroman/PowerShellTraps/tree/master/Basic/ThrowTerminatingError/Catch-is-not-called

@mklement0 mklement0 mentioned this issue Jan 5, 2018
Closed
@strawgate
Copy link

strawgate commented Jan 13, 2018 •
edited
Loading

@mklement0 I see your recommendation to utilize $PSCmdlet.WriteError() instead of write-error to fix the issue with $? but it looks like these have different behaviors when the "ErrorActionPreference" of the script/module is set to "stop".

Im testing this in Powershell 5.1

If erroractionpreference = stop and you use $PSCmdlet.WriteError() it appears to produce a locally non-terminating error but when it leaves the function it appears to be terminating. That error is not catch-able in the function producing the error but is catche-able by the caller.

If erroractionpreference = stop and you use write-error it appears to produce a statement terminating error which can be caught by the function.

$ErrorActionPreference="stop"
$VerbosePreference="continue"

function Test-WriteError {
    [CmdletBinding()]
    Param( )
    try {
        $errorRecord = New-Object Management.Automation.ErrorRecord (([Exception]'some-error'), $null, 0, $null)
        $PSCmdlet.WriteError($errorRecord) # This does not get caught when erroraction is set to stop
    } catch {
        write-verbose "Caught"
    }
}


function Test-Write-Error {
    [CmdletBinding()]
    Param( )
    try {
        write-error "Test" #This gets caught if ErrorActionPreference is Stop or -erroraction Stop is passed to this command
    } catch {
        write-verbose "Caught"
    }
}

Running Test-WriteError will terminate the script:

PS > test-writeerror; write-host "print"
some-error
At line:1 char:1
+ test-writeerror; write-host "print"
+ ~~~~~~~~~~~~~~~
    + CategoryInfo          : NotSpecified: (:) [Test-WriteError], Exception
    + FullyQualifiedErrorId : Test-WriteError

Running Test-Write-Error will not terminate the script:

PS > test-write-error; write-host "print"
VERBOSE: Caught
print

In other words, $PSCmdlet.WriteError does not appear to throw a terminating error when $ErrorActionPreference = Stop whereas write-error does

@mklement0
Copy link
Contributor Author

mklement0 commented Jan 21, 2018 •
edited
Loading

@strawgate: Thanks for pointing out that difference - I have no explanation for it.

To me, your discovery suggests that one currently should always use $PSCmdlet.WriteError() instead of Write-Error (which, needless to say, is very unfortunate):

While you could argue that technically Write-Error exhibits the correct behavior - with Stop in effect, writing a non-terminating error should throw an exception - pragmatically speaking, the $PSCmdlet.WriteError() behavior is much more sensible:

You don't want to catch your own attempts with try / catch inside an advanced function to report a non-terminating error.
Instead, you want the caller to handle this, as implied by the calling context's $ErrorActionPreference value or the -ErrorAction value passed to the advanced function (which function-internally is translated into a local-scope $ErrorActionPreference variable reflecting that value).

There is no good solution to this problem with Write-Error:

  • if you used Write-Error -ErrorAction Continue, your own catch handler wouldn't get triggered, but execution would continue, which is undesired.

  • The only solution is to reissue the Write-Error from the catch block, which is obviously cumbersome and awkward.

@alx9r
Copy link

alx9r commented Feb 3, 2018

@mklement0 Thank you very much for your excellent post. It has been invaluable to me for making sense of a number of things.

Script Terminating Errors vs Exceptions

Is there a difference between "script-terminating errors" and exceptions? There seems to be places in the PowerShell project that distinguish between "script-terminating errors" and exceptions. I haven't, however, been able to observe a difference between them. Is there a difference?

What's the "Script" in "Script-Terminating Error"?

What is "script" meant to refer to in "script-terminating errors"? It seems like "script-terminating errors" often (almost always in my use of PowerShell) do something other than terminate a script. It seems like the only case where the stack unwinding caused by a "script-terminating error" stops at something that would be called a "script" is where there happens to be nothing that catches the "script-terminating error". Am I missing something? Is there some other definition of "script" that applies to "Script-Terminating Error"?

.ThrowTerminatingError does not a Statement-Terminating Error Make

The original post includes the following statements:

Statement-terminating errors truly only terminate the statement at hand. By default, the enclosing scripts continues to run.
...
PowerShell has NO keyword- or cmdlet-based mechanism for generating a statement-terminating error.
The workaround - available in advanced functions only - is to use $PSCmdlet.ThrowTerminatingError():

Consider, however, that

function e {
    try { 1 }
    catch{ Write-Host 'catch upstream' }
}

function f {
    param ( [Parameter(ValueFromPipeline)]$x )
    process {
        # nonexistentcommand
        $PSCmdlet.ThrowTerminatingError(
            [System.Management.Automation.ErrorRecord]::new(
                'exception message',
                'errorId',
                [System.Management.Automation.ErrorCategory]::InvalidOperation,
                $null
            )
        )
    }
}

try
{
    e | f
    Write-Host 'statement following statement with .ThrowTerminatingError'
}
catch
{
    Write-Host 'outer catch'
}

outputs outer catch. If it were true that "statement-terminating errors truly only terminate the statement at hand" and .ThrowTerminatingError() caused a statement-terminating error, it seems to me that the output would have been statement following statement with .ThrowTerminatingError.

@alx9r alx9r mentioned this issue Feb 3, 2018
Closed
@mklement0
Copy link
Contributor Author

mklement0 commented Feb 7, 2018 •
edited
Loading

@alx9r: Thanks for the nice feedback and digging deeper. I'll need more time to give your post the attention it deserves, but let me say up front that it was I who came up with the term "script-terminating", out of necessity, given that the existing docs made no distinction between the sub-types of terminating errors.

Thus, there's nothing authoritative about this term, and if it turns out to be a misnomer, we should change it.

@sdwheeler sdwheeler removed their assignment Feb 14, 2018
@mklement0
Copy link
Contributor Author

mklement0 commented Feb 17, 2018 •
edited
Loading

@alx9r:

Disclaimer: Little of what I state below has been arrived at through source-code analysis. All readers are welcome to point out any misinformation.

Is there a difference between "script-terminating errors" and exceptions?

From what I understand, all PowerShell errors (as opposed to stderr output from external utilities) are exceptions under the hood.
A script-terminating error is simply an exception that isn't caught by PowerShell.

What's the "Script" in "Script-Terminating Error"?

Presumably, the more technically accurate term would be runspace-terminating error, but the term runspace is not a term a PowerShell(-only) user is necessarily expected to be familiar with. [Update: It's hard to come up with a succinct, technically accurate term, though calling such errors fatal is a pragmatic alternative - for the technical underpinnings, see PowerShell/PowerShell#14819 (comment)]

In practice, what I call a script-terminating error, when uncaught:

  • aborts a running script, including its callers.

  • aborts a runspace created with [powershell]::Create().

  • aborts a single command line submitted at the command prompt - even if composed of multiple,
    ;-separated statements. (Presumably, such a command line runs in a runspace created ad hoc).

Again, the most descriptive term is open to debate.
Session-terminating error is perhaps an alternative, but that also could be confusing with respect to behavior at the command prompt.

If it were true that "statement-terminating errors truly only terminate the statement at hand" and .ThrowTerminatingError() caused a statement-terminating error, it seems to me that the output would have been statement following statement with .ThrowTerminatingError.

In this context, the pipeline as a whole is the statement, even though a pipeline is itself by definition composed of sub-statements (an expression or command as the 1st segment, and commands as the subsequent segments).

Again, the terminology is open to debate: If an actual pipeline is involved, you could argue that pipeline-terminating error is the better choice, but note that that doesn't apply to expression-only statements such as 1 / 0 that also generate a statement-level-only error - it is for that reason that I stuck with statement as the qualifier.

Another example to demonstrate the pipeline behavior of a statement-terminating error:

# Advanced function that generates a statement-(pipeline-)terminating error
# when the input object is integer 2
function f {
  param ([Parameter(ValueFromPipeline)] [int] $i)
  begin { write-host 'before2' }
  process {
    if ($i -eq 2) {
      $PSCmdlet.ThrowTerminatingError(
        [System.Management.Automation.ErrorRecord]::new(
            'exception message',
            'errorId',
            [System.Management.Automation.ErrorCategory]::InvalidOperation,
            $null
        )
      )
    }
    $i
  }
  end { write-host 'after2' }
}

# Send an 3-element array through the pipeline whose 2nd element triggers the statement-terminating error.
1, 2, 3 | % { write-host 'before1' } { $_ } { write-host 'after1' } | f

The above yields:

before1
before2
1
f : exception message
...

That is, the entire pipeline was instantly terminated when the statement-terminating error was generated (no further input objects were processed, and the end blocks didn't run).

@alx9r
Copy link

alx9r commented Feb 25, 2018 •
edited
Loading

Since I came across this post months ago I've been trying to arrive at empirical proof of the various statements in the OP about "terminating", "non-terminating", "script-terminating", and "statement-terminating" errors. Currently I am doubtful that there exists definitions of those terms that would result in a useful taxonomy of PowerShell errors.

I have found few useful generalizations about PowerShell errors that stand up to testing. Most hypotheses involving generalizations and nuance (including some in OP) can be disproven with a small amount of testing. The truth about PowerShell errors seems remarkably resistant to simplicity.

I am fairly confident, however, that the following generalizations about PowerShell errors are true:

  1. A PowerShell expression or statement might do all, some subset, or none of the following error-related actions:
    a. write records to the $global:Error variable
    b. write records to a non-global copy of $Error
    c. write records to a variable named in an -ErrorVariable argument
    d. output to PowerShell's error stream
    e. update the value of $?
    f. terminate a pipeline
    g. terminate a statement
    h. throw an exception
  2. Whether each of the actions in (1) occurs depends at least on one or more of the following:
    a. the value of $ErrorActionPreference
    b. the value passed to -ErrorAction or the value of $ErrorAction
    c. the presence of an enclosing try{} block (eg. How, exactly, does PowerShell decide whether .ThrowTerminatingError() terminates only the statement? PowerShell/PowerShell#6098)
    d. the precise way an error is reported (eg. Write-Error vs. $PSCmdlet.WriteError() vs. throw $PSCmdlet.ThrowTerminatingError())

More nuanced generalizations would, of course, be useful. Currently, however, I am skeptical that such generalizations can be found and proven. Indeed my attempts at doing so have seemed sisyphean.

From my perspective the useful takeaway from this exercise is currently the following:

  • there is little that can be known without testing about which error actions any given PowerShell expression or statement takes
  • test plans should be informed by the possible actions in (1) and predicate conditions in (2)

@mklement0
Copy link
Contributor Author

mklement0 commented Feb 25, 2018 •
edited
Loading

@alx9r:

My writing the OP had multiple intents:

  • I wanted to point that the current documentation is hopelessly inadequate.

  • I wanted to provide a pragmatic summary of the de-facto behavior, so as to help us poor souls who have to live with and make some sense of the current behavior.

    • In the course of doing so, I coined terms that made sense to me where none were to be found in the docs.

    • While I don't doubt that my summary doesn't cover all nuances - it was arrived at empirically as well, not through studying source code - my hope was that it covers most aspects and gets at least the fundamentals right.

    • Believe me, I understand that you ran out of steam trying while investigating, but if you have pointers as to where my summary is fundamentally off, please share them.

  • I wanted to start a conversion about cleaning up the byzantine mess that PowerShell error handling currently is - save for linking to various issues highlighting specific problematic behaviors, I didn't do a good job of making that explicit, however.

    • Sadly, with the release of v6 - which had many breaking changes - an opportunity was missed to radically simplify PowerShell's error handling in favor of something that is consistent and fit's into a user's head.

@alx9r
Copy link

alx9r commented Feb 25, 2018

@mklement0 I think we are on the same page in striving for a better record of PowerShell error behavior.

I wanted to provide a pragmatic summary of the de-facto behavior, so as to help us poor souls who have to live with and make some sense of the current behavior.

As far as I'm concerned your OP succeeds at this. And I am very thankful that you wrote it.

While I don't doubt that my summary doesn't cover all nuances - it was arrived at empirically as well, not through studying source code - my hope was that it covers most aspects and gets at least the fundamentals right.

I think your OP is probably as close to "fundamentally" correct as I've seen. The problem is that there are so many exceptions and nuances to the "fundamental" behavior that "knowing" the "fundamentals" is not particularly useful.

...if you have pointers as to where my summary is fundamentally off, please share them.

The truth of PowerShell errors is sufficiently messy that I am not convinced that there are objective fundamentals that your summary could be "off" from.

I think it is probably useful for me to continue to report repros of newly-surprising error handling behavior. I'm not sure what the best venue for that is, but when I encounter behavior that is sufficiently inscrutable, I expect to continue reporting it to the PowerShell/PowerShell repository.

I wanted to start a conversion about cleaning up the byzantine mess that PowerShell error handling currently is..

It seems to me that making PowerShell errors less byzantine would involve numerous breaking changes. I don't think that would be a good way forward. It seems like we are mostly stuck with the current PowerShell error behavior.

@vexx32
Copy link
Contributor

vexx32 commented Jun 17, 2020

@Kriegel if you want to change that, Write-Error is not the best tool for it.

Instead, mark the function with [CmdletBinding()] like so:

function Get-Err {
    [CmdletBinding()]
    param()

    # function code
}

And from there, you can use $PSCmdlet.WriteError($errorRecord) and it will automatically override the invocationinfo to point at the function which called WriteError() rather than the original error source. 🙂

@Kriegel
Copy link

Kriegel commented Jun 18, 2020

@vexx32
I like that $PSCmdlet.WriteError($errorRecord) automatically overrides the invocationinfo.
I some one dislike it, then we are here: PowerShell/PowerShell#10733 (LOL)

@chasewilson chasewilson self-assigned this Aug 3, 2020
jberezanski added a commit to jberezanski/ChocolateyPackages that referenced this issue Oct 28, 2020
@jberezanski
Update-VisualStudio2017Packages: Fix Invoke-WebRequest error handling…
6b52fe2 
… to support PS Core

Background:
PowerShell/PowerShell#4534
MicrosoftDocs/PowerShell-Docs#1583
@mklement0 mklement0 mentioned this issue Jan 11, 2021
Closed
@mklement0 mklement0 mentioned this issue Feb 17, 2021
Closed
@Silex Silex mentioned this issue Mar 15, 2021
Closed
@chasewilson chasewilson removed their assignment Jun 21, 2021
@Kalmalyzer Kalmalyzer mentioned this issue Jul 12, 2021
Closed
@johndog
Copy link
Contributor

johndog commented Jul 23, 2021

Strict modes don't seem to have been discussed, but they introduce yet another nuance into the domain.

Here's one dichotomy...

First, context:

> $ErrorActionPreference = "Continue"
> Set-StrictMode -Version Latest

Then this shows that using an undefined variable generates an error but doesn't stop execution:

> & {$nonExistantVariable;"got here"}

InvalidOperation: The variable '$nonExistantVariable' cannot be retrieved because it has not been set.
got here

But the same script block run with Invoke() generates an exception, not just an error:

> {$nonExistantVariable;"got here"}.Invoke()

MethodInvocationException: Exception calling "Invoke" with "0" argument(s): "The variable '$nonExistantVariable' cannot be retrieved because it has not been set."

It would have been much nicer of course if strict mode always produced an exception, regardless of ErrorActionPreference.

@johndog johndog mentioned this issue Jul 23, 2021
Open
@mklement0
Copy link
Contributor Author

@johndog, good idea to cover strict-mode violations too; I've just updated the initial post.

The short of it is that strict-mode violations result in statement-terminating errors by default.

This applies to both your examples, which only differ in the scope of what constitutes the statement being terminated:

  • In & {$nonExistentVariable;"got here"}, $nonExistentVariable is a statement by itself, therefore "got here" still executes (with Continue or SilentlyContinue in effect).

  • In {$nonExistentVariable;"got here"}.Invoke() the scope of the statement is the method call (as an aside: from PowerShell code, there's usually no reason to call .Invoke()), so that & { {$nonExistentVariable;"won't get here"}.Invoke(); "got here" } would again still execute "got here".

    • Note: When using .Invoke(), both flavors of terminating errors inside the script block cause the method to fail with a statement-terminating error - even if the triggering error was a script-terminating one; try & { {throw 'a fit'; "won't get here"}.Invoke(); "got here" }

Note that while PowerShell's error handling is built on .NET exceptions (in the case of the two flavors of terminating errors), it's better to discuss it in terms of non-terminating vs. statement-terminating vs. script-terminating errors.

@gerardog gerardog mentioned this issue Jan 31, 2022
Closed
@michaeltlombardi michaeltlombardi mentioned this issue Mar 9, 2022
Closed
2 tasks
@MarkCallow MarkCallow mentioned this issue Apr 11, 2022
Closed
@mklement0 mklement0 mentioned this issue Oct 22, 2022
Closed
5 tasks
@sdwheeler sdwheeler removed the Pri2 Priority - Medium label Dec 16, 2022
@mklement0 mklement0 mentioned this issue Apr 13, 2023
Closed
5 tasks
@mklement0 mklement0 mentioned this issue May 2, 2023
Closed
5 tasks
@HowardWolosky HowardWolosky mentioned this issue Jun 28, 2023
Open
@jantari jantari mentioned this issue Jul 8, 2023
Closed
@davidtrevor23
Copy link

Trying to wrap my head around pipeline execution and statement-terminating behavior. From all the posts and comments, I gathered the understanding that the following two methods can both be used to provoke a statement-terminating behavior.

  • expression runtime errors; e.g. 1 / 0
  • $PSCmdlet.ThrowTerminatingError()

But in my tests they do not behave the same. The first example continues the pipeline, whereas the second terminates the pipeline. Can someone explain where my understanding is wrong?

function calc {
    param( [Parameter(ValueFromPipeline=$true)]$i )
    PROCESS { 1 / $i }
}
-1,0,1 | calc | Write-Output

-1
RuntimeException: Attempted to divide by zero.
1

function calc {
    param( [Parameter(ValueFromPipeline=$true)]$i )
    PROCESS { if ($i -ne 0) { 1 / $i } else { $PSCmdlet.ThrowTerminatingError([System.Management.Automation.ErrorRecord]::new('exception message','errorId',[System.Management.Automation.ErrorCategory]::InvalidOperation,$null)) } }
}
-1,0,1 | calc | Write-Output

-1
RuntimeException: Attempted to divide by zero.

On another note, can someone point me to a resource that explains how exactly objects get processed when passed through the pipeline operator (in what order). I do not quite understand the order of operation in this comment:
#1583 (comment)

@mklement0
Copy link
Contributor Author

mklement0 commented Oct 12, 2023 •
edited
Loading

  • An exception that occurs in a stand-alone expression (i.e. an expression that acts as its own statement) terminates only that expression itself.
  • $PSCmdlet.ThrowTerminatingError() by design terminates the enclosing advanced function/script.

Thus,1 / 0 in PROCESS { 1 / $i } terminates just 1 / $i and continues execution (even inside the script block; placing something after would still execute, e.g. PROCESS { 1 / $i ; 'after' }

A better comparison of the two scenarios would be:

# First statement on each line prints error only; 'after' still prints.
1, (1 / 0),    2 | Write-Output; 'after'
1, (0 | calc), 2 | Write-Output; 'after'

Now both errors terminate the entire statement (which happens to be a pipeline in this example, but it equally applies to an expression) and there is no success output. However the next statement ('after') executes.

Perhaps the following addresses your second question:

Note that something like 1, (1 / 0), 2 is an array-construction expression, which means that all elements are evaluated up front. The same applies to a pipeline or expression enclosed in (...)
Thus, the statement-terminating error occurs right away and terminates the entire pipeline.

By contrast, if you use $(...), @(...) or & { ... } with separate statements (separated with ; or just newlines), it is again just the statement that triggers the error alone that is terminated:

# First statement on each line prints 1, then error, then 2; 'after' still prints.
& { 1; (1 / 0); 2 }    | Write-Output; 'after'
& { 1; (0 | calc); 2 } | Write-Output; 'after'

1 and 2 get to print, because they're separate statements; (1 / 0) and (0 | calc) only terminate themselves.
Note: With $(...) and @(...) the error message prints first, because these two operators collect all success output first, whereas & { ... } streams its output.

@sdwheeler sdwheeler mentioned this issue Dec 20, 2023
Closed
gstreamer-github pushed a commit to sdroege/gstreamer-rs that referenced this issue Dec 21, 2023
@nirbheek @sdroege
ci: Fix pwsh scripts not exiting on error
52c764b 
We all know that external utilities returning a non-zero exit code do
not terminate a powershell script. However, most do not know (and
neither did I) that it is impossible to promote error exit codes to
script-terminating errors with ErrorActionPreference.

Explicitly check the return codes and Exit.

MicrosoftDocs/PowerShell-Docs#1583

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer-rs/-/merge_requests/1368>
@mklement0 mklement0 mentioned this issue Mar 14, 2024
Closed
5 tasks
@jazzdelightsme jazzdelightsme mentioned this issue May 26, 2024
Open
@alpetric alpetric mentioned this issue Oct 8, 2024
Merged
@denis-komarov
Copy link

@mklement0
Thank you for such a fundamental and in-depth description of error handling mechanisms. Just for the sake of completeness, it seems that another separate class of errors could be mentioned. This is "Fatal error. Internal CLR error" against which even try catch is powerless. However, I am not sure of the practical value of this addition.

@alx9r
Copy link

alx9r commented Nov 6, 2024

Origin Location of Errors from Explicitly-Invoked Scriptblocks

The resulting error objects that arise in a PowerShell catch{} block contain location information for the actual source of the error that is, at best, elusive. The contents of the error object are affected by at least the following:

  • How the error is created: Write-Error, throw, or some lower-level exception like 1/0
  • The calling method: the dot-source operator ., the call operator &, ForEach-Object, Scriptblock.Invoke

Consider files containing scriptblocks according to the following table:

Script File Name Scriptblock
divide by zero.ps1 { 1/0 }
error_action_stop.ps1 {$PSCommandPath | Get-Item | % BaseName | Write-Error -ErrorAction Stop}
throw.ps1 {throw $($PSCommandPath | Get-Item | % BaseName)}

By invoking each of those scriptblocks using one of the calling methods inside a try{}, the resultant error object can be caught and examined. The object tree of each such an error object can be traversed to find any mentions of the file where the scriptblock is defined. The paths of possible mentions in these object trees is numerous, and the prevalence of most such mentions across the matrix of call methods and error causes is sparse. There are, however, four paths that together contain mention of the originating scriptblock for all of these combinations of error and calling method. Those paths are summarized in the following table:

Mentioned In
Error Cause Method $.
ScriptStackTrace		 $.
InvocationInfo.
ScriptName		 $.
Exception.
InnerException.
ErrorRecord.
ScriptStackTrace		 $.
Exception.
ErrorRecord.
InvocationInfo.
ScriptName
divide by zero call_operator ✓ ✓ ✓
divide by zero dot_source_operator ✓ ✓ ✓
divide by zero ForEach-Object ✓ ✓
error_action_stop call_operator ✓
error_action_stop dot_source_operator ✓
error_action_stop ForEach-Object ✓
throw call_operator ✓ ✓
throw dot_source_operator ✓ ✓
throw ForEach-Object ✓ ✓
divide by zero Scriptblock.Invoke ✓
error_action_stop Scriptblock.Invoke ✓
throw Scriptblock.Invoke ✓

This suggests the following:

  • Errors resulting from any code downstream of ForEach-Object in particular will either correctly show the file where the error arose or not depending the cause of the error and whether only $.InvocationInfo.ScriptName is shown. (PowerShell 7 shows the correct location when using ForEach-Object, whereas Windows PowerShell does not, for example.)
  • The PowerShell 7 error view seems to be getting position information for the ForEach-Object error_action_stop scenario from somewhere other than what is contained by the caught error object. The method I'm using to traverse the error object only finds mention of the correct file in $.ScriptStackTrace.
  • The original location of errors can, in principle, be surfaced for all of the above combinations of calling method and error cause.
  • Simple rules can select the correct location from the paths.
  • $ErrorView='NormalView' shows the location of the call site not the offending scriptblock for error_action_stop called with . or & but shows does show the offending scriptblock using for error_action_stop with ForEach-Object. $.ScriptStackTrace probably should be shown for . and & for error_action_stop in $ErrorView='NormalView' instead. The same seems to apply to all errors arising from Scriptblock.Invoke() regardless of their cause.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
area-error-handling Area - Error handling
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests