terraform/tfdiags/diagnostics.go

331 lines
10 KiB
Go
Raw Normal View History

tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
package tfdiags
import (
"bytes"
"fmt"
"path/filepath"
"sort"
"strings"
tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
"github.com/hashicorp/errwrap"
multierror "github.com/hashicorp/go-multierror"
"github.com/hashicorp/hcl2/hcl"
)
// Diagnostics is a list of diagnostics. Diagnostics is intended to be used
// where a Go "error" might normally be used, allowing richer information
// to be conveyed (more context, support for warnings).
//
// A nil Diagnostics is a valid, empty diagnostics list, thus allowing
// heap allocation to be avoided in the common case where there are no
// diagnostics to report at all.
type Diagnostics []Diagnostic
// Append is the main interface for constructing Diagnostics lists, taking
// an existing list (which may be nil) and appending the new objects to it
// after normalizing them to be implementations of Diagnostic.
//
// The usual pattern for a function that natively "speaks" diagnostics is:
//
// // Create a nil Diagnostics at the start of the function
// var diags diag.Diagnostics
//
// // At later points, build on it if errors / warnings occur:
// foo, err := DoSomethingRisky()
// if err != nil {
// diags = diags.Append(err)
// }
//
// // Eventually return the result and diagnostics in place of error
// return result, diags
//
// Append accepts a variety of different diagnostic-like types, including
// native Go errors and HCL diagnostics. It also knows how to unwrap
// a multierror.Error into separate error diagnostics. It can be passed
// another Diagnostics to concatenate the two lists. If given something
// it cannot handle, this function will panic.
func (diags Diagnostics) Append(new ...interface{}) Diagnostics {
for _, item := range new {
if item == nil {
continue
}
switch ti := item.(type) {
case Diagnostic:
diags = append(diags, ti)
case Diagnostics:
diags = append(diags, ti...) // flatten
case diagnosticsAsError:
diags = diags.Append(ti.Diagnostics) // unwrap
case NonFatalError:
diags = diags.Append(ti.Diagnostics) // unwrap
tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
case hcl.Diagnostics:
for _, hclDiag := range ti {
diags = append(diags, hclDiagnostic{hclDiag})
}
case *hcl.Diagnostic:
diags = append(diags, hclDiagnostic{ti})
case *multierror.Error:
for _, err := range ti.Errors {
diags = append(diags, nativeError{err})
}
case error:
switch {
case errwrap.ContainsType(ti, Diagnostics(nil)):
// If we have an errwrap wrapper with a Diagnostics hiding
// inside then we'll unpick it here to get access to the
// individual diagnostics.
diags = diags.Append(errwrap.GetType(ti, Diagnostics(nil)))
case errwrap.ContainsType(ti, hcl.Diagnostics(nil)):
// Likewise, if we have HCL diagnostics we'll unpick that too.
diags = diags.Append(errwrap.GetType(ti, hcl.Diagnostics(nil)))
default:
diags = append(diags, nativeError{ti})
}
default:
panic(fmt.Errorf("can't construct diagnostic(s) from %T", item))
}
}
// Given the above, we should never end up with a non-nil empty slice
// here, but we'll make sure of that so callers can rely on empty == nil
if len(diags) == 0 {
return nil
}
return diags
}
// HasErrors returns true if any of the diagnostics in the list have
// a severity of Error.
func (diags Diagnostics) HasErrors() bool {
for _, diag := range diags {
if diag.Severity() == Error {
return true
}
}
return false
}
// ForRPC returns a version of the receiver that has been simplified so that
// it is friendly to RPC protocols.
//
// Currently this means that it can be serialized with encoding/gob and
// subsequently re-inflated. It may later grow to include other serialization
// formats.
//
// Note that this loses information about the original objects used to
// construct the diagnostics, so e.g. the errwrap API will not work as
// expected on an error-wrapped Diagnostics that came from ForRPC.
func (diags Diagnostics) ForRPC() Diagnostics {
ret := make(Diagnostics, len(diags))
for i := range diags {
ret[i] = makeRPCFriendlyDiag(diags[i])
}
return ret
}
tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
// Err flattens a diagnostics list into a single Go error, or to nil
// if the diagnostics list does not include any error-level diagnostics.
//
// This can be used to smuggle diagnostics through an API that deals in
// native errors, but unfortunately it will lose naked warnings (warnings
// that aren't accompanied by at least one error) since such APIs have no
// mechanism through which to report these.
//
// return result, diags.Error()
func (diags Diagnostics) Err() error {
if !diags.HasErrors() {
return nil
}
return diagnosticsAsError{diags}
}
// ErrWithWarnings is similar to Err except that it will also return a non-nil
// error if the receiver contains only warnings.
//
// In the warnings-only situation, the result is guaranteed to be of dynamic
// type NonFatalError, allowing diagnostics-aware callers to type-assert
// and unwrap it, treating it as non-fatal.
//
// This should be used only in contexts where the caller is able to recognize
// and handle NonFatalError. For normal callers that expect a lack of errors
// to be signaled by nil, use just Diagnostics.Err.
func (diags Diagnostics) ErrWithWarnings() error {
if len(diags) == 0 {
return nil
}
if diags.HasErrors() {
return diags.Err()
}
return NonFatalError{diags}
}
// NonFatalErr is similar to Err except that it always returns either nil
// (if there are no diagnostics at all) or NonFatalError.
//
// This allows diagnostics to be returned over an error return channel while
// being explicit that the diagnostics should not halt processing.
//
// This should be used only in contexts where the caller is able to recognize
// and handle NonFatalError. For normal callers that expect a lack of errors
// to be signaled by nil, use just Diagnostics.Err.
func (diags Diagnostics) NonFatalErr() error {
if len(diags) == 0 {
return nil
}
return NonFatalError{diags}
}
// Sort applies an ordering to the diagnostics in the receiver in-place.
//
// The ordering is: warnings before errors, sourceless before sourced,
// short source paths before long source paths, and then ordering by
// position within each file.
//
// Diagnostics that do not differ by any of these sortable characteristics
// will remain in the same relative order after this method returns.
func (diags Diagnostics) Sort() {
sort.Stable(sortDiagnostics(diags))
}
tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
type diagnosticsAsError struct {
Diagnostics
}
func (dae diagnosticsAsError) Error() string {
diags := dae.Diagnostics
switch {
case len(diags) == 0:
// should never happen, since we don't create this wrapper if
// there are no diagnostics in the list.
return "no errors"
case len(diags) == 1:
desc := diags[0].Description()
if desc.Detail == "" {
return desc.Summary
}
return fmt.Sprintf("%s: %s", desc.Summary, desc.Detail)
tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
default:
var ret bytes.Buffer
fmt.Fprintf(&ret, "%d problems:\n", len(diags))
for _, diag := range dae.Diagnostics {
desc := diag.Description()
if desc.Detail == "" {
fmt.Fprintf(&ret, "\n- %s", desc.Summary)
} else {
fmt.Fprintf(&ret, "\n- %s: %s", desc.Summary, desc.Detail)
}
tfdiags: new package for normalizing error and warning messages Currently we lean heavily on the Go error type as our primary means of describing errors, and along with that use several more specialized implementations of it in different spots for additional capabilities such as multiple errors in one object, source code range references, etc. We also have a rather ad-hoc approach of returning an array of warnings from certain functions along with one or multiple errors. This rather-disorganized approach makes it hard for us to present user-facing error messages consistently. As a step towards mitigating this, package tfdiags provides a model for user-facing error and warning messages and helper functions for creating them from various other error and warning types used elsewhere in Terraform. This mechanism is intended to be used to report errors and warnings where the audience is the Terraform user, and so it may go a few layers deep down the call stack into codepaths like config parsing, interpolation, etc but is primarily a UX concern. The deepest reaches of Terraform core will continue using "error" as normal, with higher layers preparing error messages for presentation to the user. To avoid needing to change the interface of every function that might generate error diagnostics, the Diagnostics type can be "smuggled" via an error value through other APIs and then unwrapped at the other end, though it will lose any naked warnings (without at least one error) along the way, and so codepaths that are expected to generate warnings (validation, primarily) should use the concrete Diagnostics type throughout the call chain.
2017-10-05 02:13:29 +02:00
}
return ret.String()
}
}
// WrappedErrors is an implementation of errwrap.Wrapper so that an error-wrapped
// diagnostics object can be picked apart by errwrap-aware code.
func (dae diagnosticsAsError) WrappedErrors() []error {
var errs []error
for _, diag := range dae.Diagnostics {
if wrapper, isErr := diag.(nativeError); isErr {
errs = append(errs, wrapper.err)
}
}
return errs
}
// NonFatalError is a special error type, returned by
// Diagnostics.ErrWithWarnings and Diagnostics.NonFatalErr,
// that indicates that the wrapped diagnostics should be treated as non-fatal.
// Callers can conditionally type-assert an error to this type in order to
// detect the non-fatal scenario and handle it in a different way.
type NonFatalError struct {
Diagnostics
}
func (woe NonFatalError) Error() string {
diags := woe.Diagnostics
switch {
case len(diags) == 0:
// should never happen, since we don't create this wrapper if
// there are no diagnostics in the list.
return "no errors or warnings"
case len(diags) == 1:
desc := diags[0].Description()
if desc.Detail == "" {
return desc.Summary
}
return fmt.Sprintf("%s: %s", desc.Summary, desc.Detail)
default:
var ret bytes.Buffer
if diags.HasErrors() {
fmt.Fprintf(&ret, "%d problems:\n", len(diags))
} else {
fmt.Fprintf(&ret, "%d warnings:\n", len(diags))
}
for _, diag := range woe.Diagnostics {
desc := diag.Description()
if desc.Detail == "" {
fmt.Fprintf(&ret, "\n- %s", desc.Summary)
} else {
fmt.Fprintf(&ret, "\n- %s: %s", desc.Summary, desc.Detail)
}
}
return ret.String()
}
}
// sortDiagnostics is an implementation of sort.Interface
type sortDiagnostics []Diagnostic
var _ sort.Interface = sortDiagnostics(nil)
func (sd sortDiagnostics) Len() int {
return len(sd)
}
func (sd sortDiagnostics) Less(i, j int) bool {
iD, jD := sd[i], sd[j]
iSev, jSev := iD.Severity(), jD.Severity()
iSrc, jSrc := iD.Source(), jD.Source()
switch {
case iSev != jSev:
return iSev == Warning
case (iSrc.Subject == nil) != (jSrc.Subject == nil):
return iSrc.Subject == nil
case iSrc.Subject != nil && *iSrc.Subject != *jSrc.Subject:
iSubj := iSrc.Subject
jSubj := jSrc.Subject
switch {
case iSubj.Filename != jSubj.Filename:
// Path with fewer segments goes first if they are different lengths
sep := string(filepath.Separator)
iCount := strings.Count(iSubj.Filename, sep)
jCount := strings.Count(jSubj.Filename, sep)
if iCount != jCount {
return iCount < jCount
}
return iSubj.Filename < jSubj.Filename
case iSubj.Start.Byte != jSubj.Start.Byte:
return iSubj.Start.Byte < jSubj.Start.Byte
case iSubj.End.Byte != jSubj.End.Byte:
return iSubj.End.Byte < jSubj.End.Byte
}
fallthrough
default:
// The remaining properties do not have a defined ordering, so
// we'll leave it unspecified. Since we use sort.Stable in
// the caller of this, the ordering of remaining items will
// be preserved.
return false
}
}
func (sd sortDiagnostics) Swap(i, j int) {
sd[i], sd[j] = sd[j], sd[i]
}