error
Wrap errors with more context.
Inspiration
This module is inspired by the go error libraries that have simple
functions for creating & wrapping errors.
This is based on libraries like eris & pkg/errors
Older version of error
If you are looking for the older v7 version of error you should
check v7.x branch
Motivation
Wrapping errors when bubbling up instead of just doing
if (err) return cb(err)
allows you to pass more context
up the stack.
Common example include passing along parameters from the DB
read related to the failure or passing along any context
from the user in a HTTP request when doing a failure.
This can give you nice to read messages that include more
information about the failure as it bubbles up.
There is more information about how to handle errors in this
article Don't just check errors, handle them gracefully
examples:
const { wrapf } = require('error')
function authenticatRequest(req) {
authenticate(req.user, (err) => {
if (err) {
return cb(wrapf('authenticate failed', err))
}
cb(null)
})
}
or
const { wrapf } = require('error')
function readFile(path, cb) {
fs.open(path, 'r', (err, fd) => {
if (err) {
return cb(wrapf('open failed', err, { path }))
}
const buf = Buffer.alloc(64 * 1024)
fs.read(fd, buf, 0, buf.length, 0, (err) => {
if (err) {
return cb(wrapf('read failed', err, { path }))
}
fs.close(fd, (err) => {
if (err) {
return cb(wrapf('close failed', err, { path }))
}
cb(null, buf)
})
})
})
}
Structured errors
const { SError } = require('error')
class ServerError extends SError {}
class ClientError extends SError {}
const err = ServerError.create(
'{title} server error, status={statusCode}', {
title: 'some title',
statusCode: 500
}
)
const err2 = ClientError.create(
'{title} client error, status={statusCode}', {
title: 'some title',
statusCode: 404
}
)
Wrapped Errors
const net = require('net');
const { WError } = require('error')
class ServerListenError extends WError {}
var server = net.createServer();
server.on('error', function onError(err) {
if (err.code === 'EADDRINUSE') {
throw ServerListenFailedError.wrap(
'error in server, on port={requestPort}', err, {
requestPort: 3000,
host: null
}
)
} else {
throw err;
}
});
server.listen(3000);
Comparison to Alternatives.
There are alternative existing libraries for creating typed
and wrapped errors on npm. Here's a quick comparison to some
alternatives.
This module takes inspiration from verror
and adds improvements.
- You can pass extra fields as meta data on the error
- The templating forces dynamic strings to be extra fields.
- Uses ES6 classes for inheritance. This gives your errors unique
class names and makes them show up in heapdumps.
- Has JSON.stringify support
This package used to have a completely different API on the
7.x branch.
- New
error
module uses actual classes instead of dynamically
monkey patching fields onto new Error()
- Implementation is more static, previous code was very dynamic
- Simpler API, see the message & properties in one place.
wrapf
& errorf
helpers for less boilerplate.
Hand writing Error
sub classes.
You can create your own Error classes by hand. This tends to lead
to 10-20 lines of boilerplate per error which is replace with
one line by using the error
module; aka
class AccountsServerFailureError extends SError {}
class ConnectionResetError extends WError {}
The ono
package has similar functionality with a different API
ono
encourages plain errors instead of custom errors by defaulterror
has zero dependencieserror
is only one simple file. ono
is 10.error
implementation is more static, ono is very dynamic.
Documentation
This package implements three classes, WError
; SError
&
MultiError
You are expected to subclass either WError
or SError
;
SError
stands for Structured Error
; it's an error base
class for adding informational fields to your error beyond
just having a message.WError
stands for Wrapped Error
; it's an error base
class for when you are wrapping an existing error with more
information.
The MultiError
class exists to store an array of errors but
still return a single Error
; This is useful if your doing
a parallel operation and you want to wait for them all to finish
and do something with all of the failures.
Some utility functions are also exported:
findCauseByName
; See if error or any of it's causes is of
the type name.fullStack
; Take a wrapped error and compute a full stack.wrapf
; Utility function to quickly wraperrorf
; Utility function to quickly create an error
WError
Example:
class ServerListenError extends WError {}
ServerListenError.wrap('error in server', err, {
port: 3000
})
When using the WError
class it's recommended to always call
the static wrap()
method instead of calling the constructor
directly.
Example (without cause message):
class ApplicationStartupError extends WError {}
ApplicationStartupError.wrap(
'Could not start the application cleanly: {reason}',
err,
{
skipCauseMessage: true,
reason: 'Failed to read from disk'
}
)
Setting skipCauseMessage: true
will not append the cause
error message but still make the cause object available.
const werr = new WError(message, cause, info)
Internal constructor, should pass a message
string, a cause
error and a info
object (or null
).
WError.wrap(msgTmpl, cause, info)
wrap()
method to create error instances. This applies the
string-template
templating to msgTmpl
with info
as a parameter.
The cause
parameter must be an error
The info
parameter is an object or null
.
The info
parameter can contain the field skipCauseMessage: true
which will make WError
not append : ${causeMessage}
to the
message of the error.
werr.type
The type
field is the machine readable type for this error.
Always use err.type
and never err.message
when trying to
determine what kind of error it is.
The type
field is unlikely to change but the message
field
can change.
werr.fullType()
Calling fullType
will compute a full type for this error and
any causes that it wraps. This gives you a long type
string
that's a concat for every wrapped cause.
werr.cause()
Returns the cause
error.
werr.info()
Returns the info
object passed on. This is merged with the
info of all cause
errors up the chain.
werr.toJSON()
The WError
class implements toJSON()
so that the JSON
serialization makes sense.
WError.fullStack(err)
This returns a full stack; which is a concatenation of this
stack trace and the stack trace of all causes in the cause chain
WError.findCauseByName(err, name)
Given an err and a name will find if the err or any causes
implement the type of that name.
This allows you to check if a wrapped ApplicationError
has
for example a LevelReadError
or LevelWriteError
in it's cause
chain and handle database errors differently from all other app
errors.
SError
Example:
class LevelReadError extends SError {}
LevelReadError.create('Could not read key: {key}', {
key: '/some/key'
})
When using the SError
class it's recommended to always call
the static create()
method instead of calling the constructor
directly.
const serr = new SError(message, info)
Internal constructor that takes a message string & an info object.
SError.create(messageTmpl, info)
The main way to create error objects, takes a message template
and an info object.
It will use string-template to apply the
template with the info
object as a parameter.
serr.type
Returns the type field. The err.type
field is machine readable.
Always use err.type
& not err.message
when trying to compare
errors or do any introspection.
The type
field is unlikely to change but the message
field
can change.
serr.info()
Returns the info object for this error.
serr.toJSON()
This class can JSON serialize cleanly.
MultiError
Example:
class FanoutError extends MultiError {}
function doStuff (filePath, cb) {
fanoutDiskReads(filePath, (errors, fileContents) => {
if (errors && errors.length > 0) {
const err = FanoutError.errorFromList(errors)
return cb(err)
}
})
}
When using the MultiError
class it's recommended to always
call the static errorFromList
method instead of calling the
constructor directly.
Installation
npm install error
Contributors
MIT Licenced