A cute, immutable, and unusually type safe logger for TypeScript & ES6+, NodeJS & Web with some unique features.
npm i -S daslog
// src/log.ts
import {logger} from 'daslog';
const log = logger().setCategory('Example');
// src/some/file.ts
import log from 'log'; // assuming baseUrl: "src"
log.info('Initializing app'); // [ 2019-10-03 02:42:19 | INFO | Example ] Initializing app
Other loggers I tried:
- were intended for node.js and used runtime require statements, which did not play well with bundlers
- did not maintain line numbers when running in a browser console
- did not support hierarchical categories
- supported custom log levels but did not expose that type information (which doesn't help when you work in typescript).
Daslog is a fun little project that solves these problems for me. It lets us do neat things like
const log = logger()
.setCategory('Utilities')
.subCategory('Promises')
log.info('Hello World');
// [ 2019-10-03 02:42:19 | INFO | Utilities > Promises ] Hello World
// This line is type safe and updates the signature of newLog's chain to
// [Sigil.Time, Sigil.Category, Sigil.Level]
const newLog = log.reformat(([time, level, category]) => [time, category, level] as const);
const onePunchLevels = levels([
'wolf',
'tiger',
'demon',
'dragon',
'god',
]);
const animeDisaster = newLog.setLevels(onePunchLevels).setCategory('Disaster');
animeDisaster.demon('the city is in danger');
// [ 2019-10-03 02:50:03 | DEMON | Disaster ] the city is in danger
This logger
- Supports node.js and browser runtime environments
- Allows for simple bundling with zero dynamic requires
- Provides a default console logger that retains line numbers.
- and the ability to create your own such appenders.
- Features customizable log levels
- and more importantly customizable log levels
- and yet more importantly typings that update as you create loggers that use these custom levels
- and more importantly customizable log levels
- Features categories and subcategories
- Is designed to be immutable and creates new logger instances leaving the old loggers in place.
reformat<T extends Tuple<Sigil>, U extends Tuple<Sigil>>(func: (current: T) => U)
: Completely restructure the chain.
Note the use of as const
may be necessary to maintain a well-typed sigil chain. If it's missing the tuple will instead become Sigil[]
, preventing well-typed reformats in the future.
Resets the category of the logger to the parameter.
The structure of options is:
interface SetCategoryOptions {
/**
* If my logger's categories are [A, B] and I call setCategory('D'),
* should my new list be [D] instead of [A, D]? (false by default)
*/
reset?: boolean;
/**
* If my logger does not have a category in its chain, should this
* command add it to the end? (true by default)
*/
append?: boolean;
}
Categories are internally represented as a linked list. subCategory adds a new node to the end of that list.
2019-10-03 02:42:19 | info | Utilities > Promises | Hello World
The magic. The levels argument must extend LogLevels i.e. implement {[level: string]: number}
. See the default log levels
Some people care about keeping a good spread in the values. It's not a bad idea and I've done so for convenience but in a pinch remember you can always remap the values to have wider gaps in the same order.
If this is set to 'Infinity'
logging is effectively turned off (unless you happen to have a log level with a weight of Infinity
which... don't do that). Frankly I included this for parity with other loggers, I don't have a use case where I turn logging off fully.
The logger will also include log functions. By default these will correspond to the default log levels, i.e.
trace
debug
info
warn
error
fatal
This project was inspired by asking "I wonder if you can make a logger that updates its type signature when you set custom levels". Being familiar with typescript it seemed like it should have been possible which made it strange that the logging frameworks I researched at the time (the js loggers with @types/
declarations) didn't do it. Log4j would do this if it could. Alas, doomed is the java programmer. Thankfully they're used to that.
The log functions are generated from an object acting as a key-value pair of log name -> weight.
Mapped types existed at the time I originally wrote this, but as const
and the more mature handling of literal tuples that's capable now did not.
Note: Log levels will be assigned on the logger meaning they will conflict with the log functions. Do not use log levels of:
append
,reformat
,setCategory
, etc.
Sigils are responsible for the 2019-10-03 02:42:19 | info | Utilities > Promises
fragment of the log message in the example. The logger handles this by maintaining a list of markers or sigils indicating the items the logger is expected to fill in and the order they are in. The above fragment is generated by calling .info()
on a logger with the sigil chain [Sigil.Time(), Sigil.Level(), Sigil.Category()]
. Depending on the sigil, you may be able to pass extra information into the sigil constructor. For example, Sigil.Time()
accepts the dateformat format strings.
Sigil types:
- Level
(format: string) => string
-- function to apply a modifier to the levels. By default,.toUpperCase()
- Category
- Time
(format: string)
-- dateformat format string - Label
(label: string)
- Function
(func: (meta: AppenderFactoryMeta) => string)
Wrapper functions obfuscate the call site of the log message. That alone makes it worth it to me. By having a factory that returns a preconfigured partially bound console.log()
I retain the information of the call site and trivially support all of the powerful console
functionality. Note that this doesn't prevent a user from creating custom factories that generate a wrapper, it just provides a mechanism for users to create factories that don't.
This was a natural consequence of using factories to get line numbers. We can no longer perform the logic to decide whether or not to print the message when the log function is called so we must do so when the logger is created/configured. Besides, it's kinda neat that turning off logging means you call the empty func () => {}
every so often and that's the only performance impact.