WebGuy is a Guy for the Web

Table of Contents generated with DocToc

• WebGuy is a Guy for the Web
  • props
  • time
    • Configuration
    • Performance Considerations
  • environment
  • trm
  • types
    • API
    • Type Signatures
  • To Do
  • Is Done

WebGuy is a Guy for the Web

props

• public_keys = ( owner ) ->: return a list of property names, including inherited ones, but excluding non-enumerables, symbols, and non-userland ones like constructor.

• get_prototype_chain = ( x ) ->: return a list containing object x and all the objects encountered when walking down x's' prototype chain (using Object.getPrototypeOf() repeatedly). If x is null or undefined, return an empty list because in that case there are zero objects where property lookup could happen. The reverse()d prototype chain is used by the two *depth_first*() method, below.

• walk_depth_first_property_descriptors = ( x ) ->: Given a value x, return an iterator

• acquire_depth_first = ( source, cfg ) ->: given a source object, walk the property chain from the bottom to the top (using walk_depth_first_property_descriptors()) and transfer all properties to a new or a given target object. This is most useful when used with a filter to select, a generator function to generate new, and / or a decorator to modify accepted and generated properties. acquire_depth_first() will keep the relative ordering: (1) 'top-down' for each object (properties declared earlier will appear, on the target, before ones declared later); (2) w.r.t. inheritance in the sense that the prototype of a given object x in the prototype chain will be looked at before properties on x itself is considered. Later properties may shadow (replace) earlier ones but it's also possible to forbid shadowing or ignore it altogether (see overwrite, below).

  When a cfg object is given as second arguments, it may have the below settings, all of which are optional:

  • filter: An optional function that will be called with an object { target, owner, key, descriptor, } for (1) each found property; it should return either true (to keep the property) or false (to skip the property); non-Boolean return values will cause an error.

  • descriptor: An optional object containing updates to each property's descriptor. Use e.g. descriptor: { enumerate: true, } in the call to acquire_depth_first() to ensure that all acquired properties on the target object will be enumerable.

  • target: the 'static' or 'default target', i.e. the object to which the properties are to be assigned to. If not given, a new empty object {} will be used. It is also possible to set a 'dynamic target' (that will override the static target) in the yielded values of generator, for which see below.

  • overwrite: controls how to deal with property keys that appear more than once in the prototype chain. Since acquire_depth_first()'s raison d'être is doing depth-first 'anti-inheritance', there are several ways to deal with repeated properties, as the case may be:

    • false (default): Throw an error when an overriding key is detected
    • true: Later key / value pairs (that are closer to the source value) override earlier ones, resulting in a key resolution that is like inheritance (but without the possibility to access a shadowed value).
    • 'ignore': Silently ignore later keys that are already set; only the first mention of a key / value pair is retained.

  • generator: if given, must be a generator function gf() (a function using the yield keyword). The generator function will be called with an object { target, owner, key, descriptor, } for each property found and is expected to yield any number of values of the format { key, descriptor, }. Optionally, this object may also have target set (the 'dynamic target'), which will be the object that the current property will be set on. This is useful e.g. to distribute multiple derived properties over a number of target objects.

    gf() will only be called if the property has not been not filtered out. Yielded keys and descriptors will be used to call decorator if that is set.

    Points to keep in mind:

    • The most trivial setting for generator, a generator that doesn't yield anything—( d ) -> yield return null; JS: function*( d ) { return null; }—has the effect of preventing any property to be set on the target. This is because the original key / value pair is not treated specially in any way, so the user can (and must) freely decide whether and where they want the original property to appear in the target.
    • Take care not to re-use the descriptor that was passed in without copying it. Instead, always use syntax like yield { key: 'foo', descriptor: { descriptor..., value: foo, } } to prevent leakage of (most importantly) the value from one property to another.

  • decorator: An optional function that will be called with an object { target, owner, key, descriptor, } for (1) each found property and (2) each generated property, too. The decorator function may return null or undefined to indicate no change for the given property; otherwise, it should return an object that will be used (like cfg.descriptor) to update settings in the property's descriptor—in other words, the returned object needs only to mention those parts of the decorator that should be changed, and most commonly, an object like { value: 'helo', } where only value is set will suffice. In case both cfg.descriptor and the return value of the decorator function mention the same descriptor settings, the ones returned by the latter (the decorator function) will overwrite those of the former (i.e. the decorator always has the last word).

time

WEBGUY.time contains facilities to create timestamps for purposes like logging or to create dated DB records.

Timestamps are one of those things that seem easy (like, +new Date() or Date.now()) but get quite a bit harder when you bring in typical constraints. One wants one's timestamps to be:

• precise: Computers are fast and a millisecond is sort of a long time for a CPU. Naïve JS timestamps only have millisecond precision, so one can easily end up with two consecutive timestamps that are equal. This leads to

• monotonous: You don't want your timestamps to ever 'stand still' or, worse, decrement and repeat at any point in time. Because new Date() is tied to civil time, they are not guaranteed to do that.

• relatable: Ideally, you want your timestamps to tell you when something happened. A 13-digit number can do that—in theory. In practice, only some nerds can reliably tell the difference between timestamps from today and those from last week or last year.

• durable: Time-keeping is complicated: Timezones are introduced and abolished, daylight saving dates can vary within a single country and may get cancelled in some years or split into two separate durations within a year; some people count years ab urbe condita, some days since CE 1900, some seconds and others milliseconds from CE 1970; in some years, you get a leap second and so on. For these reasons, local civil time is not a good choice for timestamps.

• Suffixes:

  • methods ending in f return floats,
  • methods ending in s return strings;
  • methods ending in 1 return a single value, contrasted with
  • methods ending in 2 which return a list of two values.

• stamp_f = -> utc_timestamp = performance.timeOrigin + performance.now(): return a float representing present time as milliseconds elapsed since the Unix epoch (1970-01-01T00:00:00Z), including microseconds as a fraction. This is the recommended way to measure time for performance measurements and so on, as it is reasonably precise and monotonic (i.e. it is unaffected by system time updates and will only ever increase). Here included as a convenience method.

• stamp_s = ( stamp_s = null ) -> ( stamp_s ? @stamp_f() ).toFixed 3: return the numeric timestamp or time.stamp_f() as a string with exactly 3 decimals; suitable for IDs, logs &c.

• monostamp_f2 = ->: return a list containing the result of

• monostamp_s2 = ( stamp_f = null, count = null ) ->: return a list containing the result of time.stamp_s() and a monotonic zero-based, zero-padded counter which will be shared across all callers to this method. Sample return value: [ '1693992062544.423', '000' ]; should time.stamp_and_count() get called within the same microsecond, it'd return [ '1693992062544.423', '001' ] &sf. Especially for testing purposes, one can pass in the fractional timestamp and a value for the counter.

• monostamp_s1 = ( stamp_f = null, count = null ) ->: return the same as monostamp_s2(), but concatenated using cfg.counter_joiner.

• stamp() is a convenience equivalent to monostamp_s1().

Configuration

cfg =
 count_digits:    3     # how many digits to use for counter
 counter_joiner:  ':'   # comes between timestamp and counter
 ms_digits:       13    # thirteen digits should be enough for anyone (until November 2286)
 ms_padder:       '0'   # padding for short timestamps (before 2001)
 format:          'iso' # should be 'iso', or 'milliseconds', or custom format

• format:
  • milliseconds: timestamps look like 1693992062544.423:000
  • iso: timestamps look like 1970-01-01T00:00:00.456789Z:000
  • compact: timestamps look like 19700101000000456789:000
  • dense: timestamps look like 19700101@000000.456789:000 for readability
  • any other string will be interpreted by the format() method of dayjs, with the addition of µ U+00b5 Micro Sign, which symbolizes 6 digits for the microseconds part. A minimal template that doesn't leave out any vital data and still sorts correctly is YYYYMMDDHHmmssµ, which produces compact format timestamps like 20230913090909275140:000 (the counter being implicitly added).

Performance Considerations

A quick test convinced me that I'm getting around 170 calls to time.monostamp_s1() into a single millisecond; these timestamps then look like

1694515874596.967:000
1694515874596.976:000
1694515874596.981:000
1694515874596.990:000
1694515874596.995:000

— that is, a repetition in the tens and hundredths of milliseconds is quite likely, but a repetition in the thhousandths of milliseconds (i.e. microseconds) is unlikely. It's a rare event (estimated to less than one in a million) that the counter ever goes up to even one. This tells me that on my (not up-market, not fast) laptop it should be more than safe to use three digits for the counter; however that may not be true for faster machines.

environment

( require 'webguy' ).environment is an object like { browser: false, node: true, webworker: false, jsdom: false, deno: false, name: 'node', } with boolean and one text properties that tell you in what kind of environment the code is running. Observe that there may be environments where no boolean property is true and name is null.

trm

• rpr = ( x ) ->: return a formatted textual representation of any value x.

types

API

• validate.t x, ...—returns true on success, throws error otherwise
• isa.t x, ...—returns true on success, false otherwise

Type Signatures

• string of variable length reflecting the results of a minimal number of tests that never fail and give each type of values a unique name

• Tests are:

  • the result of typeof x
  • the shortened Miller Device Name (MDN) obtained by Object::toString.call x, but replacing the surrounding (and invariably constant) [object (.*)]
  • the value's constructor.name property or 0 where missing
  • the value's Denicola Device Name (DDN), which is the constructor property's name or, if the value has no prototype, the digit zero 0.
  • the value's Carter Device Name (CDN), which is class for ES6 classes, fn for functions, and other for everything else. It works by first looking at a value's Miller Device Name; if that is not indicative of a function, the value's CDN is other. Else, the property descriptor dsc of the value's prototype is retrieved; if it is missing, the CDN is other, too. If dsc.writable is true, the CDN is fn; otherwise, the CDN is class.
  • N if Number.isNaN x is true, digit zero 0 otherwise

Results are joined with a slash /.

### TAINT test for class instances?

( typeof x )
( x?.constructor.name ? '-' )
( Number.isNaN x ) ].join '/'

( ( Object::toString.call x ).replace /^\[object (.+?)\]$/u, '$1' )
( x?.constructor.name ? '0' )
( if Number.isNaN x then 'N' else '-' )
###

xxx The [Carter Device (by one Ian Carter, 2021-09-24)](https://stackoverflow.com/a/69316645/7568091) for
those values whose Miller Device Name is `[object Function]`:

Also see [this detailed answer in the same discussion](https://stackoverflow.com/a/72326559/7568091).

[Link to specs](https://tc39.es/ecma262/#sec-runtime-semantics-classdefinitionevaluation)

###
get_carter_device_name = ( x, miller_device_name = null ) ->
  miller_device_name ?= Object::toString.call x
  return '-'  unless miller_device_name is '[object Function]'
  return 'fn' unless ( descriptor = Object.getOwnPropertyDescriptor x, 'prototype' )?
  return 'fn' if descriptor.writable
  return 'class'
console.log '^4234-1^', isa_class ( class D )
console.log '^4234-2^', isa_class ( -> )
f = -> new Promise ( resolve , reject ) ->
  console.log '^4234-3^', isa_class resolve
  console.log '^4234-4^', isa_class reject
  console.log '^4234-5^', Object.getOwnPropertyDescriptor resolve, 'prototype'
  resolve null
await f()
###
https://stackoverflow.com/a/69316645/7568091 (2021-09-24 Ian Carter)
https://stackoverflow.com/a/72326559/7568091
coffee> ( Object.getOwnPropertyDescriptor d, 'prototype' )?.writable ? false
{ value: {}, writable: false, enumerable: false, configurable: false }
coffee> Object.getOwnPropertyDescriptor (->), 'prototype'
{ value: {}, writable: true, enumerable: false, configurable: false }
###

To Do

• [–] types.isa.sized(), types.isa.iterable() test for 'existence' of x (x?) but must test for non-objects as well or catch exception (better)
• [–] define what iterable and container are to mean precisely, as in, provide the defining characteristic. Somehow we can e.g. iterate over a string as in x for x in 'abc' and d = [ 'abc'..., ] but Reflect.has 'abc', Symbol.iterator still fails with an exception ('called on non-object').
  • [–] In the same vein, what exactly is an object in JS? Maybe indeed anything that is not a primitive value (i.e. not null, undefined, true, false, number including Infinity and NaN (but not BigInt)). As such, maybe primitive, nonprimitive would be OK?
    • Maybe any d for which [ ( typeof d ), ( Object::toString.call d ), ( d instanceof Object ), ] gives [ 'object', '[object Array]', true ]. This would include instances of a plain class O; which are implicitly (but somehow different from explicitly?) derived from Object. One could throw the Dominic Denicola Device i.e. d.constructor.name into the mix which would then exclude instances of class O;.
• [–] implement in WEBGUY.errors custom error classes with refs, use them in WEBGUY.types
• [–] disallow overrides by default when extending class Isa to avoid surprising behavior (might want to implement with set of type names; every repetition is an error unless licensed)
  • [–] might later want to allow overrides not for entire instance but per type by adding parameter to declaration object
• [–] in props.acquire_depth_first(), fix handling of descriptors
• [–] use an instance of Types in its methods ('dogfeeding')
  • [–] consider to instantiate Types from Pre_types passing in an instance of itself (Types), thus allowing the instance to use 'itself' / 'a clone of itself' without incurring infinite regress

Is Done

• [+] in the Isa standard types, should e.g. integer only refer to integer floats (4.0) or to floats and BigInts (4.0 and 4n)? Could / should that be configurable? remove all mentions of BigInts in isa tests with a view to establish separate types for them in the future (bigint, zero_bigint &c)
• [+] in types.validate, return input value such that x is types.validate.whatever x is always satisfied unless x doesn't validate
• [+] in props.acquire_depth_first(), do not silently overwrite earlier properties with later ones; instead, use cfg.overwrite to determine what should happen (true overwrites, function calls back, false throw an error).
• [+] in props.acquire_depth_first(), add cfg.generator() (?) option to allow generation of any number of additional members in addition to seen ones. This should be called before cfg.decorator() gets called. Should probably require cfg.generator() to be a generator function.
• [+] in props.acquire_depth_first(), allow both generator and decorator to produce a 'local' value for target that will override cfg.target; this will allow to distribute properties over a number of targets.
• [+] WEBGUY.types.declare: consider to prohibit adding, removing types from the default export instance as it may be considered too brittle: declaring a type can potentially change results of type_of, too, so even consumers that do not make use of the new type could be affected. A dependent module may or may not see the same instance of WEBGUY.types, depending on their precise dependency declarations and depending on the package manager used. Types are now always declared at instantiation time, later declarations are not (and likely will not be) implemented.