Saturday , January 16 2021

Go’s Features of Last Resort, Hacker News

A “Feature of Last Resort” (FOLR) is a useful feature which solves certain otherwise hard-to-solve problems, but are often best avoided.

I stole this term from Mercurial’sFeatures of Last Resortwiki page. As far as I know it originated on that page (and isn’t widely used outside of it). Many complex systems have these kind of FOLRs, and being explicit about it makes things easier for new users, who do not yet have the experience to distinguish them from regular features.

At any rate, this is a list of what I consider FOLRs in Go.[1]

Struct tags allow annotating struct fields with extra metadata to be retrieved at runtime with reflection:

type T struct {     F int `json:" f "` }

Why it exists and when to use– provide name aliases for marshalling and unmarshalling, which would be tricky otherwise (pass amap [string] stringtojson.Marshal ()?)

Problems– lack of type checking; typos can pass silently.vetcatches a bunch of spacing errors, but won’t catch misspellings. Libraries that use struct tags often don’t error out at runtime (you can never be quite sure if a struct tag is intended for you or not, so erroring out may cause conflicts).

Especially some of the more creative uses of struct tags is essentially “Programming by comments”. There is no “discoverability” in the form of API docs, code completion, etc. If I see the struct tagvalid: "yikes"then how do I figure out what this means to who? With regular Go code I can “jump to definition ”or grep, but for struct tags this is much harder.

It also rather hard to read (non-aligned) and edit (follows from being hard to

Alternatives– it depends on what exactly you’re doing, but data (eg map) or function calls can cover almost all struct tag cases.

Empty interface

The empty interface (interface {}) is an interface with no methods, and is satisfied by any type.

Why it exists and when to use– sometimes you really need to accept any type (but this is not very common for most types of programs).

Problems– no compile-time type checks; limited tooling support. Reflection is hard, may be quite slow, and often incomplete (are you suremap [string] map [int] [] struct {}will work for your function that accepts “any” type?)

Alternatives– this is theGreat Go Generics Debate. Alternatives include:

  • Write it multiple times for each type (possibly generated).

  • Use an actual interface with methods.

  • Convert to other type; e.g. just acceptint 64instead ofintand call asfun (int 64 (i)).

SeeSummary of Go Generics Discussionsfor some more details and Examples.

Imports aliases, dot-imports

Imports can be aliased asb "foo / bar", so that instead ofbar.Xyou useb.X. All identifiers can be imported to the current namespace with. "foo / bar", in which case you’d use justX.

Why it exists and when to use– sometimes you need to alias packages to prevent conflicts. Ideally package names should be unique, but sometimes it happens when combining third-party packages. It’s also needed when a directory name contains invalid identifier characters (eggo-pkg).

The dot-imports exist mostly for tests that run outside the package they’re testing (pkg_testpackages).

Problems– names often assume the package name, for examplezip.Fileis clearer thanz.FileorFile. It almost always hurts readability. Especially when used to shorten longer package names it can be unclear what something likebrefers to.

Alternatives– just use regular imports. *******

panic ()

Thepanic ()builtin displays a message, stack trace, and halts the program, unlessrecover ()’d.

Why it exists and when to use– sometimes the program really cannot continue and the only thing left to do is to “panic”. It’s a way to indicate that something impossible has happened (e.g. exiting an infinite loop), initialisation errors (“can’t find foo.html”), or some types of programmer errors (“can’t pass empty string”.

I think “never use panic”, or even “never use panic in libraries” is too strict. Using panics to signal initialisation errors is often convenient, especially for functions which would otherwise not have error returns.

Problems– using panics as if they’re exceptions is probably the most common mistake I’ve seen new Go programmers make. It’s understandable (I did it myself!), but it’s not how Go is expected to be used.

Panicscanbe used like exceptions, but this does not fit Go’s design aesthetics very well. Exceptions are generally considered to not be worth the cost; see e.g.Exception Handling Considered Harmful

Panics are a bit tricky to recover from and are often unexpected since it’s not the standard way to deal with errors in Go. In general it’s best to forget thatrecover ()exists and consider panics as a shortcut for:

fmt.Fprintln (os.Stderr, "oh noes!") debug.PrintStack () os.Exit (1)

Are yousurethis is what you want to do? Sometimes it is, but often it’s not.

Alternatives– most of the time just return errors.

init ()

Allinit ()functions in a package are automatically run when it’s imported.

Why it exists and when to use– initialize data and set up state, for example lookup tables or some package-global object.

Problems– there are a bunch of (potential) problems:

  • It’s “hidden” code that gets run, which can be surprising and unexpected especially if the code is non-trivial.

  • Relying on global state is often not the best solution.

  • It can make testing needlessly difficult.

  • It’s hard to signal errors (outside of (panic).

  • Can be a waste of resources ifinit ()sets up something forA (), and you just wantB ()

Alternatives– depending on what you want to do:


    Instead of relying on package globals, usingNewFoo () Fooconstructors and passingFooas a function argument is often a good alternative. SeeA theory of modern Go.




    For setting up a package-level variable a self-executing function will work just as well, and is clearer about what it’s doing instead of relying on side-effects:


    var geodb=func () * geoip2.Reader {     g, err:=geoip2.FromBytes (pack.GeoDB)     if err!=nil {         panic (err)     }     return g } ()


    Which is even shorter than theinit ()solution:


    var geodb * geoip2.Reader  func init () {     var err error     geodb, err=geoip2.FromBytes (pack.GeoDB)     if err!=nil {         panic (err)     } }


    Note I don’t agree these kind of package-level variables shouldalwaysbe avoided likeA theory of modern Goadvocates. e.g. in the example abovegeodbis essentially just a fancy IP address → country code lookup table, and passing that through several layers of functions for just one call is IMO less clear.



cgo allows calling C code from Go.

Why it exists and when to use– interacting with C libraries, which are ubiquitous. Perhaps the most commonly used cgo library is SQLite, which is a perfectly valid use and would be impossible to implement without cgo.

Problems– it’s comparatively slow, it inherits some of C’s problems such as unsafe memory, can be tricky to understand by Go programmers, and it makes builds slower and more tricky. Seecgo is not Gofor a more detailed breakdown of possible issues.

Alternatives– write it in Go if at all feasible.



    The list is not 100% comprehensive. For example Go compiler directives      and arguably type aliases could also be included, but I have not      witnessed them being abused often so I excluded them for brevity.


Brave Browser
Read More

About admin

Check Also

Samsung Galaxy S20 Plus' upcoming camera specs, features – Android Authority,

Samsung Galaxy S20 Plus' upcoming camera specs, features – Android Authority,

Cash KaroThe Samsung Galaxy S20 series’ launch event is just around the corner. Now, just under a month ahead of the big day, we’ve seen some leaked information about the Galaxy S20 Plus’ upcoming camera specs and features thanks to XDA Developers.According to XDA, the primary rear shooter on the Galaxy S20 Plus will feature…

Leave a Reply

Your email address will not be published. Required fields are marked *