[randomdata]

> It is absolutely not idiomatic to give any meaning to non-error values under error conditions in the usual case.

Based on what? We have come to see that it is beneficial to make zero values useful. In fact, Go Proverbs even says so. Likewise, we have learned it is useful to return the zero value when you have an error. Most commonly, this means returning nil, which is packed full of all kinds of useful information. Therefore, the T value can be expected to useful if the code is idiomatic.

I'd love to see some real-world code you think is idiomatic, but doesn't return a useful T value when there is an error.

> If a function returns an error, callers must treat all non-error return values as unspecified unless explicitly documented otherwise

This is not at odds with that. This merely warns that not all code you may call will be idiomatic. In fact, if I recall correctly, doesn't os.Open (maybe os.Create) return an invalid file handle in some error cases? The standard library is old and what helped us eventually see what is idiomatic. It is decidedly not idiomatic for the most part. If you rely on a function being written idiomatically, then you are going to run into trouble, as that is not a guarantee (unless the documentation provides such a guarantee).

But in the case of this Either wrapper, it explicitly states it is for use with idiomatic code, not any old code you can throw at it.

[awused]

>Based on what? ... In fact, Go Proverbs even says so.

And the Go style guide says otherwise: if err is non-nil, you shouldn't even check the other return values. The Go proverbs are just words, it doesn't make them true or even good ideas. When Go proverbs don't agree with how Go code is written, including idiomatic Go code, reality wins over a bad theory. The Go style guide happens to better match real code written by real people, even the people directly responsible for the proverbs.

>I'd love to see some real-world code you think is idiomatic

To be fair, I have been talking about the Go programming language and idiomatic code written in that language. You seem to have a different idea of "Idiomatic Go" from everyone else's. Most of the stdlib, including the newest additions, is idiomatic as judged by other people but evidently not by you.

[randomdata]

> if err is non-nil, you shouldn't even check the other return values.

Yes, it says you cannot trust functions, unless documented, to be idiomatic. Which is reasonable as not all code is idiomatic. The language goes to no lengths to enforce how the code is written in this regard (obviously). But we are talking about code that is known to be idiomatic.

> Most of the stdlib, including the newest additions, is idiomatic as judged by other people but evidently not by you.

1. I don't see how it could be. What is idiomatic emerges from writing code and seeing what works and what doesn't. Most of the stdlib was written in the early days before anyone understood what works best. And, thanks to the go1 guarantee, modifying it now is out of the question.

2. If you believe that the stdlib is idiomatic, then you have to accept that returning an int (-1) to represent an error is idiomatic. The standard library is full of that. Which violates the premise of FP-Go that (T, error) is idiomatic. That is not my claim, that is theirs.

[awused]

>Yes, it says you cannot trust functions, unless documented, to be idiomatic.

That is not what it says, in fact, it almost says the opposite. It is idiomatic Go to never return useful values if error is non-nil unless explicitly documented. It never affirms your particular, personal, definition of what "Idiomatic Go" is and directly contradicts it.

>If you believe that the stdlib is idiomatic

I believe most of the stdlib is idiomatic Go, especially the newer stuff. Your excuse earlier was that the older stuff in the stdlib isn't idiomatic but the newer stuff is. But the newer stuff doesn't match your personal standards for being idiomatic either. I would challenge you to point to a large body of code that actually matches your definition of idiomatic, I do not think it exists.

[randomdata]

> It is idiomatic Go to never return useful values if error is non-nil unless explicitly documented.

No, it is very much written from a consumer perspective. There is known, non-idiomatic, code where relying on T in its error state is problematic, so the guidance is reasonable and logical.

But we're talking about this from the producer perspective. These do not challenge each other and can exist in harmony.

> believe most of the stdlib is idiomatic Go, especially the newer stuff.

I wholeheartedly agree, at least with respect to the newer stuff, and have already said as such. Now this is where, in that newer work which is idiomatic, you point to a good example of where you can find meaningless return values when there is an error.

Clearly the vast majority of the standard library, especially in the newer stuff, does return meaningful values upon error, so we need to identify those which buck the trend if we want to hold it as a counterexample of this being an idiomatic practice.

[awused]

>No, it is very much written from a consumer perspective. We're talking about from the producer perspective.

This is just a lie. It's written about Go and doesn't split its perspectives, the producer should only ever return meaningful non-error values with an error if explicitly documented, and the consumer should only ever expect those when documented.

>you point to a good example of where you can find un-meaningful return values when there is an error.

Okay, skimming the release notes for 1.21. The slog package, brand new in 1.21, contains a MarshalText method (and others, but this is the first I noticed). It returns ([]bytes, err) and it is not documented to be idiomatic as per your personal definition. Therefore, by your interpretation of the style guide and your own personal standards, it is not idiomatic. What's more, a nil/empty slice is potentially valid output from marshaling something, so the slice is entirely useless and must be ignored if err is non-nil. It is not valid to try to "derive meaning" (your terminology) from the slice being nil or not, so it is well and truly useless if err isn't nil.

Moreover, I'll assert that if Either existed in the stdlib, it would return Either[bytes[], err] instead of ([]bytes, err), and it is a limitation of Go that they're using (T, error) which does not offer the appropriate semantics.

>Clearly the vast majority of the standard library, especially in the newer stuff, does return meaningful values upon error

This, too, is simply a lie.

[randomdata]

> The slog package, brand new in 1.21, contains a MarshalText method

At quick glance, I found three MarshalText implementations in the slog package. Which implementation are you referring to specifically?

And, for what it is worth, none of them return useless values on error, so where do I find the fourth which does?

> This, too, is simply a lie.

Let's hope. I love nothing more than being wrong. That means I get to learn something new! But I haven't found it yet, which is sorrily disappointing. Looking forward to you clarifying the above so we can put this to rest.