never understood the issue with go error handling. As soon as you start dumping exceptions as a valid error forwarding mechanism ( which seems like a totally acceptable design decision), you end up manually having to check every call you make that can raise an error, on each line.
I don't really see any alternative. It also makes you carefully think about how you plan on managing errors in your codebase, which also seems like a very sane thing to enforce.
You don't have to trap exceptions separately on every line. You trap a whole block of code and match on the exception type to determine how to handle the problem. It isn't perfect, nothing is, but I prefer systems languages where typical failures cannot easily be ignored and yet you are not burdened with constantly thinking about it.
> I prefer systems languages where typical failures cannot easily be ignored and yet you are not burdened with constantly thinking about it
This is self-contradictory. In particular, the only way you can have reliable error handling is if you are forced to think about each possible failure.
I assume by "cannot easily be ignored" you mean the way exceptions blow up at runtime? I don't find that an acceptable default for any non-scripting language.
I'd get annoyed too, because I know there are much simpler alternatives. Language design should encourage doing things right by making it more convenient than the shortcuts.
In Rust that entire check can be a single "?" symbol.
How much syntactic sugar is too much is a matter of preference, but I personally think that properly handling all errors without syntactic sugar turns into an unreadable mess because there's just a lot of things which could go wrong.
> I think just forwarding all low-level errors is a really bad habit
Why exactly is that a bad habit? In almost all situations where I return an error I already have enough context, I'm just wondering what else I'd add to that.
most errors you encounters are with i/o and are stupid "can't read, can't write or can't serialize".
In a network environment (which is originally what go was made for) you often need to add tracing information, business-level identifiers or processing information related to your state etc.
I'm currently writing a fairly complex api in go, and to be honest this really hasn't bothered me once.
Not to say it doesn't exists, but with time i've come very suspicious of people complaints over go. Most of the time those complaints come from people that didn't realize they missed an opportunity to have written a much much more elegant solution to their problem.
I never said it is, but to handle an error you have to pass it to the caller that can actually do something with it, whether it's trying alternatives, repeating the step, just logging it, or whatever else - a lot of functions just need to pass it on a couple of times and making this verbose in every single case doesn't make sense to me.
I don't really see any alternative. It also makes you carefully think about how you plan on managing errors in your codebase, which also seems like a very sane thing to enforce.