We were scratching out own itch by doing this. This is currently used by several of our projects.
It does feel like the HTTP libraries should be better and handling these various exceptions. Since they don't, we've ended up writing this.
For the most part, whenever we do any HTTP call, we are expecting a specific return code and we want to raise an easy to handle exception when this happens. However, many HTTP call don't even get to the point where an error code is return when the network is not reliable. We've found that in most case, we want to handle that exception in the same way than an unexpected error code.
Handy! A lot of Ruby libraries do a pretty bad job protecting their users from underlying exceptions that should be handled within the library. I'm hoping that with exception causes in Ruby 2.1, we'll see more libraries wrapping common exceptions.
Another option for dealing with sloppy exception handling by libraries is toe_tag, which we open-sourced at my work:
It allows you to declare groups of exceptions that can be caught with a single rescue clause. This has been extremely useful when dealing with ActiveRecord's tendency to bubble up exceptions from the underlying drivers -- the native and JDBC drivers for Postgres raise completely different exceptions for the same errors.
In some cases, exception classes aren't enough to determine whether an exception actually should be caught. A lot of libraries end up forcing you to make decisions based on the text of the message, or on some custom method (#error_code or what have you). toe_tag helps clean up those checks too.
`begin...rescue` is almost syntactic sugar for `case...when`. Actually, for a brief period it was possible to use any object that responded to #=== in `rescue`. That turned out to be an accidental feature and was reverted, but the first version of toe_tag was built on top of it. Now I just make sure to inherit from Module and that's enough.
It's good to know people are interested in this! That's enough to convince me to get it hooked up with Travis CI and give it an official, supported 1.0.1 release.
This definitely works, and I've done it a lot in the past (and still would if my requirements are simple enough), but there's a few useful things that toe_tag does that this doesn't.
First, you can pass it an array of exception names, and it will ignore ones that aren't defined. This is really useful if you need to group together exceptions from multiple libraries that may or may not be loaded (this happens a lot if you're writing an app expected to work both in MRI and JRuby).
You also don't need to know anything special about toe_tag's synthetic exception classes to use them. No splat operator needed. You can also combine exception groups into larger groups, which you couldn't do with nested arrays and splat.
Oh, definitely. I didn't mean to diminish the cool things toe_tag does - I just tend to dislike the tendency for novices to just use gems to do things without understanding why they work, o I like to point out where solutions can be accomplished in bare Ruby. :)
toe_tag sounds like a great tool for the toolbox, but if the need is "I want to rescue multiple exceptions at once in a DRY fashion", there's no magic needed.
Understood and agreed, as a fellow "just use Ruby"-ist. Once you've done something a few dozen times or so by hand, you start thinking about whether it might be handy to have a library for it.
But as someone who is employed to maintain and improve more than one Ruby app with more than 200 gem dependencies apiece, I can say with certitude that there is such a thing as going too far.
I have actually found that even fairly experienced Ruby developers have a limited understanding of how Ruby exceptions work. Usually there's one or two gaps -- like not knowing that `rescue Exception` is a bad idea, or not knowing `rescue => err` works (as opposed to `rescue StandardError => err`), or that multiple rescues can be added to a single begin block (I've seen nested `begin...rescue` constructs for exactly that reason), or that `rescue A, B` works. Or what the return value of `begin...rescue...ensure` is.
(Or that `begin...rescue` can take an `else`, which is another example of its shared behavior with `case...when`. I still have yet to see begin-else used in real code, though, and I wouldn't encourage it.)
9 comments
[ 2.8 ms ] story [ 23.7 ms ] threadWe were scratching out own itch by doing this. This is currently used by several of our projects.
It does feel like the HTTP libraries should be better and handling these various exceptions. Since they don't, we've ended up writing this.
For the most part, whenever we do any HTTP call, we are expecting a specific return code and we want to raise an easy to handle exception when this happens. However, many HTTP call don't even get to the point where an error code is return when the network is not reliable. We've found that in most case, we want to handle that exception in the same way than an unexpected error code.
This little gem let us do that easily.
Enjoy :)
Another option for dealing with sloppy exception handling by libraries is toe_tag, which we open-sourced at my work:
https://github.com/crowdcompass/toe_tag
It allows you to declare groups of exceptions that can be caught with a single rescue clause. This has been extremely useful when dealing with ActiveRecord's tendency to bubble up exceptions from the underlying drivers -- the native and JDBC drivers for Postgres raise completely different exceptions for the same errors.
In some cases, exception classes aren't enough to determine whether an exception actually should be caught. A lot of libraries end up forcing you to make decisions based on the text of the message, or on some custom method (#error_code or what have you). toe_tag helps clean up those checks too.
I'll use your gem next time I find myself doing that :)
It's good to know people are interested in this! That's enough to convince me to get it hooked up with Travis CI and give it an official, supported 1.0.1 release.
First, you can pass it an array of exception names, and it will ignore ones that aren't defined. This is really useful if you need to group together exceptions from multiple libraries that may or may not be loaded (this happens a lot if you're writing an app expected to work both in MRI and JRuby).
You also don't need to know anything special about toe_tag's synthetic exception classes to use them. No splat operator needed. You can also combine exception groups into larger groups, which you couldn't do with nested arrays and splat.
toe_tag sounds like a great tool for the toolbox, but if the need is "I want to rescue multiple exceptions at once in a DRY fashion", there's no magic needed.
But as someone who is employed to maintain and improve more than one Ruby app with more than 200 gem dependencies apiece, I can say with certitude that there is such a thing as going too far.
I have actually found that even fairly experienced Ruby developers have a limited understanding of how Ruby exceptions work. Usually there's one or two gaps -- like not knowing that `rescue Exception` is a bad idea, or not knowing `rescue => err` works (as opposed to `rescue StandardError => err`), or that multiple rescues can be added to a single begin block (I've seen nested `begin...rescue` constructs for exactly that reason), or that `rescue A, B` works. Or what the return value of `begin...rescue...ensure` is.
(Or that `begin...rescue` can take an `else`, which is another example of its shared behavior with `case...when`. I still have yet to see begin-else used in real code, though, and I wouldn't encourage it.)
Thank you for the kind words!