koa shouldn't special-case ENOENT errors

[julienw]

I noticed that koa special-cases ENOENT errors: https://github.com/koajs/koa/blob/ed84ee50da8ae3cd08056f944d061e00d06ed87f/lib/context.js#L146-L147 I think this is a mistake: ENOENT errors could happen for a lot of reasons (this happens with google's storage library when the key file isn't found, for example) that shouldn't get a 404 status. And BTW it's not even documented.

[Usamaliaquat123]

ENOENT errors could happen for a lot of reasons (this happens with google's storage library when the key file isn't found, for example) that shouldn't get a 404 status.

So, what exactly are you getting :)

[fl0w]

I think the semantics are pretty spot on. ENOENT is thrown when something (usually a file or a directory) isn't found, i.e. 404.

[...] this happens with google's storage library when the key file isn't found, for example [...]

I don't get it, isn't 404 a perfect match for this? Requested resource not found.

[julienw]

404 means "page not found", it doesn't mean "we got some error about a file not found". Koa doesn't know the context. Especially in this case it's a configuration error (some configuration file isn't found).

Probably the google library should throw a better error, but still.

[dead-horse]

I'm ok to remove the 'ENOENT' case, but it would be a breaking change.

[fl0w]

@julienw I guess it's a matter of subjective interpretation, but I guess my way of thinking is perhaps too "HTTP-minded". Either way, I think special casing should either be configurable or removed. But as dead-horse pointed out, it's a breaking change so at the very least, lets reference #1114 for future evaluation.

[julienw]

Yeah this is a breaking change, I totally agree with that. Note that this is alleviated by the fact that this behavior is documented nowhere. I had to look at the source code :-) (but still it's a breaking change).

I believe this was done initially to accommodate use cases like:

ctx.body = fs.createReadStream(filePath);

So I think we need a replacement for this. Maybe a middleware that would be part of koa (or another package) that would be then easy to add for somebody who wants this functionality? Would it be worth having a "compat-2.0" middleware that would cancel out all differences from the 3.0 changes, possibly configurable with flags to enable/disable individual bits?

[jonathanong]

👍 this should be done downstream

[jkomyno]

Has any decision been taken on this?

[IlyaSemenov]

I was hit by that as well. My perfectly legitimate API handler started to return 404 for no apparent reason.

What happened was, deep inside, a third party library (soap in my case) was doing fs.readFile for some of its internal stuff and was failing for unrelated reason. This should be generating a proper 500 Internal Server Error (which is then logged accordingly by upstream layers), not 404 how Koa presents it.

I don't see the problem with introducing a breaking change... May be it's time to fully use semver and don't be afraid of major version bumps? Many of us have done that and survived. 😆