Exception raised due to missing directory

[cgalant]

Exception raised due to missing directory

[landonreed]

@cgalant, you're right that this needs fixing, but I wonder if a better approach would be to initialize the FeedStore when the application starts up. It probably shouldn't be a static field on FeedVersion (we've been refactoring away a lot of the static fields/methods on these model classes).

@abyrd what are your thoughts here? If we move feedStore = new FeedStore(); to DataManager::main, we could have the application fail fast with a clear warning. I think I might prefer the application failing on this rather than accidentally creating directories where they're not supposed to be, but I am open to creating them automatically (or some other approach) if the above seems incorrect.

[cgalant]

@landonreed, any news ? I still got the error in the last mongo-pojo branch update !

[landonreed]

A simple fix for your implementation is to only reference existing directories in the config files. If you're meaning to have the feeds persist locally (rather than stored on s3), the directory for gtfs should not be placed in /tmp (for other implementations we typically place them somewhere like var/data/gtfs/.

I think the true issue here is that the application does not fail on startup if the directory does not exist. I stand by my initial preference that the application should error if a directory for gtfs files does not exist, rather than creating them automatically. Do you have any input on this approach @cgalant?

[abyrd]

@landonreed I don't think it's a problem to require directories to exist, but we do want to fail fast in that case. As I understand it you already began refactoring to remove static fields and could change this behavior in that set of changes, which is nonbreaking and can be applied as a point release after our main release incorporating RDBMS functionality.

[landonreed]

That's correct @abyrd. I will share these changes in a PR after we release the mongo-pojo version (and reference this issue). As a quick fix, however, I did add a change (which has been removed in the set of changes I'm working on for the point release) that makes the application fail fast.