The first argument to the error callback is *usually* a string code
from a list in the filedrop source; but sometimes it was the text
the server sent in the HTTP status line, instead. The latter isn't
predictable, and so it's not possible to write app code that uses it
to handle error consistently.
Instead, use that parameter for the numeric HTTP status code. This
still isn't totally clean in that sometimes it's internal filedrop
errors, as strings, and sometimes it's HTTP status codes, as numbers;
but at least both of those are things we can sanely handle with a
`switch` statement.
Also pass through `serverResponse`, which for a nice JSON error from
the server will contain meaningful information about the error which
the calling code can use for nice error handling. And just drop the
HTTP status text, which at best is redundant with the numeric code.
In passing, fix one case where for no obvious reason filedrop was
passing the file object but not the index.
This should be a pure refactor.
We'll replace this primarily with per-realm quotas (plus the simple
per-file limit of settings.MAX_FILE_UPLOAD_SIZE, 25 MiB by default).
We do want per-user quotas too, but they'll need some more management
apparatus around them so an admin has a practical way to set them
differently for different users. And the error handling in this
existing code is rather confused. Just clear this feature out
entirely for now; then we'll build the per-realm version more cleanly,
and then we can later add back per-realm quotas modelled after that.
The migration to actually remove the field is in a subsequent commit.
Based in part on work by Vishnu Ks (hackerkid).
Greg Price
Brock Whittaker
Cory Lynch
Philip Skomorokhov
Umair Khan
Leo Franchi
Tim Abbott