> For now, why don't we both put it on our task list, and if one of us
> feels like making the first draft, email the other to old-fashioned
> "check out" that task.

Sounds good.

> First, though, we should start publicly collecting desiderata.  Offhand:
>
> As previously mentioned, it should be a union of "system errors", which
> I define to be stuff in the stack that the Scheme system sits on or uses
> (this *might* include a Boehm GC library), and errors library writers
> need to signal when using system code.  I.e. making a simple test and
> raising an error instead of handing it to an FFI which will segfault.

I still think "system" is quite vague, and if we go that far we might as
well go the whole way and make it a generic errors-with-metadata SRFI.

> (Optional) indication of where the error happened!!!  Not just what
> procedure, but allow for a Scheme to insert a file name and line number,
> or the user to say "That port you think you gave me?  It isn't a
> port....", that is, a string that can be searched on.  A scheme like
> the one below could also provide the same thing if not for the cut
> and paste disease:

This is nice. It's starting to look more like a generic error SRFI,
giving the ability to tack arbitrary metadata onto errors (some of it
standardized), than an OS-error-only SRFI.

> As previously discussed, we can use a short abbreviation to make the
> source of error codes more obviously, like Oracle's "ORA-".

Since Scheme has rich hierarchical data structures we could have
pre-parsed codes such as ("ORA" . "WHATEVER") or ("ORA" . "WHATEVER").
Of course, in the specific case of Oracle if the standard convention is
to call them "ORA-1234" then we should probably follow that.