In the past, once things got past the HTTP parser, all API misuses
were *LP developer bugs*.
Now though, folk can use the model code directly via the LP API, and
so things that previously were a programmer error on the LP team are
now a programmer error on a consumer of the API.
It makes sense to me that these are *still AssertionErrors*, its still
a programming error.
(That said, I try not to use assert much if at all, ValueError is a
better way of signalling an incorrect input to a function) - I'm
simply noting that all that has happened here is the population of
'programmers' has increased, whether something is a programmer error
not not hasn't altered.
I think there is an undisclosed tension here.
In the past, once things got past the HTTP parser, all API misuses
were *LP developer bugs*.
Now though, folk can use the model code directly via the LP API, and
so things that previously were a programmer error on the LP team are
now a programmer error on a consumer of the API.
It makes sense to me that these are *still AssertionErrors*, its still
a programming error.
(That said, I try not to use assert much if at all, ValueError is a
better way of signalling an incorrect input to a function) - I'm
simply noting that all that has happened here is the population of
'programmers' has increased, whether something is a programmer error
not not hasn't altered.