major slowdown since 2.5, backing out ???

[Bob Carpenter]

This is unrelated to the last message.

You may want to suspend work on current branches because
we're going to have to back out changes on develop that may
break new branches (I don't know enough about git to
say at this point).

Something that's been committed since 2.5 has slowed down
Stan by a factor of 2 fitting a simple logistic regression,
and we need to first isolate what it is. On the plus side,
the answer is stable --- we get the same draws with the same
seed.

Likely culprits we're exploring are

* change to header only
* updated OperandsAndPartials
* change to memory stack to plug leak

Argh!

- Bob

[Bob Carpenter]

I guess I should've guessed that our guesses
would be wrong.

The good news is that we should just be able to fix
what's there without backing anything out.

Daniel tracked the problem down to a change in the
way error-handling routines were called. The signatures
went from roughly

void foo(const char* bar, ...);

to

void foo(const std::string& bar, ...);

with calling pattern:

foo("baz", ...);

which is bad news because the string for the argument has
to be constructed on the fly.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
>