Ask HN: What would you change about Python for scientific computing

I'm working on scientific / technical computing in Python and having been thinking about what the ideal technical computing language might look like, geared towards scientists not programmers. A lot of technical computing is simply library support (e.g., scypy and numpy), but some language level features are helpful.

I'd be interested in what HNers wish Python had (and didn't have) at the language level. My preliminary list is as follows:

* Matrix and data.frame type in core, similar to numpy or R.

* IEEE 754 floating point behavior. 1/0 = Inf, -1/0 = -Inf, 0/0 = NaN, etc. This is how numpy (matlab, R) works.

* Elementwise and objectwise matrix operators (PEP 225)

* Coroutines (PEP 380, PEP 3152 extending python with yield from) to enable better language level async programming and things like agent based models.

* Better access to expression trees or a unified way to push computation to GPU, database, etc. Perhaps something like linq and expressions trees support rather than Python's list comprehensions. LINQ is becoming unifying for Dryad map/reduce module, PLINQ, data access, and reactive programming (Rx). I'm less sure about the best model here...

Others?