PR #6447 adds a public API to get the maximum number of registers per thread (numba.cuda.Dispatcher.get_regs_per_thread()). There are other attributes that might be nice to provide - shared memory per block, local memory per thread, const memory usage, maximum block size.

These are all available in the FuncAttr named tuple: https://github.com/numba/numba/blob/master/numba/cuda/cudadrv/drive

Should probably throw a nicer error message like (with-decorator) does

Support &, |, ^, etc.

Make a concise print function for the QueryPlanningAnalysis class https://github.com/kensho-technologies/graphql-compiler/blob/v2.0.0.dev25/graphql_compiler/cost_estimation/analysis.py#L381

In most cases, the printout of the analysis passes is enough to explain why a particular query plan was chosen.

The current behavior is unnecessarily verbose and irritating.

artiq_ctlmgr
^CTraceback (most recent call last):
  File "/home/rj/src/conda/envs/py35/bin/artiq_ctlmgr", line 11, in <module>
    load_entry_point('artiq', 'console_scripts', 'artiq_ctlmgr')()
  File "/home/rj/src/artiq/artiq/frontend/artiq_ctlmgr.py", line 84, in main
    loop.run_until_complete(rpc_server.wait_terminate(