Device spec API implementation

[thisac]

Context: The device spec API need to be implemented so that it is used/called when creating/compiling a program.

Description of the Change: Three major changes have been made:

1. CircuitSpecs have been renamed to Compiler.

2. A new property eng.spec has been added to RemoteEngine that pings the device spec API and downloads the hardware device specifications and stores it in a DeviceSpec object.

3. The Program.compile method has been updated to support DeviceSpec objects as inputs as well as an optional compiler argument.

Benefits:

Possible Drawbacks:

Related GitHub Issues:

[codecov[bot]]

Codecov Report

Merging #432 into master will increase coverage by 0.11%. The diff coverage is 99.27%.

@@            Coverage Diff             @@
##           master     #432      +/-   ##
==========================================
+ Coverage   97.84%   97.96%   +0.11%     
==========================================
  Files          58       67       +9     
  Lines        6784     6932     +148     
==========================================
+ Hits         6638     6791     +153     
+ Misses        146      141       -5     

Impacted Files	Coverage Δ
strawberryfields/apps/plot.py	100.00% <ø> (ø)
strawberryfields/apps/train/embed.py	100.00% <ø> (ø)
...awberryfields/backends/gaussianbackend/__init__.py	100.00% <ø> (ø)
strawberryfields/backends/gaussianbackend/ops.py	100.00% <ø> (+2.70%)	:arrow_up:
strawberryfields/utils/post_processing.py	100.00% <ø> (ø)
strawberryfields/utils/program_functions.py	97.77% <ø> (ø)
strawberryfields/utils/random_numbers_matrices.py	100.00% <ø> (ø)
strawberryfields/utils/states.py	100.00% <ø> (ø)
strawberryfields/io.py	97.70% <80.00%> (-1.13%)	:arrow_down:
strawberryfields/backends/fockbackend/ops.py	94.04% <94.11%> (+2.04%)	:arrow_up:
... and 60 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update f77e879...20bf121. Read the comment docs.

[josh146]

@thisac, I've worked a bit on this PR today, finishing the circuitspec to compiler refactor, and integrating in the device specs. In particular, I've made the following changes:

• Factored out the parameter validation into a separate DeviceSpec.validate_parameters method.

• Added support for validating gate parameters against the device spec by calling DeviceSpec.validate_parameters in Program.compile().

• Renamed the BaseBackend.circuit_spec property to BaseBackend.compiler

• Deleted the TensorFlow circuit spec (since it is identical to the Fock circuit spec).

• Deleted X8Spec and X12Spec.

• Updated the compilation section in doc/introduction/circuits.rst.

• Removed modes, remote, parameter_ranges, and local from the Compiler class, since they have never been used (and belong in the DeviceSpec class instead).

• Previously, sf.io.to_program would attempt to both convert a blackbird program to a SF program, and compile it using the bb.target attribute. This no longer makes sense, since device targets are no longer local, so implicit compilation has been removed. This probably should not have been occurring at all; a program loader should not implicitly be transforming the program.

• Finally, the engine would previously skip compilation if prog.target == eng.target. This also does not make sense; having the target attribute of a program set does not guarantee that a program has been correctly compiled, and is easily spoofed. For now, every submitted program is recompiled, but we should determine a better approach of certifying a program as 'compiled'.

Still to do:

• Add more tests to cover the new logic

• Currently, the API isn't returning default compilers for X8 and X12. We need to work out what to do for the time being --- add this to the API, or have a short term hot-fix in SF, whereby we assume compiler = "Xcov" if device_spec.default_compiler == [].

[josh146]

@antalszava @thisac I think this is now ready for review! @nquesada, what should the default compiler be for the X chips? Xcov or Xunitary?

[thisac]

@josh146 @antalszava I've updated the signature of Program.compile to compile(self, *, device=None, compiler=None) along with all the places that it was called. I've also added this to the docstring + documentation.

The only thing I think is left now is fixing the pylint issues regarding the tf-backend not overriding inherited methods, but I'm not sure why this is happening now (nor what would be the best way to handle it; disabling the error, etc.)

[josh146]

[ch263]