Circular dependency when importing from utils in ops about openfermion

[idk3]

@kevinsung I'm late to the party, but one solution when all else fails is to import within the function. It looks like you've found a better solution through restructuring :)

[babbush]

Great. Write "Fixes #94" in that PR so that when we merge it will close this issue automatically.

[babbush]

These sorts of circular dependancies can be really annoying to deal with. You may need to change the order of imports in __init__ somewhere (breaking the alphabetical order) or you may need to do a much more specific import somewhere that does the import just inside a function instead of at the top of the file. In this sort of situation, I recommend just hacking to try to make things work. I do not think there is a nice general strategy to avoid this.

@Strilanc @maffoo @Spaceenter do you guys have any more enlightened software engineering tips for avoiding this sort of thing? @idk3 I know you dealt with this a lot in FermiLib. Any ideas?

[Spaceenter]

My past experience is that circular dependencies indicate design issues, and could usually be solved by a bit re-design.
For this case, it seems like 'util' should be the lowest level. I agree things like "swap_column" and "swap_rows" should belong to 'util'. But I feel trotter_error.py should probably belong to higher level, maybe even higher than Hamiltonion.

I think the layer should be util -> ops -> hamiltonion -> high-level "util" such as the trotter_error thing.

@idk3 @babbush - WDYT?

[Strilanc]

I agree with @Spaceenter

[babbush]

Actually I think that in most cases op is lower level than util. Kevin's example is a somewhat rare exception.

…

On Oct 30, 2017 19:04, "Craig Gidney" ***@***.***> wrote: I agree with @Spaceenter <https://github.com/spaceenter> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#94 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ANlTf5gPvL6n0-WWumU1bteSbpf7-Dt_ks5sxoA4gaJpZM4QL_H3> .

[kevinsung]

In that case one solution is to move swap_rows and swap_columns to ops/_quadratic_hamiltonian.py and import these functions in utils/_slater_determinants.py.

[babbush]

That sounds like a very reasonable solution.

…

On Oct 30, 2017 19:11, "Kevin Sung" ***@***.***> wrote: In that case one solution is to move swap_rows and swap_columns to ops/_quadratic_hamiltonian.py and import these functions in utils/_slater_determinants.py. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#94 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ANlTf-gwjZG_e4_u6KP80Ln_JWPhv5SXks5sxoHCgaJpZM4QL_H3> .

[kevinsung]

However, I realized that in my pull request, I import other things from utils in ops/_quadratic_hamiltonian.py.

[kevinsung]

Namely, I need to use givens_decomposition and fermionic_gaussian_decomposition from utils/_slater_determinants.py in ops/_quadratic_hamiltonian.py.

[kevinsung]

Perhaps utils/_slater_determinants.py needs a new home.

[kevinsung]

Okay, I think I have a satisfactory solution. I will restructure the code to resolve this in PR #95 .