Is there a way to get consistent (or nearly consistent) comparable output?
My specific use case is that we have protobufs, and expect to have the graph files checked into the repo as part of the commit. The simplest way to do this (we do it for generated language bindings) is to run protodot, and check if the file is different from what is checked in.
Except the output isn't consistent.
From my checks, I found several things:
- date for
generated by comment at the end; we can ignore that
subgraph cluster_<id>: the ID appears to be generated using some data about the content. As far as I can tell, this is consistent- node names, e.g.
Node_Ja_<number>: the numbers appear to be consistent with the content in them
- subgraph sort order: this appears to be one major source of discontinuity
- node sort order within a subgraph: this appears to be another major source of discontinuity
I don't know the internals of protodot well enough, but it looks like consistent sorting of the subgraphs and nodes within subgraph might go a long way?
Is there a way to get consistent (or nearly consistent) comparable output?
My specific use case is that we have protobufs, and expect to have the graph files checked into the repo as part of the commit. The simplest way to do this (we do it for generated language bindings) is to run
protodot, and check if the file is different from what is checked in.Except the output isn't consistent.
From my checks, I found several things:
generated bycomment at the end; we can ignore thatsubgraph cluster_<id>: the ID appears to be generated using some data about the content. As far as I can tell, this is consistentNode_Ja_<number>: the numbers appear to be consistent with the content in themI don't know the internals of
protodotwell enough, but it looks like consistent sorting of the subgraphs and nodes within subgraph might go a long way?