Skip to main content

context: syntactic sugar for traversing `map`

In the multi-branch example in the previous section, both branches may be skipped. To record the status of each branch and aggregate the results into the final result, many key-value pairs and complex map configurations are required.

In practical business scenarios, we find that this can be very cumbersome. This section introduces a best practice for sharing context.

Custom Global Shared Struct​

Define a struct in r_1 that can record the results of all branches and write it to the context key:

struct OurStruct{
    bool a_skiped = false;
    bool b_skiped = false;
    std::vector<float> a_result;
    at::Tensor b_result;
}

(*input)["context"] = std::make_shared<OurStruct>(); //input is a dict

In each branch, only process its own part of the data. In this way, we only need one key-value pair context to accommodate all data.

Data Race

Unless it is read-only, multiple branches are prohibited from accessing the same part of the context.

context Syntax Sugar in map​

In the figure above, if b_2 does not configure map, it will share the same data (dict) with b_1. If a_2_1 does not configure map, it will get a copy of the data from a_1. As long as map is not configured, the node can always access context. However, when map is used, a brand new dict will be created and the specified key-value pairs will be copied over, which may not necessarily contain the context item.

Syntax Sugar

For map configuration, if the data source has a context attribute and map does not have a mapping for context, map will automatically add a mapping for context.