GitLab

Branch csv2json_with_repeating_template has an initial attempt at using a repeating streamset template in the csv2json application.

I am unsure about how to use this repeating stream set facility. Please add some comments to the header file or the wiki to document.

I have made a guess that you need to provide a vector of individual element values for a single stream, so I have tried to use the bytes of the template strings in csv2json as the values for pattern_t.

If this is correct, then the next question is whether it is possible to apply a kernel such as S2P to a repeating streamset to produce another repeating streamset. Maybe not, because I currently get an error in compiling csv2json on the new branch:

csv2json: /home/cameron/parabix-devel/lib/kernel/pipeline/compiler/analysis/relationship_analysis.cpp:1006: void kernel::PipelineAnalysis::combineDuplicateKernels(kernel::BuilderRef, kernel::ProgramGraph&): Assertion `isa<StreamSet>(G[relationship].Relationship)' failed.

If was not the intention to support the latter, then should I be breaking down the individual bytes into their bits and supply 8 vectors of bits?

Please review this application and advise the path forward for making use of repeating streamsets.

Actually, the ideal method for csv2json would probably be this:

1. We create a memory buffer having the concatenation of the template strings.
2. We set this up as a memory source kernel
3. We transpose to produce bit streams using S2P.
4. We have a casting facility that can cast the result streamset to an repeating streamset.