Choose timezone
Your profile timezone:
Powered by Indico
v3.3.3-pre
Present: Danilo, Enric, Guilherme, Enrico, Pere, Jim, Philippe
TDF Sprint
TDF From an Empty source
- Interface: In practice translatable to a constructor with the number of entries to be created
- Implementation: two routes
1. Base everything on TTree and create for example an TTreeEmpty
2. Write sources also not based on TTrees, such as in the empty source case
- If we go for 2. we can add an if in the Run method and rely in TThreadedExecutor (prototype already available)
- If we go for 1. we can use all as it is and modify the TTreeProcessorMT which has to deal with a file which does not really exists
- It is understood that we'll discuss the data sources after 6.10
- Axel's comment: https://sft.its.cern.ch/jira/browse/ROOT-8757?focusedCommentId=81030 "Stop when this number of entries after N filters has been achieved".
- We recognise the importance of the feature
- We shift the discussion to the next release: we don't have an implementation plan (analogy with Ranges) and a clear picture of the usecase.
Decision: We will go for the "special TTrees" TTreeProcessorMT. We add a constructor where we pass a name of a file or an sql db or a csv file. We'll look into the header of the file to infer its type or/and decide based on its extension if this is not possible.
Parallel tree writing
- PR on github: 533
- We take note of the comments of Vassil and agree with them
- We want a factory pattern where the "merging files" are created starting from a server
- We want to expose the THreadedObject and the 2 classes as well
- The queue shall be put in ROOT::Internal in core/cont if at all
- Metaquestion: used by GV, ROOT. Do we need to have this and other containers in veccore?
Generalised jit driven type inference
- PR 532
- Extend the mechanism to all methods
- Type inference shall be there also for Snapshot and Generic fill: we will look into the jitting of Define and Filter for that
We did not have the time to discuss the split of libThread into two components, one for threading utilities and one for expression of parallelism, where the latter depends on TBB.