Repository navigation
Check how gaia is using tskit/C++/R #141
Description
Activity
gaia implements almost exactly the infrastructure RcppTskit is meant to centralise:
-
It vendors and compiles its own copy of tskit.
-
Its S4 treeseq object stores external pointers to both a tsk_treeseq_t and a current tsk_tree_t: treeseq_class.R.
-
It manually allocates, wraps, finalises, and frees those structures: tsk_treeseq_load.c.
-
It writes separate C wrappers that copy tskit tables into R data frames: tsk_table_nodes.c, tsk_table_edges.c.
-
Its application algorithms extract the native pointer and work directly with the tskit C API.
-
It currently carries tskit 1.1.2 in-tree, so it owns the ongoing upgrade and security burden.
This validates RcppTskit’s basic design: downstream packages should be able to receive a TreeSequence$xptr, safely access tsk_treeseq_t in compiled code, and avoid implementing their own loader, finalisers, vendoring workflow, and routine wrappers.
Concrete lessons for RcppTskit
The largest missing downstream-oriented capability is tree iteration. gaia keeps a tsk_tree_t next to its tree sequence because its algorithms repeatedly traverse local trees. RcppTskit should eventually offer either:
-
a safe public C++ example for constructing and iterating tsk_tree_t from TreeSequence$xptr; or
-
an owned tree-iterator wrapper that preserves its parent tree sequence for its entire lifetime.
The lifetime requirement matters: a child tree or iterator must not outlive the tree sequence it references.
gaia also demonstrates the value of bulk table access. Returning nodes or edges as R data frames is useful for inspection and analysis.
-
@LynxJinyangii implemented iterator in #132 so that would then cover the suggestions and I logged data.frame export.
https://github.com/blueraleigh/gaia
https://github.com/blueraleigh/gaia-paper