thoughts about arenas

1 files changed, 14 insertions, 0 deletions

diff --git a/proc.c b/proc.c
index ff1197f..0877a91 100644
--- a/proc.c
+++ b/proc.c
@@ -3,6 +3,20 @@
 
 /* procedures */
 
+/* the tangle of arenas here will need to be sorted out eventually: currently
+ * we have two arenas for every single procedure, a permanent one lasting the
+ * full length of compilation, and a scratch arena reset after every statement.
+ * so, long-lasting node data and very temporary allocations.
+ *
+ * this works fine when there's just one procedure, but long-term there should
+ * be a couple arenas per _worker thread_, not per procedure.  and for this,
+ * there's a third kind of data --- stuff that lasts the length of parsing one
+ * procedure, but no longer; scope information, mainly.
+ *
+ * so in the end we might have like
+ * _Thread_local Arena graph_arena, scratch_arena, parse_arena;
+ */
+
 void proc_init(Proc *proc, Str name) {
 	memset(proc, 0, sizeof(Proc));
 	Graph *g = &proc->graph;