From a37dc8a8462b2f7ae30a0226581ec29b2cf5b633 Mon Sep 17 00:00:00 2001
From: pompolic <pompolic@special-circumstanc.es>
Date: Fri, 7 Apr 2023 20:53:24 +0200
Subject: [PATCH] More comments
---
gdb-port/parser.py | 10 ++++++++--
1 file changed, 8 insertions(+), 2 deletions(-)
diff --git a/gdb-port/parser.py b/gdb-port/parser.py
index 3fca7d7..7eaff97 100644
--- a/gdb-port/parser.py
+++ b/gdb-port/parser.py
@@ -275,13 +275,19 @@ class ParserStack:
print("push_allocations:", push_allocations) # DEBUG
print("push_allocations_with_names:", {parser.name: pushes[index+1][1]-pushes[index][1] for index, parser in enumerate(parsers)}) # DEBUG
# TODO: looks like it needs to account for the case where top of stack is a string of pushes, but there are runs of pushes without corresponding pops (yet?) deeper in the stack
- # TODO: seems this is possible, but check if it really is:
- #print("all stack events:", self.stack_events) # DEBUG
+ print("top 30 stack events:", [ev[0] for ev in self.stack_events[-30:]]) # DEBUG
#for index, ev in enumerate(self.stack_events[::-1]):
# if ev[0] == StackEvent.POP:
# last_pop = index
+ # The list of stack events is split up into two kinds of "runs", runs of "push" events which don't have a corresponding "pop" yet
+ # And runs of push/pop events which are assumed to be balanced
+ # These latter can be straightforwardly accounted for by recursively removing the "outer" frame corresponding to a push/pop pair
+ # "push" events are trickier, because to get the allocations, they need to be compared to the *next* time hammer memory stats were sampled, be it push or pop
+
+ # TODO: perhaps a "pop" should act as a checkpoint, and commit every allocation up to that point
+
def calculate_and_clear_pushes_at_end(self):
# Throws StopIteration if no more items in the iterator match the condition
last_pop = next((index for index,ev in enumerate(self.stack_events[::-1]) if ev[0] == StackEvent.POP), None)
--
GitLab