evolution: 5-MOAD scan; evolution-0002 CWE-407 search_hit_cache dedup O(N^2) 433x at N=1000

cal_searching_got_instance_cb() scans entire search_hit_cache GSList with
g_slist_find_custom() for each new calendar instance: O(1)+O(2)+...+O(N)
= O(N^2). Fix: parallel GHashTable for O(1) membership test. 433x op-count
speedup at N=1000. MOADs 0002/0003/0004/0005 CLEAN.
This commit is contained in:
russell@unturf.com 2026-04-03 14:24:12 -04:00
parent 358360d744
commit 094c19c745
4 changed files with 245 additions and 1 deletions

View file

@ -35,7 +35,7 @@ Rule: clone, scan, delete clone after. Keep disk under 90%.
- [x] Pidgin/Finch (C, chat client) — pidgin-0001 CWE-407 499.5x, pidgin-0002 CWE-312
- [ ] HexChat (C, IRC client)
- [ ] Evolution (C, email/calendar)
- [x] Evolution (C, email/calendar) — evolution-0002 MOAD-0001 CWE-407 search_hit_cache g_slist_find_custom O(N^2) 433x at N=1000; MOADs 0002/0003/0004/0005 CLEAN
- [x] Thunderbird (C++, email) — see Priority 1 entry above; 8 defects total
- [x] Calibre (deeper, Python ebook manager) — calibre-0003 MOAD-0001 CWE-407 depth_first flat.index() O(L*F) 38x at F=500; calibre-0004 MOAD-0001 CWE-407 HTMLFile.find_links list dedup O(L^2) 124x at L=1000; MOADs 0002/0003/0004/0005 CLEAN
- [ ] Okular/Evince/Zathura (PDF viewers)

View file

@ -0,0 +1,61 @@
# evolution-0002: Calendar Search Hit Cache Dedup O(N^2) in cal_searching_got_instance_cb
## Target
GNOME Evolution email/calendar client
## Severity
MEDIUM
## MOAD
0001 (CWE-407: Inefficient Algorithmic Complexity)
## Location
`src/modules/calendar/e-cal-shell-view-private.c``cal_searching_got_instance_cb()`
## Description
`cal_searching_got_instance_cb()` is the per-instance callback for Evolution's
calendar "next/previous occurrence" search. It is invoked once per calendar
event instance as `e_cal_client_generate_instances_for_object()` expands
recurring events over a search window.
Inside the callback, each new instance is deduplicated against
`priv->search_hit_cache` (a GSList of `time_t *`) using `g_slist_find_custom()`.
As instances accumulate the list grows, so the I-th instance check scans up to
I-1 entries: O(1) + O(2) + ... + O(N-1) = O(N^2/2) total comparisons.
A search over multiple subscribed calendars each containing long-running weekly
recurring events (e.g., a team standup repeated since 2020) easily reaches
hundreds to thousands of instances per search invocation. Navigation
("next occurrence") triggers this search repeatedly.
## Fix
Add a parallel `GHashTable *search_hit_cache_set` (keyed on `time_t` cast to
`gpointer`) to `ECalShellViewPrivate`. Membership check in
`cal_searching_got_instance_cb()` becomes an O(1) `g_hash_table_contains()`.
Clear and destroy the hash table wherever the GSList is freed.
## Speedup
| N instances | Ops defective | Ops fixed | Ratio |
|-------------|---------------|-----------|-------|
| 50 | 1225 | 50 | 24.5x |
| 100 | 4950 | 100 | 49.5x |
| 200 | 19900 | 200 | 99.5x |
| 500 | 124750 | 500 | 249.5x |
| 1000 | 499500 | 1000 | 499.5x |
## MOAD-0002 (Intertangle)
See evolution-0001 SCAN-NOTES.md — `mail-send-recv.c` static globals
(`send_data`, `glob_ongoing_downsyncs`) are shared mutable state with no mutex.
Architectural finding, not patched here.
## MOAD-0003
CLEAN. No ThreadLocal or GPrivate request-scoped identity leakage beyond
what was identified in evolution-0001 scan.
## MOAD-0004
CLEAN. No credential values (passwords, tokens) logged verbatim. Debug
messages reference password operations by description only.
## MOAD-0005
CLEAN. Evolution uses GLib main loop (single UI thread). No concurrent
cache-get+put race conditions.

View file

@ -0,0 +1,53 @@
# UNDF: TBD
--- a/src/modules/calendar/e-cal-shell-view-private.h
+++ b/src/modules/calendar/e-cal-shell-view-private.h
@@ -113,7 +113,8 @@ struct _ECalShellViewPrivate {
gint search_direction; /* negative value is backward, positive is forward, zero is error; in days */
GSList *search_hit_cache; /* pointers on time_t for matched events */
+ GHashTable *search_hit_cache_set; /* time_t values as gpointer keys for O(1) dedup */
/* Event/Task/Memo transferring */
gpointer transfer_alert; /* weak pointer to EAlert * */
--- a/src/modules/calendar/e-cal-shell-view-private.c
+++ b/src/modules/calendar/e-cal-shell-view-private.c
@@ -878,10 +878,14 @@ cal_searching_got_instance_cb (ICalComponent *icomp,
priv = gid->cal_shell_view->priv;
value = g_new (time_t, 1);
*value = start;
- if (!g_slist_find_custom (priv->search_hit_cache, value, cal_time_t_ptr_compare))
+ if (!priv->search_hit_cache_set)
+ priv->search_hit_cache_set = g_hash_table_new (g_direct_hash, g_direct_equal);
+ if (!g_hash_table_contains (priv->search_hit_cache_set, GINT_TO_POINTER ((gint) start))) {
+ g_hash_table_add (priv->search_hit_cache_set, GINT_TO_POINTER ((gint) start));
priv->search_hit_cache = g_slist_append (priv->search_hit_cache, value);
- else
+ } else {
g_free (value);
+ }
return TRUE;
}
@@ -1271,6 +1275,10 @@ e_cal_shell_view_search_start (ECalShellView *cal_shell_view,
if (priv->search_hit_cache) {
g_slist_free_full (priv->search_hit_cache, g_free);
priv->search_hit_cache = NULL;
}
+ if (priv->search_hit_cache_set) {
+ g_hash_table_destroy (priv->search_hit_cache_set);
+ priv->search_hit_cache_set = NULL;
+ }
cal_iterate_searching (cal_shell_view);
}
@@ -1300,6 +1308,10 @@ e_cal_shell_view_search_stop (ECalShellView *cal_shell_view)
if (priv->search_hit_cache) {
g_slist_free_full (priv->search_hit_cache, g_free);
priv->search_hit_cache = NULL;
}
+ if (priv->search_hit_cache_set) {
+ g_hash_table_destroy (priv->search_hit_cache_set);
+ priv->search_hit_cache_set = NULL;
+ }
priv->search_direction = 0;
}

View file

@ -0,0 +1,130 @@
"""
Unit test for evolution-0002: Calendar search_hit_cache dedup O(N^2) fix.
Simulates cal_searching_got_instance_cb() collecting unique time_t values
into a dedup'd cache. Measures op-count for defective (GSList linear scan)
vs fixed (GHashTable O(1)) implementations.
Asserts speedup > 3x at N=100 and N=1000.
"""
import time
import sys
PASS = True
def collect_instances_defective(instances):
"""
Defective: linear scan for each new instance.
Equivalent to g_slist_find_custom on a growing GSList.
Returns (cache_list, op_count).
"""
cache = []
ops = 0
for val in instances:
# Scan entire cache for duplicate — O(k) per insert
found = False
for c in cache:
ops += 1
if c == val:
found = True
break
if not found:
cache.append(val)
return cache, ops
def collect_instances_fixed(instances):
"""
Fixed: O(1) hash set membership check.
Equivalent to g_hash_table_contains before g_slist_append.
Returns (cache_list, op_count).
"""
cache = []
cache_set = set()
ops = 0
for val in instances:
ops += 1 # One hash lookup per insert
if val not in cache_set:
cache_set.add(val)
cache.append(val)
return cache, ops
def run_test(label, n_instances, expected_speedup):
global PASS
# Generate N unique timestamps (weekly recurring event instances)
# Mix in some duplicates: same event across two calendars = same timestamp
unique = list(range(n_instances))
# 20% duplicates simulating overlapping calendar subscriptions
duplicates = unique[: n_instances // 5]
instances = unique + duplicates
# Shuffle deterministically for reproducibility
# (keep order as-is; order doesn't change big-O result)
_, ops_defective = collect_instances_defective(instances)
cache_fixed, ops_fixed = collect_instances_fixed(instances)
ratio = ops_defective / max(ops_fixed, 1)
status = "PASS" if ratio >= expected_speedup else "FAIL"
if status == "FAIL":
PASS = False
print(
f"[{status}] {label}: N={n_instances} | "
f"defective={ops_defective} ops | fixed={ops_fixed} ops | ratio={ratio:.1f}x "
f"(expected >= {expected_speedup}x)"
)
# Verify correctness: both produce same unique set
cache_defective, _ = collect_instances_defective(instances)
if sorted(cache_defective) != sorted(cache_fixed):
print(f" [FAIL] correctness: cache contents differ!")
PASS = False
else:
print(f" correctness: caches match ({len(cache_fixed)} unique timestamps)")
def run_wall_time_test(label, n_instances):
"""Wall-clock speedup check for large N."""
unique = list(range(n_instances))
duplicates = unique[: n_instances // 5]
instances = unique + duplicates
t0 = time.perf_counter()
collect_instances_defective(instances)
t1 = time.perf_counter()
collect_instances_fixed(instances)
t2 = time.perf_counter()
defective_ms = (t1 - t0) * 1000
fixed_ms = (t2 - t1) * 1000
ratio = defective_ms / max(fixed_ms, 1e-9)
print(
f" wall-clock {label}: defective={defective_ms:.2f}ms fixed={fixed_ms:.2f}ms ratio={ratio:.1f}x"
)
if __name__ == "__main__":
print("evolution-0002: Calendar search_hit_cache dedup O(N^2) fix")
print("=" * 65)
run_test("N=50 op-count", 50, 3.0)
run_test("N=100 op-count", 100, 10.0)
run_test("N=500 op-count", 500, 50.0)
run_test("N=1000 op-count", 1000, 100.0)
print()
run_wall_time_test("N=1000 wall-clock", 1000)
run_wall_time_test("N=5000 wall-clock", 5000)
print()
if PASS:
print("ALL TESTS PASSED")
sys.exit(0)
else:
print("SOME TESTS FAILED")
sys.exit(1)