mirror of
https://github.com/elicpeter/nyx.git
synced 2026-06-21 20:18:06 +02:00
23 lines
1.1 KiB
Markdown
23 lines
1.1 KiB
Markdown
|
# cross_file_alias_bounded_graph
|
||
|
|
|
||
|
|
## Purpose
|
||
|
|
Stability test for the bounded alias-graph promotion path. The fixture crosswires
|
||
|
|
five aliased containers so that the helper overflows the dense-alias budget and
|
||
|
|
falls back to the conservative "any arg taints any other arg" rule. The primary
|
||
|
|
guarantee is that the scanner **terminates** under overflow, not that a specific
|
||
|
|
finding fires.
|
||
|
|
|
||
|
|
## Expectations
|
||
|
|
- **required**: (none — terminating scan is the assertion)
|
||
|
|
- **forbidden**: (none — overflow promotion is conservative over-approximation;
|
||
|
|
forbidden_findings would contradict the whole point of the fixture)
|
||
|
|
- **noise_budget**: max_total=20, max_high=10
|
||
|
|
|
||
|
|
## Why `noise_budget` stays
|
||
|
|
Because the dense-alias promotion deliberately over-approximates, the exact set
|
||
|
|
of findings can shift when the budget is tuned. The loose upper bound catches
|
||
|
|
any regression that explodes finding counts (a real failure mode if the
|
||
|
|
promotion logic is broken) without requiring lockstep alignment with whatever
|
||
|
|
happens to fire today. Dropping `noise_budget` here would remove the only
|
||
|
|
upper-bound check this fixture has.
|