Have you done a speed comparison between low-memory and normal mode? or are you not able to due to the limitations of your host?

@blacktop Yeah unfortunately I cannot perform normal mode test. My machine just doesn't have enough RAM.

This PR is stale because it has been open 45 days with no activity. Remove stale label or comment or this will be closed in 10 days.

Hi, have you got a chance to look at this further? @blacktop

I reviewed this and found two issues we should fix before merge:

1. Low-memory matching edge case can misclassify files as New

In --low-memory mode we use one map both for “exists in old” and “still unmatched old”.
Trigger: if the same relative path is encountered more than once during the new-side walk (for example via duplicate mount-relative paths/symlink traversal), the first match deletes the key, and later occurrences of that same path are treated as New even though they existed in old.
Impact: false positives in New/Removed classification.

(though I think this might be a pre-existing issue?)

2. DSC diff fidelity regression due to dropped local/private symbol merge

The refactor no longer calls ParseLocalSymbols(false) and no longer appends local symbols to the Mach-O symtab before GenerateDiffInfo.
Trigger: any dyld shared cache image where useful symbol identity comes from local/private symbols (common when exported symbols are sparse/obfuscated).
Impact: weaker symbol/function naming in the diff output and potentially noisier/less actionable DSC diffs vs current master.

Right, thanks for pointing out.

Low-memory matching edge case can misclassify files as New

I changed the low-memory path to track match state instead of deleting keys (map switched from map[string]struct{} -> map[string]bool, set to true once matched). This prevents later duplicate occurrences of the same new-side path from being classified as New.

DSC diff fidelity regression due to dropped local/private symbol merge

I added back ParseLocalSymbols(false) call.