diff --git a/.github/workflows/update.yaml b/.github/workflows/update.yaml
index 7c179e3..d15119e 100644
--- a/.github/workflows/update.yaml
+++ b/.github/workflows/update.yaml
@@ -14,6 +14,9 @@ jobs:
       with:
         ref: ${{ github.head_ref }}
         submodules: true
+    # pypy is much faster at generating lots of JSON
+    # and canonicaljson also uses pypy-json instead 
+    # of simplejson (https://github.com/matrix-org/python-canonicaljson/pull/25/files)
     - uses: actions/setup-python@v4
       with:
         python-version: 'pypy3.9'
diff --git a/generate.py b/generate.py
index 4844567..9e831ef 100644
--- a/generate.py
+++ b/generate.py
@@ -3,6 +3,7 @@ from glob import glob
 import subprocess
 import markdown
 import json
+import canonicaljson
 import urllib.request
 from datetime import datetime
 import copy
@@ -52,12 +53,14 @@ def get_osv(cve_data_all_versions):
         def cve_fixed_version(package, cves, os_version, advisory):
             # list of fixed versions with a matching
             # CVE/pkg/OS combination
-            fixed_versions = set([
-                x["res_ver"]
-                for cve in cves
-                for x in cve_data_all_versions.get(cve, list())
-                if (x and x["os"] == os_version and x["pkg"] == package)
-            ])
+            fixed_versions = set(
+                [
+                    x["res_ver"]
+                    for cve in cves
+                    for x in cve_data_all_versions.get(cve, list())
+                    if (x and x["os"] == os_version and x["pkg"] == package)
+                ]
+            )
             # There should only be a single such reference
             if len(fixed_versions) != 1:
                 f = ", ".join(list(fixed_versions))
@@ -96,7 +99,7 @@ def get_osv(cve_data_all_versions):
                                 {"introduced": "0"},
                                 {"fixed": fixed_version},
                             ],
-                            "type": "ECOSYSTEM"
+                            "type": "ECOSYSTEM",
                         }
                     return r
 
@@ -105,10 +108,7 @@ def get_osv(cve_data_all_versions):
                     "modified": modified.isoformat("T") + "Z",
                     "published": published.isoformat("T") + "Z",
                     "related": cves,
-                    "affected": [
-                        affected(pkg, cves, os_version)
-                        for pkg in packages
-                    ],
+                    "affected": [affected(pkg, cves, os_version) for pkg in packages],
                     "references": [
                         {"type": "ADVISORY", "url": ADVISORY_URL.format(slug=slug)}
                     ],
@@ -116,18 +116,24 @@ def get_osv(cve_data_all_versions):
 
 
 def merge_advisories(advisory_file, data):
+
+    def dedup_dicts(items):
+        dedupped = [ json.loads(i) for i in set(canonicaljson.encode_canonical_json(item) for item in items)]
+        return dedupped
     # read the current advisory data as json
     with open(advisory_file, "r") as f:
         original = json.load(f)
         current = copy.deepcopy(original)
     # merge the data
     assert current["id"] == data["id"]
-    current["affected"].extend(data["affected"])
-    current["references"].extend(data["references"])
-    current["related"].extend(data["related"])
-
-    # Make sure no CVE references are duplicated
-    current["related"] = list(set(current["related"])).sort()
+    # Add any new data, but use a set, to avoid
+    # duplicate entries
+    for key in ['affected', 'references', 'related']:
+        if current[key]:
+            current[key].extend(data[key])
+            current[key] = dedup_dicts(current[key])
+        elif data[key]:
+            current[key] = data[key]
 
     # Pick the earlier published date
     # and the later modified date
@@ -149,9 +155,13 @@ def merge_advisories(advisory_file, data):
 
     no_important_changes = True
     # One of the important keys has changed
-    for key in ["id", "affected", "references", "related", "published"]:
-        if current[key] != original[key]:
+    for key in ["affected", "references", "related", "published"]:
+        if canonicaljson.encode_canonical_json(
+            original[key]
+        ) != canonicaljson.encode_canonical_json(current[key]):
+            print(f"Found changes in {current['id']} / {key}")
             no_important_changes = False
+            break
 
     if no_important_changes:
         return None
@@ -182,19 +192,25 @@ def fetch_cve_metadata(PHOTON_VERSIONS):
             print(f"[+] CVE metadata for Photon OS {branch}.0: Added {len(data)} CVEs")
     return cve_metadata
 
-def __main__():
+
+def __main__(advisory_id = None):
     cve_metadata = fetch_cve_metadata(PHOTON_VERSIONS)
 
-    for advisory in glob("advisories/*.json"):
-        os.remove(advisory)
     for d in get_osv(cve_metadata):
+        # If we are only running for a single advisory
+        # Check and continue if it doesn't match
+        if advisory_id and d['id'] != advisory_id:
+            continue
         fn = f"advisories/{d['id']}.json"
         if os.path.exists(fn):
             d = merge_advisories(fn, d)
         if d:
-            with open(fn, "w") as f:
-                f.write(json.dumps(d, indent=4, sort_keys=True))
+            with open(fn, "wb") as f:
+                f.write(canonicaljson.encode_pretty_printed_json(d))
 
 
 if __name__ == "__main__":
-    __main__()
+    if len(sys.argv) >=2:
+        __main__(sys.argv[1])
+    else:
+        __main__()