Install URLs: X.zip.md5?md5=<hex> downloads X.zip directly

CLAUDE.md

@@ -35,7 +35,7 @@ python3 -I tools/i18n.py check             # language files: unused strings, los
 
 pugwash flags: `--quiet --no-check --debug --no-colour --force-colour --no-log --offline --no-harbour`.
 
-To test on a device: `./do_release.sh test`, copy `dist/PortMaster.zip` over (stream it over ssh: `scp` has truncated files on some devices), then on the device run `./PortMaster/harbourmaster --no-check install ./PortMaster.zip` (the `./` prefix makes it install a local file). `harbourmaster install` also takes a URL: `<url>/X.zip.md5` fetches the md5 file and then `X.zip`, and `<url>/X.zip?md5=<hex>` checks the download against that md5 (the `md5` parameter isn't sent to the server); both must agree when given together (`raw_download` in `source/__init__.py`).
+To test on a device: `./do_release.sh test`, copy `dist/PortMaster.zip` over (stream it over ssh: `scp` has truncated files on some devices), then on the device run `./PortMaster/harbourmaster --no-check install ./PortMaster.zip` (the `./` prefix makes it install a local file). `harbourmaster install` also takes a URL: `<url>/X.zip.md5` fetches the md5 file and then `X.zip`, and `<url>/X.zip?md5=<hex>` checks the download against that md5 (the `md5` parameter isn't sent to the server). Given together, `<url>/X.zip.md5?md5=<hex>` downloads `X.zip` straight away and checks it against the link's md5, without fetching the md5 file; older PortMasters fetch the md5 file and then `X.zip`, so DeviceHub sends this form to every version (`raw_download` in `source/__init__.py`).
 
 ## Local-dev mode (HM_TESTING)
 

PortMaster/pylibs/harbourmaster/source/__init__.py

@@ -52,7 +52,13 @@ def raw_download(save_path, file_url, callback=None, file_name=None, md5_source=
     if file_name is None:
         file_name = url_info.path.rsplit('/', 1)[-1]
 
-    if file_name.endswith('.md5') or file_name.endswith('.md5sum'):
+    if (file_name.endswith('.md5') or file_name.endswith('.md5sum')) and md5_query is not None:
+        ## "<url>/X.zip.md5?md5=<hex>": the md5 is right here, so the .md5 file isn't fetched; X.zip is.
+        ## (Older PortMasters fetch the .md5 file and then X.zip, so the one link works for both.)
+        file_name = file_name.rsplit('.', 1)[0]
+        file_url = urlunparse(url_info._replace(path=url_info.path.rsplit('.', 1)[0]))
+
+    elif file_name.endswith('.md5') or file_name.endswith('.md5sum'):
         ## If it is an md5 file, we assume the actual zip is sans the md5/md5sum
         md5_source = net.fetch_text(file_url)
         if md5_source is None:
@@ -63,14 +69,6 @@ def raw_download(save_path, file_url, callback=None, file_name=None, md5_source=
 
         md5_source = md5_source.strip().split(None, 1)[0].lower() if md5_source.strip() else ""
 
-        if md5_query is not None and md5_source != md5_query:
-            ## Two answers to "what should it be": neither can be trusted.
-            if callback is not None:
-                callback.message_box(_("Download validation failed."))
-
-            logger.error(f"Unable to download file: {file_url!r} [md5 file {md5_source!r} != url md5 {md5_query!r}]")
-            return None
-
         file_name = file_name.rsplit('.', 1)[0]
         file_url = urlunparse(url_info._replace(path=url_info.path.rsplit('.', 1)[0]))
 

tests/test_source.py

@@ -235,17 +235,21 @@ def test_raw_download_md5_file(tmp_path, callback, served):
 
 
 def test_raw_download_md5_file_and_query(tmp_path, callback, served):
+    ## The md5 is in the link, so the .md5 file isn't fetched: the zip is downloaded and checked.
+    ## (DeviceHub sends these: older PortMasters fetch the .md5 file, then the zip, from the same link.)
     url = f"https://example.com/port.zip.md5?md5={served.md5}"
-    assert hm_source.raw_download(tmp_path, url, callback=callback) is not None
-    assert served.got == ["https://example.com/port.zip"]
+    result = hm_source.raw_download(tmp_path, url, callback=callback)
+    assert result is not None and result['name'] == "port.zip"
+    assert served.fetched == [] and served.got == ["https://example.com/port.zip"]
 
 
-def test_raw_download_md5_file_and_query_disagree(tmp_path, callback, served):
-    ## Which one is right? Neither is trusted: nothing is downloaded.
+def test_raw_download_md5_file_and_wrong_query(tmp_path, callback, served):
+    ## The link's md5 is what the download is checked against.
     result = hm_source.raw_download(tmp_path, f"https://example.com/port.zip.md5?md5={'0' * 32}", callback=callback)
 
-    assert result is None and served.got == []
+    assert result is None and served.fetched == []
     assert callback.message_boxes == ["Download validation failed."]
+    assert not (tmp_path / "port.zip").exists()
 
 
 def test_raw_download_no_md5(tmp_path, callback, served):