Add SplitArtifacts=metainfo

[bluca]

GNOME Software parses .metainfo.xml files to provide nice metadata for artifacts it downloads/updates. Follow the spec and add parameter for it, filling the content from os-release.

https://www.freedesktop.org/software/appstream/docs/sect-Metadata-OS.html

[behrmann]

This handles home_url in a single branch and hopefully also addresses @septatrix comment.

Suggested change

	if home_url:
	url = urlparse(home_url)
	if not url.netloc:
	home_url = None
	if context.config.appstream_id:
	id = context.config.appstream_id
	elif home_url:
	netloc = url.netloc.split(".")
	netloc.reverse()
	netloc.remove("www")
	id = ".".join(netloc)
	id += f".{osid}"
	else:
	id = osid
	if context.config.appstream_id:
	id = context.config.appstream_id
	elif home_url:
	url = urlparse(home_url)
	if url.netloc:
	netloc = url.netloc.removeprefix("www.").split(".")
	id = ".".join(reversed(netloc)) + f".{osid}"
	else:
	home_url = None
	id = osid
	else:
	id = osid

[bluca]

it's the same result, doesn't make any difference

[behrmann]

I think having the logic for home_url in a single place is cleaner and removing the leading www. with removeprefix is shorter, and I'd argue cleaner, with the added check whether www is actually in there, that is needed by remove.

That being said, why do you want to strip the www in the first place? Above you said it's not a valid tld or domain and that is true, but neither is the fallback linux and in the reverse url scheme, it would end up at the end, just before osid, anyway. Also, this is only a fallback path, people should set the appstream id in the first place (maybe using specifiers), and singling out www seems a bit odd.

I generally agree that it wouldn't make sense to have www in an appstream ID, but neither would other subdomains like download, as when I grab current systemd from OBS. Can we do without stripping www or what is the logic here that I don't see?

@septatrix list.remove only removes the first occurence, so this also behaves the same. I misremembered this as well.

[septatrix]

Can we do without stripping www or what is the logic here that I don't see?

This is from the spec:

This means the reversed-DNS HOME_URL value from /etc/os-release (without any https/www parts) combined with the ID of the operating system will produce the operating system's component id.

---

@septatrix list.remove only removes the first occurence, so this also behaves the same. I misremembered this as well

No they do not behave the same.
It removes the first occurrence, not the occurrence iff it is at index 0. Therefore a "www" subdomain somewhere in the middle of a URL would also be stripped which I am pretty sure is not how the spec is supposed to be interpreted (because then it would likely also mean removing any "www" component, not only the first occurrence.

[behrmann]

Thanks @septatrix, I really shouldn't look at this before having coffee :)

[behrmann]

LGTM apart from the www stuff.

[behrmann]

Let's shorten it down to this

Suggested change

	if home_url:
	url = urlparse(home_url)
	if not url.netloc:
	home_url = None
	if context.config.appstream_id:
	id = context.config.appstream_id
	elif home_url:
	url = urlparse(home_url)
	netloc = url.netloc.split(".")
	netloc.reverse()
	if "www" in netloc:
	netloc.remove("www")
	id = ".".join(netloc)
	id += f".{osid}"
	else:
	id = osid
	id = osid
	if context.config.appstream_id:
	id = context.config.appstream_id
	elif home_url:
	url = urllib.parse.urlparse(home_url)
	if url.netloc:
	netloc = url.netloc.removeprefix("www").split(".")
	id = f"{'.'.join(reversed(netloc))}.{osid}"
	else:
	home_url = None

It's clearer and looking at the spec the intention seems to be the strip a leading www but not the first. The alternative interpretation would be to remove any www, but that doesn't seem reasonable to me.

[behrmann]

I still think that's pretty much an eyesore, but well, it's XML. Here's one last attempt to make this readable without sigil soup around it

Suggested change

	metainfo = '<?xml version="1.0" encoding="UTF-8"?>\n'
	metainfo += '<component type="operating-system">\n'
	metainfo += f" <id>{id}</id>\n"
	metainfo += f" <name>{name}</name>\n"
	metainfo += f" <summary>{osid} image built with mkosi</summary>\n"
	metainfo += f" <description><p>{description}</p></description>\n"
	if home_url:
	metainfo += f' <url type="homepage">{home_url}</url>\n'
	if icon:
	metainfo += f' <icon type="remote">{icon}</icon>\n'
	metainfo += " <metadata_license>FSFAP</metadata_license>\n"
	metainfo += " <releases>\n"
	metainfo += (
	f' <release version="{context.config.image_version}" date="{timestamp}" type="development">\n'
	)
	metainfo += " <description></description>\n"
	metainfo += " </release>\n"
	metainfo += " </releases>\n"
	metainfo += ' <bundle type="systemd" class="sysupdate"></bundle>\n'
	metainfo += "</component>\n"
	metainfo = textwrap.dedent(
	f"""\
	<?xml version="1.0" encoding="UTF-8"?>
	<component type="operating-system">
	<id>{id}</id>
	<name>{name}</name>
	<summary>{osid} image built with mkosi</summary>
	<description><p>{description}</p></description>
	{f'<url type="homepage">{home_url}</url>' if home_url else ""}
	{f'<icon type="remote">{icon}</icon>' if icon else ""}
	<metadata_license>FSFAP</metadata_license>
	<releases>
	<release version="{context.config.image_version}" date="{timestamp}" type="development">
	<description></description>
	</release>
	</releases>
	<bundle type="systemd" class="sysupdate"></bundle>
	</component>
	"""
	)

The only downside is, that the optional things home_url and icon become empty lines when unset. I think that's not unreasonable, since in XML it's syntactically insignificant whitespace.