Use final crawl path in HTML parsing message

This commit is contained in:
I-Al-Istannen 2021-07-03 17:05:48 +02:00
parent 91200f3684
commit 89be07d4d3

View File

@ -247,13 +247,12 @@ instance's greatest bottleneck.
maybe_cl = await self.crawl(path)
if not maybe_cl:
return None
return self._crawl_ilias_page(url, parent, path, maybe_cl)
return self._crawl_ilias_page(url, parent, maybe_cl)
async def _crawl_ilias_page(
self,
url: str,
parent: IliasPageElement,
path: PurePath,
cl: CrawlToken,
) -> None:
elements: List[IliasPageElement] = []
@ -267,7 +266,7 @@ instance's greatest bottleneck.
while next_stage_url:
soup = await self._get_page(next_stage_url)
log.explain_topic(f"Parsing HTML page for {fmt_path(path)}")
log.explain_topic(f"Parsing HTML page for {fmt_path(cl.path)}")
log.explain(f"URL: {next_stage_url}")
page = IliasPage(soup, next_stage_url, current_parent)
if next_element := page.get_next_stage_element():