Querying for html parameter

[Racro]

I get the following error when I query with HTML for www.google.com. I generate the graphml using brave nightly binary and pagegraph-crawl.

[FHantke]

@pes10k I checked into this issue. The problem arises because the security origin is calculated for all contexts, including those without a parent node. This leads to the assertion being triggered when the parent_domroot_node function is called.

To prevent this, adding the following check before the parent_domroot_node call would resolve the issue:

if self.is_top_level_domroot():
  return None

I'm unsure about the best approach for the fix. Should we keep the assertion as is and simply add the snippet above, or should parent_domroot_node return a specific error value that can be handled more explicitly?

Cheers

[pes10k]

@Racro can you confirm you're using an up-to-date version of pagegraph-query and brave-nightly?

@FHantke i think thats a fine bandaid fix. I think tbh probably the right thing for me to do is to record the frame's security-origin brave/pagegraph side, instead of re-caculating it from the graph (even though those things should end up the same). Let me add this to my TODO for this week.

In the meantime though i think the correct thing to do is probably to fix the html query in pagegraph-query since we likely dont mean to be calculating information about the HTML elements in a frame w/o a security origin (since those frames should be either those created in isolated origins, or as initialization-phase frames that the page can't interact with). And so i'd prefer to fix things by omitting such frames from the query all together.

[pes10k]

@Racro just to compare, i just did the following and the report generates correctly with me (this is using the 0.9.5 tagged version of pagegraph-query)

> pagegraph-crawl@1.2.1 crawl
> node ./built/run.js --url https://www.google.com -o /tmp/output -t 10

DEBUG:Received arguments: Namespace(binary=undefined, recursive_depth=1, output='/tmp/output', url='https://www.google.com', existing_user_data_dir=undefined, persist_user_data_dir=false, extensions_path=undefined, shields='down', secs=10, logging='info', interactive=false, user_agent=undefined, proxy_server=undefined, extra_args=undefined, crawl_duplicates=false, screenshot=false, store_har=false, store_har_body=false, compress=false)
DEBUG:Running with settings: {"executablePath":"/Applications/Brave Browser Nightly.app/Contents/MacOS/Brave Browser Nightly","outputPath":"/tmp/output","url":"https://www.google.com/","recursiveDepth":1,"seconds":10,"withShieldsUp":false,"loggingLevel":"info","stealth":false,"interactive":false,"crawlDuplicates":false,"screenshot":false,"storeHar":false,"storeHarBody":false,"compress":false}
DEBUG:Starting crawl with URL: https://www.google.com/ and with previously seen urls: []
DEBUG:Profile exists at /var/folders/th/0d_r1l6j0pz5xr2vlwjmgkj00000gn/T/pagegraph-profile-brmINF, so deleting.
DEBUG:Running on darwin, Xvfb not supported
DEBUG:Closing 2 pages that are already open.
DEBUG:  - closing tab with url chrome://newtab/
DEBUG:  - closing tab with url about:blank
DEBUG:Navigating to https://www.google.com/
DEBUG:Loading https://www.google.com/ bc it is the first top frame page load
DEBUG:Loaded https://www.google.com/
DEBUG:Waiting for 10s
DEBUG:calling generatePageGraph
DEBUG:generatePageGraph { size: 5170291 }
DEBUG:Writing PageGraph file to: /tmp/output.graphml
DEBUG:Closing page
DEBUG:Closing the browser

<CHANGE TERMIMAL>

(pagegraph-query) ➜ ~/Code/pagegraph-query/pagegraph-query: (main) ✗ ./run.py html /tmp/output.graphml

And I get the below report: google-html-report.json

Either way, no argument here, i think the issue you're seeing suggests a useful fix to pagegraph itself, but just wanted to see if an update might unblock you in the meantime

[FHantke]

@pes10k Interesting that it works on your side. I just cloned the latest version of both crawl and query, and get the same error behavior. I thought it might be the Brave nightly version, but I'm also using the latest, 1.78.43.

> pagegraph-crawl@1.2.1 crawl
> node ./built/run.js --url https://www.google.com -o /tmp/output -t 10

DEBUG:Received arguments: Namespace(binary=undefined, recursive_depth=1, output='/tmp/output', url='https://www.google.com', existing_user_data_dir=undefined, persist_user_data_dir=false, extensions_path=undefined, shields='down', secs=10, logging='info', interactive=false, user_agent=undefined, proxy_server=undefined, extra_args=undefined, crawl_duplicates=false, screenshot=false, store_har=false, store_har_body=false, compress=false)
DEBUG:Running with settings: {"executablePath":"/Applications/Brave Browser Nightly.app/Contents/MacOS/Brave Browser Nightly","outputPath":"/tmp/output","url":"https://www.google.com/","recursiveDepth":1,"seconds":10,"withShieldsUp":false,"loggingLevel":"info","stealth":false,"interactive":false,"crawlDuplicates":false,"screenshot":false,"storeHar":false,"storeHarBody":false,"compress":false}
DEBUG:Starting crawl with URL: https://www.google.com/ and with previously seen urls: []
DEBUG:Profile exists at /var/folders/j0/1ldfhh6x2td_hvlwmzg6h_1m0000gn/T/pagegraph-profile-trYdbf, so deleting.
DEBUG:Running on darwin, Xvfb not supported
DEBUG:Closing 2 pages that are already open.
DEBUG:  - closing tab with url chrome://newtab/
DEBUG:  - closing tab with url about:blank
DEBUG:Navigating to https://www.google.com/
DEBUG:Loading https://www.google.com/ bc it is the first top frame page load
DEBUG:Loaded https://www.google.com/
DEBUG:Waiting for 10s
DEBUG:calling generatePageGraph
DEBUG:generatePageGraph { size: 4519703 }
DEBUG:Writing PageGraph file to: /tmp/output.graphml
DEBUG:Closing page
DEBUG:Closing the browser

(.venv) florianhantke@Florians-MacBook-Pro pagegraph-query % python3 run.py html /tmp/output.graphml                           
Traceback (most recent call last):
  File "/private/tmp/pagegraph-query/run.py", line 250, in <module>
    RESULT = command.execute()
  File "/private/tmp/pagegraph-query/pagegraph/commands/html.py", line 57, in execute
    reports.append(node.to_report())
                   ~~~~~~~~~~~~~~^^
  File "/private/tmp/pagegraph-query/pagegraph/graph/node/dom_root.py", line 70, in to_report
    self.url(), self.security_origin(), self.blink_id())
                ~~~~~~~~~~~~~~~~~~~~^^
  File "/private/tmp/pagegraph-query/pagegraph/graph/node/dom_root.py", line 81, in security_origin
    if parent_domroot_node := self.parent_domroot_node():
                              ~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/private/tmp/pagegraph-query/pagegraph/graph/node/dom_root.py", line 119, in parent_domroot_node
    assert not self.is_top_level_domroot()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError

If you want to reproduce it from the crawl output, you can find the pagegraph here (.txt only to upload it to Github).
output.graphml.txt

Anyways, as you said, the right thing to do is probably to record the frame's security origin in pagegraph itself.

[Racro]

@pes10k I cannot add anything more than what @FHantke suggested but I was curious if it could be the linux environment by any chance? I think your crawl logs seem pretty much the same as @FHantke also posted.

npm run crawl -- -u https://www.google.com -o control/ -b /usr/bin/brave-browser-nightly --screenshot -t 10
This was my command on ubuntu 22.04.

[pes10k]

fwiw, i merged the change to explicitly include security origin attributes in DOMRootNoe items in the PageGraph recordings, and so will push the pagegraph-query today or tomorrow too

[pes10k]

@Racro i just tried pushing an update for pagegraph-query. Please try with HEAD / the new version 0.9.6

For new graphs (those taken with the PageGraph changes I just pushed to BraveNightly) pagegraph-query uses the explicit security origin recorded into pagegraph files.

For older graphs (those recorded with older versions of Brave) pagegraph-query uses the more permissive graph parsing code @FHantke suggested

[pes10k]

and if you're still seeing problems @Racro , can you share an example graph thats causing trouble?