Github Actions search is not reliable

As an example, go to https://github.com/vadi2/vcpkg-lua-reproduction/runs/1523275802?check_suite_focus=true and search for lua.exe. Then download the raw logs and search for the same.

Actions search does not find anything for me, which makes it untrustworthy :frowning:

As far as I can tell, it’s reliable… but it only returns results in the log sections that have been expanded.

Often that’s inconvenient, especially with the lack of an “expand all” option to open all the collapsed sections of the log.

EDIT: I just remembered one thing that’s unreliable - the viewport scrolling. The number of results found is correct, but sometimes the viewport fails to scroll to the highlighted result.

I don’t think it only returns log sections that have been expanded, because it does auto-expansion by itself. Are we talking about the same thing?

It seems that the search tool only auto-expands sections that were previously expanded but then collapsed. The log page only loads the content of sections once they’ve been expanded, and the search only finds results in the loaded content. If you F5 refresh the log page and search without expanding anything then there will be no results.

Going to Show contents of github workspace · vadi2/vcpkg-lua-reproduction@5373a71 · GitHub and searching for lua.exe gives me no results, as in your screenshot above. Going to that same page, opening and closing the Run Get-ChildItem -Path $env:GITHUB_WORKSPACE -Recurse section and then searching gives me this:

In addition, expanding categories after entering a search doesn’t update the search results. Pressing enter in the search box doesn’t help either; you have to delete or type something in the search box before it will pick up new results from the expanded section.

It looks like it could use with some improvements, because it is not intuitive to use and not very trustworthy either. I always end up looking at the raw logs instead - which is a worse experience than Travis or Appveyor, where we migrated from.

1 Like