Infrastructure » Websites

In some cases, the same crash signature is collected in both Windows and Linux, but the reports are collected in distinct pages because of a minor whitespace difference.

• "IsAnchoredObjShown(SwTextFrame const &,SwFormatAnchor const &)" on Windows

• Sample report: https://crashreport.libreoffice.org/stats/crash_details/841cd041-a54e-4abc-a17e-96ef9395af2e
• Linked to Bugzilla report: https://bugs.documentfoundation.org/show_bug.cgi?id=159967
• Collected into https://crashreport.libreoffice.org/stats/signature/IsAnchoredObjShown(SwTextFrame%20const%20&,SwFormatAnchor%20const%20&)

• "IsAnchoredObjShown(SwTextFrame const&, SwFormatAnchor const&)" on Linux (notice the difference: spaces missing before "&")

• Sample report: https://crashreport.libreoffice.org/stats/crash_details/6a21d8fe-6142-4d21-aff5-c1b09ca1f0b3
• Linked to Bugzilla report: https://bugs.documentfoundation.org/show_bug.cgi?id=159379
• Collected into https://crashreport.libreoffice.org/stats/signature/IsAnchoredObjShown(SwTextFrame%20const&,%20SwFormatAnchor%20const&)

Another example in this contributor's comment https://bugs.documentfoundation.org/show_bug.cgi?id=158343#c4 (no space before the asterisk on Linux)

This leads to wasted effort:
- users opening a new Bugzilla ticket because they don't see an existing report linked to the crash signature
- QA having to triage more reports
- QA and devs thinking the issue is Linux-only or Windows-only
- marking a ticket as fixed while others are lingering and might end up eventually closed as "worksforme"
- not having a clear picture of how widespread the issue actually is, and therefore not assigning the correct importance to the bug

Cloph mentioned this is related to how each OS collects the trace, so I am unsure what the best solution is (processing the actual trace, or later merging the pages conditionally). Copying Cloph and Xisco in for advice.