Bug 7798

Summary: Intelligent matching of new error report to existing reports
Product: [Yocto Project Subprojects] Error Reporting tool Reporter: Paul Eggleton <bluelightning>
Component: Error Reporting ToolAssignee: Michael Wood <michael.g.wood>
Status: VERIFIED FIXED QA Contact: Bogdan Alexandru Voiculescu <bogdanx.a.voiculescu>
Severity: enhancement    
Priority: Medium CC: bogdanx.a.voiculescu, randy.macleod
Version: unspecified   
Target Milestone: 1.9 M1   
Hardware: All   
OS: Multiple   
Whiteboard:
OS type for building Yocto: --- Type of Regression: ---
Verified: Documentation change: Don't know

Description Paul Eggleton 2015-05-21 09:47:51 UTC
It would be useful to recognise when a new error report is similar to or represents the same underlying issue as a report that already exists in the database. Users should still be able to see all instances of a submitted error. We could start small, by showing the number of instances for a certain error and link to a list of such instances.
Comment 1 Randy MacLeod 2015-05-21 14:57:38 UTC
Michael, 

If you have some ideas on how to implement this feature, please add your notes here since I hear that you are busy with toaster. I might find someone at Wind River to work on it depending on the complexity.
Comment 2 Michael Wood 2015-05-26 17:54:16 UTC
(In reply to comment #1)
> Michael, 
> 
> If you have some ideas on how to implement this feature, please add your
> notes here since I hear that you are busy with toaster. I might find someone
> at Wind River to work on it depending on the complexity.

Currently I've got this working in a local branch using a Levenshtien distance on a prime error (the first in the database).

So the flow goes like this:

New data comes in -> Calculate levenshtien distance from the first error in the database  -> store that in the record for this error.

Then when you wish to find existing similar reports we just query the column for the Levenshtien distance where it's = this report's distance +/- some tolerance value.

This means that the lookup is trivial which is the bit that will happen very frequently. The problem I have is that calculating the distance can take seconds to several minutes on standard hardware, as the size of the log being compared varies greatly. This means we wouldn't be able to send immediate feedback on a duplicates/similar errors existing. 

I'm not sure if that's important or not, I think at some point we're going to have to do a fair bit of computation so it's just a matter of where in the flow that is acceptable.
Comment 3 Michael Wood 2015-06-05 17:06:49 UTC
This feature has now been added and available as of

69ee069a6a49d4ed9a37d22fdb7600d7f1555be2
Comment 4 Bogdan Alexandru Voiculescu 2015-09-07 11:52:46 UTC
feature implemented