-
Improvement
-
Resolution: Fixed
-
Major
-
None
If the warnings scanner is also active on failed builds (for example to get a quick overview of compiler errors) the trendgraph gets "down" dependent on where on the build progress the build failed.
It would be nice to have an Option, which allows the parsing of warnings on failed builds, but exclude this data for threndgraph and calculation of "new" warnings.
Maybe an option like: [ ] only show high priority warnings on failed builds
or something like this.
- is related to
-
JENKINS-14867 Warnings plugin Trends graph does not include failed builds except for the last one
-
- Resolved
-
[JENKINS-10682] Option for Trendgraph to ignore failed builds
Yes, I agree that failed builds should always be excluded from builds. The Problem is, that currently there are only two option:
1) Don't run the Warningsblugin on failed build (so disable the "always run this plugin" option in the advanced options
2) Enable the Option and let it run regardles of the failure of the build.
With Option 1 everything with the trendgraph/new warnings is okay, except that when a build fails you manually have to scan the console output to see what is going wrong, thats why I use option 2, this creates "High Priority Warnings" for compile errors that could be easily reviewed (and fixed of course 
.
The problem with option 1 is the following:
Assume a build with 100 Sourcefiles each one generates a warning so if everything is fine it says 100 warnings, 0 new, 0 fixed.
When the build fails, lets say after the 20th file it says 21 Warnings (20 'normal' one 'high') 1 new, 80 fixed.
When the build gets okay again it compiles all files and say 100 Warnings, 80 new 1 fixed.
The Trendgraph then has a "peak" down and up for this broken build als well as the new/fixed ration is missleading.
So what would be nice is a combination of 1+2 which parses (and shows) the warnings on a failed so they could be reviewed, but behaves like this build never happens on the trendgraph/difference calculation.
I hope this makes things a little more clean, if prefereable I can attatch some screenshots but they would be in german if thats okay.
Code changed in jenkins
User: Ulli Hafner
Path:
src/main/java/hudson/plugins/analysis/core/BuildHistory.java
src/test/java/hudson/plugins/analysis/core/BuildHistoryTest.java
http://jenkins-ci.org/commit/analysis-core-plugin/96ad36e7c6c6f06de201204131aa150785b4f52d
Log:
[FIXED JENKINS-10682] Ignore failed (or aborted) builds when computing
the history information. I.e., failed builds will not be considered
during new warnings evaluation and in trend graphs.
Do you think it makes sense to have an option for this behaviour? Shouldn't failed build be always excluded from trend graphs?
I'm not sure why this should influence the new warnings? Can you elaborate this part somewhat?