| Summary: | A number of "performance test links" need fixing or improvements | ||
|---|---|---|---|
| Product: | [Eclipse Project] Platform | Reporter: | David Williams <david_williams> |
| Component: | Releng | Assignee: | David Williams <david_williams> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | P3 | ||
| Version: | 4.4 | ||
| Target Milestone: | 4.5 M4 | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Bug Depends on: | |||
| Bug Blocks: | 374441 | ||
|
Description
David Williams
another link, is the "back to global results" link. For example, if I drill down into "jface" results, http://download.eclipse.org/eclipse/downloads/drops4/N20141203-2000/performance/org.eclipse.jface.php?fp_type=0 It's "back to global results" link points to http://download.eclipse.org/eclipse/downloads/drops4/N20141203-2000/performance/performance.php?fp_type=0 instead of http://download.eclipse.org/eclipse/downloads/drops4/N20141203-2000/performance.php?fp_type=0 Another place that need improvement is that when we run the "short set" of tests (such as for "nightly" builds) then there are *always* links for org.eclipse.jdt.core* org.eclipse.jdt.text* even though they have no results for "short tests". In other places in code, we look for the existence of some file, before displaying a link to it. In this case, we might still want to display something ... to signify they were not expected, not just "failed to run", but, a "link to no data" is very confusing. Definitely looks like something is wrong. I think (just now) have fixed these in Git. Some still broken in initial N20141206-2000 (and I20141207-2000 once tests complete) that caused performance page to not display ... fixed by N-build by hand, and will fix I-build by hand once tests are done (else, hand fixes will be over written). But I think 1208-0800 build should be correct "out of the box". Will re-open if I find spots I missed. FWIW, this fix also included general "clean up" and XHTML compliance of "log files". Also, if the three "log file displays" that we have (normal unit tests, unit tests from performance run, and unit tests from baseline run) I "factored out" a sizable chunck of common PHP code into "logPhpUtils.php". (In reply to David Williams from comment #0) > FWIW, I think the original assumption, years ago, was that the "main page", > "perforamnce.php" was in same directory as performance results > .../performance, whereas now, it is "one up" from performance directory. > > I've looked at moving performance.php "down" a level, but seems to me to be > about the same amount of work ... but, probably deserve closer insspection. > Should have mentioned ... the main part of the fix WAS to put performance.php down a level, into .../performance/ It did turn out to be fewer changes, but mostly just "cleaner". Well, that was a fast "reopening" but found a problem already. On the main performance test page, I put the "junits" from that run (to see "pass/fail" table). And, the links in there are wrong. They "point to" results in /.../performance/performance/... (which, are not there) instead of just .../performance/.... (In reply to David Williams from comment #5) > Well, that was a fast "reopening" but found a problem already. > > On the main performance test page, I put the "junits" from that run (to see > "pass/fail" table). > > And, the links in there are wrong. They "point to" results in > /.../performance/performance/... (which, are not there) instead of just > .../performance/.... I think I can fix this "junit table" problem by passing in "." to our custom ant task that computes the summary tables ... and am trying a test build with that now. Not sure I mentioned this before, but another place of error was that 'console logs' had to link "one up" in the hierarchy to find the right file ... not sure that will work quite right ... am investigating now. Similarly, the "baseline" junit results were not being displayed/linked to since it also has to point "one up" in the directory structure. That seems to work ok. Using the "." value appeared to work fine in my local test run, so committed that change ... in time for the Tuesday 11:15 I-build. (along with other, "less risky" ones). |