Bugzilla – Full Text Bug Listing
| Summary: | change platform names to contain accurate "bits" and VM Levels | ||
|---|---|---|---|
| Product: | [Eclipse Project] Platform | Reporter: | David Williams <david_williams> |
| Component: | Releng | Assignee: | Platform-Releng-Inbox <platform-releng-inbox> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | P2 | CC: | akurtakov, daniel_megert, markus.kell.r |
| Version: | 4.2 | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Bug Depends on: | 431143 | ||
| Bug Blocks: | 374441, 425071, 454921 | ||
Ok, bad news, I've discovered the list of permissible "configs" is hard coded in
TestResultsGenerator class.
To change values (or, improve so not hard coded) would take a "build" of org.eclipse.build.tools and that updated in baseBuilder. That is, not worth is :)
(at least now).
For future improvement, see line 137 of
//The four configurations, add new configurations to test results here + update
// testResults.php.template for changes
private String[] testsConfig = { "linux.gtk.x86_6.0.xml", //
"macosx.cocoa.x86_5.0.xml", //
"win32.win32.x86_7.0.xml" };
So, will revert for now the script/property values, and fix this once we are building org.eclipse.build.tools independent of basebuilder.
For future reference, for step "b", the overall console log is piped in testAll.sh/bat such as
./runtests.sh -os linux -ws gtk -arch $eclipseArch -vm "${vmcmd}" -properties ${propertyFile} $* > results/consolelogs/linux.gtk.x86_6.0_consolelog.txt
Hmm, I think this might be work a rebuild on the buildTools.jar after all. While starting to use the 64 bit mac, its VM also changed, from 5 to 6. I had, not know about _this_ bug at the time, changed its platform code to be macosx.cocoa.x86_64_6.0 They ran last night, but are not "summarized" since tool is looking for macosx.cocoa.x86_5.0 Its one thing to be inaccurate on the "bits" ... but ... VM version seems worse? I'll see if there's some "quick and easy" way to fix this up in a safe way. If not ... move back to the inaccurate name of macosx.cocoa.x86_5.0 for tonight's N build. *** Bug 427589 has been marked as a duplicate of this bug. *** This was improved while working on bug 473904. But, still not completely handled by variables ... still some "hard coded" values. A. testConfigs.php is checked into branch, in the directory ...tychoeclipsebuilder/eclipse/publishingFiles/staticDropFiles This is used for the "labels" in the test results table, and could easily be written by our test framework, where we compute the values, and in fact write similar property=value php files (the difference here is that "multiple tests" would need to write to one file ... or, change the testResults.php to "include testConfig* files (if that's even possible). The order is significant, controlling the order left to right the summary columns are created. B. the "indexer" had to be changed to accept a "testsConfigExpected" variable ... comma separated list of the "configs" which is how the indexer selects the xml files to summarize (the xml files end with the "config-value.xml"). Again, could be written to a property file ... but, would have to read it existing value first, and then add to it, if other tests had already ran. Harder still, we have to write them in same order as column "titles". Guess would be easiest to work off one set of properties, and simply sort by alpha-numeric or something? (But, order might change then, as "tests completed, and came in" to be summarized. (the built-in defaults should still work for R4_5_maintenance stream ... until generalized solution completed). C. It is unclear where best to set the testsConfigExpect variable in ant scripts. Some are set in helper.xml, but those didn't "take" ... and, I had to set directly in publish.xml (I think we sometimes call 'helper', and sometimes call publish.xml directly). D: The overall console logs still have incorrect value after changes for bug 473904. linux.gtk.x86_64_8.0_consolelog.txt macosx.cocoa.x86_64_7.0_consolelog.txt win32.win32.x86_7.0_consolelog.txt (In reply to David Williams from comment #5) > D: The overall console logs still have incorrect value after changes for bug > 473904. > linux.gtk.x86_64_8.0_consolelog.txt > macosx.cocoa.x86_64_7.0_consolelog.txt > win32.win32.x86_7.0_consolelog.txt Off hand, not sure how to take advantage of it, but the correct values are written to production.properties. (as the value of the variable testedPlatform. A possible bug, is that is seems as if the following files are left in the "archive" directory, and not copied to the "data" of each platform: $ ls -1 archive/ antDiagnostics.txt envVars.txt javaSettings.txt production.properties propertiesAllFromRunTest2.properties (In reply to David Williams from comment #5) > > D: The overall console logs still have incorrect value after changes for bug > 473904. > linux.gtk.x86_64_8.0_consolelog.txt > macosx.cocoa.x86_64_7.0_consolelog.txt > win32.win32.x86_7.0_consolelog.txt This problem turned out to be a case of remaining hard coding in Mac and Windows scripts, so changed to use computed variable, as already done in Linux scripts. http://git.eclipse.org/c/platform/eclipse.platform.releng.aggregator.git/commit/?id=6838405e40ccc4c443de3bf9f59d8dd4b1efcc94 Doing a mass "reset to default assignee" of 52 bugs to help make clear it will (very likely) not be me working on things I had previously planned to work on. I hope this will help prevent the bugs from "getting lost" in other people's queries. Feel free to "take" a bug if appropriate. This is fixed and even java level is descriptive in platform name. |
In our build/test system, we use "platform name" in several places to form log names, links, etc. Now that I've broken it, thought I'd open a bug to document all the places it "touches" ... though, still missing something. a) in the "vm.properties" files, there's a line similar to platform=linux.gtk.x86_6.0 that'd changed to platform=linux.gtk.x86_64_6.0 [to signify "64 bit"] b) the "main log" is piped or renamed (or created by ant?) to .../results/consolelogs/linux.gtk.x86_6.0_consolelog.txt; changed to .../results/consolelogs/linux.gtk.x86_64_6.0_consolelog.txt [this seemed to happen automatically, but could not find where its done]. c) during the tests (via test framework) the junit results are put in ${results}/xml/${report}_${platform}.xml [that is, controlled by 'a', above]. d) the "logs.php" file has a section with lines such as listDegailedLogs("linux.gtk.x86_6.0"); that needs to be changed to listDegailedLogs("linux.gtk.x86_64_6.0"); e) the file testManifest.txt has lines such as <logFile name="org.eclipse.releng.tests_linux.gtk.x86_6.0.xml"> changed to <logFile name="org.eclipse.releng.tests_linux.gtk.x86_64_6.0.xml"> [If not done, the test will show up as "missing results"]. f) after the tests are done, there is a step where the "summary table" is created by a custom ant task, that shows "number of failures/errors" and links to the full report. It is this "summary step" that is currently failing. Granted, I have only "just now" updated the testManifest.xml file. After doing that, I tried re-generating but didn't seem to work. So either the "regen" manual process isn't working quite right or the "summary task" can not handle a name with so many underscores. If I can't figure out soon, I will revert changes a through e.