Hi,The best way if you want to fix these flaky tests (ie non deterministic tests) is to try to reproduce them in your favorite IDE.
Regards, Raphaël Ouazana. Le 2017-06-08 13:28, cryptearth a écrit :
Hey Benoit, first: sorry for getting your name the wrong way around - I was confused by your signature. About maildir not supported on Windows: didn't knew, but IOException got thrown and the stack for it made it pretty clear whats happening. Fun fact: try to compile a class called "aux" - it will fail cause the filename "aux.class" is reserved and can't be choosen as file- or folder-name by normal user. I'm getting confused about HBase as I couldn't find any line that's system depended, so I can't figure out what's raising the error as the stack looks incomplete. Is there any other way to get a full stack for such errors? Maybe I'm able to figure out what's wrong when I get the line causing the fail. Matt ---- Benoit Tellier schrieb ----Hi, My answers inlined... Cheers, Le 08/06/2017 à 16:16, cryptearth a écrit :Hey Rapael,I know about this as Tellier told me this once the first time I reported issues when first tried to build beta5. So this is what I as a hobbiest don't understand: if the commit that is made public at least passed all tests once - why it's failing so differently on my different build setups? Java shouldn't behave like this. And as I quickly looked into the source-files causing the issues and couldn't spot any lines that maybe responsible for this I couldn't figure out what went wrong (at least I'm able to read and understand a StackTrace).ByteCode runs everywhere, true.However, you still depend on the system for fileSystem (resource names), what your operating system allows (file locks, etc...), encoding, etc...That is some kind of concerns that should be enforced in your code...I expected issues related to Docker as I don't have it installed, but not the one I encountered.Maybe some wants to know: the tests fail pretty hard on Windows for the maildir target as it seems the tests try to cteate files or folders wich aren't allowed on NTFS/Windows according to IOException: syntax error for filename.Note that maildir in James is documented as not supported for Windows.Wouldn't any problem when using database as storage - but could lead to unexpected failures during runtime.True, this can create issues at run-time.That being said, it's an open source project. Contributions are welcometo improve cross plateform situation.Matt ---- Raphael OUAZANA schrieb ----Hey Matt, Thank you for your interest in this project.You seem to have some issues building James. That's something that can happen because with have some flaky tests, and not enough time to solvethem all.But you should now that each time we are making a commit, all the tests do pass at least one time (and generally multiple times in practice), sowe are pretty confident with the quality on the commits we merge.If you have so much issues building James, please consider to build itwithout tests. It will take less than 2 minutes and you will quickly have the binaries you are looking for. To do this, just add -DskipTests=true to your current maven options. Regards, Raphaël Ouazana. Le 2017-06-08 01:07, cryptearth a écrit :Hey there,sorry for me to taking so long to reply - but it happend again: as Iwas preparing to set up some VMs for testing - I failed again thecursed game of luck to download the working commit and got the RC2 youjust pushed out. As I didn't knew it the time I just lazyly startedanother try on my root server it surprisingly went very well. At least until the test wich starts up RMI - wich failed cause the ports werealready in use by the running beta6 instance. So I shut down therunning instance - re-tried it - and it got a success. As there is nodocker installed on my root I got few docker related issues, but it seems these failed tests didn't mattered. After re-running it with-Pwith-assembly I also got the zip/tar.gz - wich I'm currently runningnow and over wich this mail is sent. As I tried to set up a VM to match my servers setup - but it still fail at apache-james-mailbox-hbase I re-run maven with -X and -e, but still didn't get more than this:Tests run: 1, Failures: 0, Errors: 1, Skipped: 0, Time elapsed: 1.217sec <<< FAILURE! - in org.apache.james.mailbox.hbase.mail.HBaseMailboxMessageMapperTest testMessageMapperScenario(org.apache.james.mailbox.hbase.mail.HBaseMailboxMessageMapperTest) Time elapsed: 0.003 sec <<< ERROR! java.lang.ExceptionInInitializerError at org.apache.james.mailbox.hbase.mail.HBaseMailboxMessageMapperTest.<clinit>(HBaseMailboxMessageMapperTest.java:69) It looks like there is the rest of the stacktrace missing - but I can't find it anywhere. Line 69 is this one here: public static final HBaseClusterSingleton CLUSTER = HBaseClusterSingleton.build();Maybe this helps - even if I still can't understand WHY this happens,as it went fine on the same setup on my root. Can't tell any difference: same maven, same jdk, same OS and patch-level. Nothingwich could cause this fail according to the source as there is nothingsystem-level called, at least nothing I can spot on first look.If anyone want a compiled RC2 buld (as long as the licence permits me to offer this) can ask me for a link to download it from my server. Beaware: it's 1.1GB in size. Any way - so long - Matt Am 05.06.2017 um 11:57 schrieb Benoit Tellier:Hi, First thanks for the feedback, it is rather interesting.So, to start this of with a meme: One does not simply - compile ApacheJames from source. :-)Call me stupidI would never. We all have different backgrounds, thus we don't consider the same things easy. About the 3 fails that you encountered: - apache-james-mailbox-hbase : I already saw our Continuous Integration system fail there. As we don't use this part of the project, we schedulea new build. As it is a very rare event, it doesn't matter much. Thequestion would then be: do you always observe it? - Concerning windows: As far as I am aware of it, all of the recent active James contributorsare using Linux. I personally use ArchLinux, some others prefer Debianand Ubuntu. We don't have a license for Windows, thus we can not guaranty we do not introduce bugs regarding it.As it is a free project, anyone can propose patches, and for instancepatches for making the test suite run well on windows. Another remark on cross-system: as we are aware of the issues, we provides docker container, continuously delivered, fully tested on eachmerge. That way one can run James in a reliable environment, whateverthe system it is running on. For your information: - https://docs.docker.com/- https://github.com/apache/james-project/tree/master/dockerfilesThat being said, I'm sorry to tell you your mail is hardly readable,and I can not extract interesting information out of it. You did a great work testing different environments, and I should thank you for this. Wouldn't you mind sending us, for each bug you encountered: - The Junit test that failed - The explanation JUnit is giving - The environment (Maven version, Java version, OS, default encoding) That would allow us to work in a rather more productive way. Cheers, -- Tellier Benoit Software engineer dedicated to OpenPaaS at Linagora PMC of the Apache JAMES project VIE in Vietnam https://twitter.com/AwesomePaaS https://medium.com/linagora-engineering Le 05/06/2017 à 16:00, cryptearth a écrit :So, to start this of with a meme: One does not simply - compile Apache James from source. Idk if and what I'm doin wrong here, but either its my hardware screwingup everything I've learned about Java (would explain random crashesinGTA5 tho) or I'm just to stupid to correctly setup the needed buildenvironment to get a successfull build. But let me start from the beginning why I'm back after abandoning this mail-list (still not used to this kind of public communication):So as I was successfull to compile one of the latest builds beforethisproject moved to Docker and have it running on my openSuSE-tumbleweed server since (and as smart as I am: deleted the built package - of course!) - I just looked up the main project page and noticed: oh,it's out of beta - RC1 available for download. But as the page shows "Docker"(sidenote: yea - I know it makes sense to "containerize" such code -torun a Java code as root is not the smartest idea one can have) I saidtomysefl: "screw it - compile from source" - and off we go from "wonderif it's still crap as last time I tried" to "what the F*?".So let me show the results first and then let me explain why I thinkmy hardware is broken: vm - opensuse tumbleweed - failed: apache-james-mailbox-hbase vm - debian 8.8.0 - failed: james-server-mailets host - win7 sp1 ulti x64 - failed: apache-james-mailbox-store I don't bother you with posting the logs - as it seems some wired random-ish but surprisingly re-produceable stuff going on here:As building james isn't more than compiling Java source into bytecode-and as Java is supposed to be platform-independent - it should failon the exact same point on each different system - but it doesn't. Unlikeearlier tries where it "crashed" random on the same system - at leastnoit's "crashing" on the same spot every time - but why and how? Theonlydifference are Linux vs Windows and openJDK8u131 vs Oracle 8u121 -and as far as I know Java as a hobbiest dev this shouldn't happen. At leastthe error should be the same accross differnt systems - no matter ifVM or real hardware.Ok, the error on windows seems to be some wired random-ish encodingissue, see the few lines of log as follows: Failed tests:DefaultTextExtractorTest.textTest:44 expected:<...e awesome texttext.[ ] "> but was:<...e awesome text text.[ ] "> I can only imagine there is something goin on with different line-endings as the build expecting only linux-style \n while my windows using \r\n - confusing the equality check to fail (some more like thisif you try to bootstrap ant from source on windows - it fails cause windows doesn't support posixfileattributes - wich could checked andhandled in a very easy way - but this should belong to the ant-maillist). The other two on the linux-based systems are very strange:On the openSuSE (ok, to be honest - it's the distro I "grew up" with- and strangely the only major distro that somehow no body seems to likeand therefore isn't really supported at all - just: WHY? cause itsgerman?) it fails with java.lang.ExceptionInInitializerError for org.apache.james.mailbox.hbase.user.HBaseSubscriptionMapperTest.<clinit> followed by java.lang.NoClassDefFoundError: Could not initialize class org.apache.james.mailbox.hbase.user.HBaseSubscriptionMapperTestOn the debian (wich went way better and further than the other two)it fails with this crap: Failed tests: RemoteDeliveryTest.remoteDeliveryShouldSplitMailsByServerWhenNoGateway:123 Expecting: <[FakeMail{msg=null, recipients=[ot...@james.apache.org, a...@james.apache.org], name=mail_name-to-james.apache.org, sender=null, state=null, errorMessage=null, lastUpdated=null, attributes={}, size=0, remoteAddr=127.0.0.1}, FakeMail{msg=null, recipients=[a...@james2.apache.org], name=mail_name-to-james2.apache.org, sender=null, state=null, errorMessage=null, lastUpdated=null, attributes={}, size=0, remoteAddr=127.0.0.1}]> to contain only: <[FakeMail{msg=null, recipients=[a...@james.apache.org, ot...@james.apache.org], name=mail_name-to-james.apache.org, sender=null, state=null, errorMessage=null, lastUpdated=null, attributes={}, size=0, remoteAddr=127.0.0.1}, FakeMail{msg=null, recipients=[a...@james2.apache.org], name=mail_name-to-james2.apache.org, sender=null, state=null, errorMessage=null, lastUpdated=null, attributes={}, size=0, remoteAddr=127.0.0.1}]> elements not found: <[FakeMail{msg=null, recipients=[a...@james.apache.org, ot...@james.apache.org], name=mail_name-to-james.apache.org, sender=null, state=null, errorMessage=null, lastUpdated=null, attributes={}, size=0, remoteAddr=127.0.0.1}]> and elements not expected: <[FakeMail{msg=null, recipients=[ot...@james.apache.org, a...@james.apache.org], name=mail_name-to-james.apache.org, sender=null, state=null, errorMessage=null, lastUpdated=null, attributes={}, size=0, remoteAddr=127.0.0.1}]> Just another encoding-issue based on non-standard non-EN/US setup (allsystems share locale de-de)? This seems to be another issue based onwrong String comparison. So - to complete this kinda sarcastic-ish mail with a few final words:1.) Does anyone of you have any clue why the same source behave so differently on different systems? This is somehow against anythingelse I know about Javas famous "write once - run everywhere".2.) Could anybody please be so kind to just get a me quick overview about how to correctly setup a working build environment to get thecurrent source built successfully? Call me stupid - or just naiv cause I don'T know anything about modern development on projects this size - but as a Windows-98 kid from times where Google didn't existed yet and just got few low-ish games by burnedCDs from school friends (online DRM wasn't a thing back then) who isjust used to "double-click on setup/install and hit next until it saysfin" - there is no clue why this to-be-simple-magic-box called "PC" under my desk behaves this way. It's just a big calculator - and I expect it to output 2 if I enter 1+1. Maybe someone can explain ...I don'T really expect any serious response/reply - nor to have any changes in the source - but as Benoit Tellier (btw: big shout outs tohim) once told me: each commit gets auto-compiled and only passed oneseven released to public - so it has to be some sort of error on mysite.--------------------------------------------------------------------- To unsubscribe, e-mail: server-user-unsubscr...@james.apache.org For additional commands, e-mail: server-user-h...@james.apache.org--------------------------------------------------------------------- To unsubscribe, e-mail: server-user-unsubscr...@james.apache.org For additional commands, e-mail: server-user-h...@james.apache.org--------------------------------------------------------------------- To unsubscribe, e-mail: server-user-unsubscr...@james.apache.org For additional commands, e-mail: server-user-h...@james.apache.org--------------------------------------------------------------------- To unsubscribe, e-mail: server-user-unsubscr...@james.apache.org For additional commands, e-mail: server-user-h...@james.apache.org
--------------------------------------------------------------------- To unsubscribe, e-mail: server-user-unsubscr...@james.apache.org For additional commands, e-mail: server-user-h...@james.apache.org