error extraction form build log (in case of bash: Permission denied error on Windows)

[Magnus Ihse Bursie]

On 2023-09-13 16:14, Baesken, Matthias wrote:

> I found  the "] Error " - advice  unfortunately AFTER  I  did some 
> searching through the log  (seems I  overlooked it at first because I 
> searched for a compile errors and not for such “hints” ) .
>
> But in this case it worked , so my idea was -  why not automatically 
> extract it using this pattern ?
>
> ( but maybe this would be too easy )
>
In this case you were lucky and the actual error was printed just a line 
above. In general, this is not the case, since output from different 
jobs can be intermingled, so a more typical scenario is that the real 
error message is printed tens or maybe hundreds of lines before the "] 
Error " line. The "] Error " line just marks the *last* line in the log 
that it is meaningful to look at -- anything relevant to the failure 
must be presented above it, but not necessarily just above it.

Maybe we can present this so it comes across more clearly to the 
developers, but I do not really know how. I've just resigned at the fact 
that no-one really reads the output from the build system. ;-)

/Magnus



> Best regards, Matthias
>
> *From:*Magnus Ihse Bursie <magnus.ihse.bursie at oracle.com>
> *Sent:* Wednesday, 13 September 2023 15:18
> *To:* Baesken, Matthias <matthias.baesken at sap.com>; build-dev at openjdk.org
> *Cc:* Zeller, Arno <arno.zeller at sap.com>
> *Subject:* Re: error extraction form build log (in case of bash: 
> Permission denied error on Windows)
>
> Hi Matthias,
>
> Let me see if I get this correct. You got a build failure, and the 
> built-in system could not diagnose it. Instead, it suggested that you 
> should search the build log for '] Error'. Instead, of doing this, you 
> scrolled through the entire log, just to find that the actual error 
> message was printed close to '] Error'..?
>
> What exactly are we to do about this?
>
> The reason we do not extract this information to the build failure 
> summary is that it is not possible (or, if possible, extremely 
> non-trivial). Instead, we direct the user on how to best go ahead to 
> try and find this information themselves. Fixing users not reading or 
> following suggestions is unfortunately out of scope for the build 
> system. ;-)
>
> There is no point in trying to search for this particular error and 
> repeat it to the user. The "No indication of failed target found. Try 
> searching the build log for '] Error'." is our generic fallback when 
> we have no way to determine the cause of failure. There are a myriad 
> ways of the build to fail in that case, if we start adding special 
> cases to this, there will be no end.
>
> /Magnus
>
> On 2023-09-08 13:19, Baesken, Matthias wrote:
>
>     Hello,
>
>     A Windows build failed because of some temporary  "Permission
>     denied" issue we see sometimes on Windows.
>
>     Unfortunately at the end of the log not much info is found, the
>     error is not reported there .
>
>     .....
>
>     if /usr/bin/grep -q "recipe for target .* failed"  .......
>
>     No indication of failed target found.
>
>     HELP: Try searching the build log for '] Error'.
>
>     printf "HELP: Run 'make doctor' to diagnose build problems.\n\n"
>
>     HELP: Run 'make doctor' to diagnose build problems.
>
>     After a lot of scrolling  through the endless build log, I found
>     in the middle of the build.log:
>
>     make[3]: /usr/bin/bash: Permission denied
>
>     make[3]: *** [Lib.gmk:62:
>     /cygdrive/C/Windows_x64/jdk-build-fastdebug/support/native/jdk.jpackage/jpackageapplauncher/MsiDb.obj]
>     Error 127
>
>     make[3]: *** Waiting for unfinished jobs....
>
>     which probably causedthe error.
>
>     Could we maybe add a grep for  "bash: Permission denied" or for
>     the recommended "] Error "    at the end of the build log ?
>
>     At least the grep/search for  "] Error "  finds the issue pretty
>     fast and looks like a good errorreporting addition in this case.
>
>     Best regards, Matthias
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/build-dev/attachments/20230914/4a2a8056/attachment-0001.htm>