[Scons-dev] Failing buildbot runs...

[William Deegan]

Dirk,

On Aug 30, 2012, at 3:13 PM, Dirk Bächle <tshortik at gmx.de> wrote:


> Hi,

> 

> On 30.08.2012 15:45, Gary Oberbrunner wrote:

>> On Wed, Aug 29, 2012 at 6:14 PM, Dirk Bächle<tshortik at gmx.de>  wrote:

>>> Hi,

>>> 

>>> the new buildslaves are doing good work...and throw a lot of fails. ;)

>>> 

>>> I'd like to open a bug for fixing a few issues, such that we can hopefully

>>> get all tests to pass again.

>>> Just wanted to check and make sure that nobody else has started to work on

>>> this.

>>> 

>>> If you do, raise your hand please...else I'll start tomorrow.

>> This would be great!  I fixed a few Windows failures last week, but

>> there are still some out there.  I think (?) all of them are test-code

>> rather than actual SCons bugs.  I know at least one is just a

>> case-sensitive string compare of drive letters (C: vs. c:), and I did

>> add a case-insensitive QMtest string matcher a few weeks ago which

>> would probably fix that one.

>> 

> 

> I started work on this (issue #2872) and will probably need a few days. ;)

> It's not that complicated overall, just a lot to do.

> 

> Anyway, we still have some real problems:

> 

>  - The translation issue, some tests expect a system message in English. This breaks with a German Windows

>    installed, or any other language. Should I try to get a full translation scheme going, using gettext?

>    I can try, but do we really want to open this can of worms?


Sounds painful.
Looks like only a small handful of tests are affected by this?
If so you might just check the local and skip them if not en_US ?



>  - Several MSVS tests expect that VisualStudio is installed and call "cl" directly. Should I simply check with a

>    test.where_is('cl') and then skip it if nothing is found (=only MinGW installed)? Is this safe enough for now?

>    Or should we provide a better way to detect the installed toolchain before the actual tests are carried out?


That sounds reasonable. I believe we do that with many tests which have no value if the tool in question isn't installed.


>  - Fortran support seems to be broken badly under Windows. Didn't have time to investigate this further so far,

>    but the suffixes (*.f77, ...) are not properly recognized.

>    Hints, pointers and solutions are welcome! :)


Yes I see that from your build logs.
Probably there's no fortran builder being installed by default on windows and/or with mingw?


> 

> Finally, a request regarding the Buildbot setup. Bill, can you please increase the timeout setting for the Fedora17 slave from 1200s to something like 5400s? Yeah, I know that's a lot...but I timed the test KeyboardInterrupt.py on the VirtualMachine without any significant load. It takes a full 18min to finish, so we need to give it more time.

> (These are 2*12 test runs with '-j' increasing from 8 up to 64…)


Timeout changed to 5400 seconds for the runtest.py step.

Thanks for all the work! 
It's great having folks provide buildslaves, and even greater when they spend some time addressing the issues they reveal!

-Bill