[gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Pacho Ramos]

The problem I hit usually:

There are some packages that rely on X for some tests, that way, when I
for example emerge tk, it's nearly impossible to keep working because I
get tons of windows opened and closed. 

Wouldn't be better to make that tests *always* use virtual X?

Thanks for your thoughts

Re: [gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Patrick Lauer]

On 07/19/2013 02:50 PM, Pacho Ramos wrote:
> The problem I hit usually:
> 
> There are some packages that rely on X for some tests, that way, when I
> for example emerge tk, it's nearly impossible to keep working because I
> get tons of windows opened and closed. 
> 
> Wouldn't be better to make that tests *always* use virtual X?

Yes. Any that do not are defective and need to be fixed.

Re: [gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Chí-Thanh Christopher Nguyễn]

Pacho Ramos schrieb:
> Wouldn't be better to make that tests *always* use virtual X?

One issue could be that Xvfb is not very representative of a modern X
server any more. I am not sure how this affects the validity of the test
results. This problem however exists already with all virtualx users on
headless machines (bug 409925 might address this to some degree).

You are right that using always Xvfb would be much less annoying to
someone running tests on his desktop computer.


Best regards,
Chí-Thanh Christopher Nguyễn

Re: [gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Diego Elio Pettenò]

[-- Attachment #1: Type: text/plain, Size: 512 bytes --]

On Fri, Jul 19, 2013 at 7:50 AM, Pacho Ramos <pacho@gentoo.org> wrote:

>
> Wouldn't be better to make that tests *always* use virtual X?


As Patrick said, yes, if they don't they are broken. Especially so if Xvfb
does not represent a modern X anymore, as that would mean that tests will
behave differently depending on where you launch them.

And non-deterministic tests are stupid, useless, broken tests.


Diego Elio Pettenò — Flameeyes
flameeyes@flameeyes.eu — http://blog.flameeyes.eu/

[-- Attachment #2: Type: text/html, Size: 1046 bytes --]

Re: [gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Fabio Erculiani]

On Fri, Jul 19, 2013 at 1:40 PM, Diego Elio Pettenò
<flameeyes@flameeyes.eu> wrote:
> [...]

> And non-deterministic tests are stupid, useless, broken tests.

Amen.
Even though there is that 1% of cases in where you want to have
non-deterministic tests. For instance, if you want to run the same
test thousands of times and randomly pick an initial state to see if
you spot a "scenario" in where you have problems (concurrency
problems)... :-)

>
>
> Diego Elio Pettenò — Flameeyes
> flameeyes@flameeyes.eu — http://blog.flameeyes.eu/



-- 
Fabio Erculiani

Re: [gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Ian Stakenvicius]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On 19/07/13 08:20 AM, Fabio Erculiani wrote:
> On Fri, Jul 19, 2013 at 1:40 PM, Diego Elio Pettenò 
> <flameeyes@flameeyes.eu> wrote:
>> [...]
> 
>> And non-deterministic tests are stupid, useless, broken tests.
> 
> Amen. Even though there is that 1% of cases in where you want to
> have non-deterministic tests. For instance, if you want to run the
> same test thousands of times and randomly pick an initial state to
> see if you spot a "scenario" in where you have problems
> (concurrency problems)... :-)

You still want those tests to be deterministic though -- ie, imo it's
better to run through a very large set of test-cases in some sort of
repeatable order, than to rely on randomization.  At least then you
know the cases where it succeeds and the cases where it is
unknown/untested.

..but this is getting off-topic.  :)

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.20 (GNU/Linux)

iF4EAREIAAYFAlHpPGwACgkQ2ugaI38ACPD4TgD+MQJwHU4U6mT98Gz28WzuCK0Q
U6P5h1PzJVWdU1Zpd58A/R1TEhIXMX2P/NV3w2wHerdVQKyrX2/CqjQyHTKhd7AC
=jUtf
-----END PGP SIGNATURE-----

Re: [gentoo-dev] Running tests using virtualx.eclass should be allowed to be forced to run in virtual X always

[Diego Elio Pettenò]

[-- Attachment #1: Type: text/plain, Size: 601 bytes --]

On Fri, Jul 19, 2013 at 1:20 PM, Fabio Erculiani <lxnay@gentoo.org> wrote:

> Even though there is that 1% of cases in where you want to have
> non-deterministic tests. For instance, if you want to run the same
> test thousands of times and randomly pick an initial state to see if
> you spot a "scenario" in where you have problems (concurrency
> problems)... :-)
>

Not during packaging. Those are developer tests. Yes there are difference
between the two. But as Ian said, we're off-topic now ;)

Diego Elio Pettenò — Flameeyes
flameeyes@flameeyes.eu — http://blog.flameeyes.eu/

[-- Attachment #2: Type: text/html, Size: 1125 bytes --]