Re: CI log verbosity

2023-11-04 Thread Ben Cooksley
On Sat, Nov 4, 2023 at 2:48 AM Ingo Klöcker  wrote:

> On Freitag, 3. November 2023 13:01:26 CET Harald Sitter wrote:
> > What are your thoughts on having the CI be less verbose by default and
> > instead have an env var or some other toggle to switch into verbose
> > mode?
>
> +1
>
> Ideally, the verbose logs would be written to an artifact. Otherwise, it
> will
> be painful to debug intermittent problems. I think the Craft jobs now do
> this
> (also in response to insanely verbose output of some builds which made it
> impossible to see the actual error).
>

Unfortunately the test log output is all governed by CMake/CTest so there
isn't too much we can do ourselves - not without doing quite a bit of work
to read XML files produced by CTest anyway.


>
> > Specifically I'm talking about the qtlogging rules that are currently
> > enabling everything and the kitchen sink. To my mind we should just
> > use the default rules by default.
> > I find that 99% of the time the output is entirely useless in finding
> > what is wrong, if anything it gets in the way because I first have to
> > find where the test failure is and then instead of reading walls of qt
> > plugin info I will just proceed to reproduce the problem locally
> > anyway.
>
> Full ACK. I'm almost always only interested in finding the error which
> caused
> the CI job to fail and sometimes in seeing compiler warnings.
>
> Regards,
> Ingo
>

Cheers,
Ben


Re: CI log verbosity

2023-11-04 Thread Ben Cooksley
On Sat, Nov 4, 2023 at 1:01 AM Harald Sitter  wrote:

> What are your thoughts on having the CI be less verbose by default and
> instead have an env var or some other toggle to switch into verbose
> mode?
>
> Specifically I'm talking about the qtlogging rules that are currently
> enabling everything and the kitchen sink. To my mind we should just
> use the default rules by default.
> I find that 99% of the time the output is entirely useless in finding
> what is wrong, if anything it gets in the way because I first have to
> find where the test failure is and then instead of reading walls of qt
> plugin info I will just proceed to reproduce the problem locally
> anyway.
>

The current approach to enabling all of the debugging information was
adopted a few years back when dfaure was trying to debug some tests.
We were back on Jenkins then though, which made it quite difficult to set
things yourself, unlike what you can do now with Gitlab CI and a work
branch.

Behaviour is changed from
https://invent.kde.org/sysadmin/ci-utilities/-/commit/67e9aaac0f540005834bd61c60abf01eed01cb12

If anyone would like to propose a set of more narrow, sane logging rules
then please send an MR (would suggest that it is wrapped behind a check for
whether the logging rules variable is set already though)

Cheers,
Ben


>
> Thoughts?
>
> HS
>


Re: CI log verbosity

2023-11-03 Thread Ingo Klöcker
On Freitag, 3. November 2023 13:01:26 CET Harald Sitter wrote:
> What are your thoughts on having the CI be less verbose by default and
> instead have an env var or some other toggle to switch into verbose
> mode?

+1

Ideally, the verbose logs would be written to an artifact. Otherwise, it will 
be painful to debug intermittent problems. I think the Craft jobs now do this 
(also in response to insanely verbose output of some builds which made it 
impossible to see the actual error).

> Specifically I'm talking about the qtlogging rules that are currently
> enabling everything and the kitchen sink. To my mind we should just
> use the default rules by default.
> I find that 99% of the time the output is entirely useless in finding
> what is wrong, if anything it gets in the way because I first have to
> find where the test failure is and then instead of reading walls of qt
> plugin info I will just proceed to reproduce the problem locally
> anyway.

Full ACK. I'm almost always only interested in finding the error which caused 
the CI job to fail and sometimes in seeing compiler warnings.

Regards,
Ingo


signature.asc
Description: This is a digitally signed message part.


CI log verbosity

2023-11-03 Thread Harald Sitter
What are your thoughts on having the CI be less verbose by default and
instead have an env var or some other toggle to switch into verbose
mode?

Specifically I'm talking about the qtlogging rules that are currently
enabling everything and the kitchen sink. To my mind we should just
use the default rules by default.
I find that 99% of the time the output is entirely useless in finding
what is wrong, if anything it gets in the way because I first have to
find where the test failure is and then instead of reading walls of qt
plugin info I will just proceed to reproduce the problem locally
anyway.

Thoughts?

HS