Re: [PATCH v6 1/2] selftests: Rename sigaltstack to generic signal

From: Dev Jain
Date: Mon Sep 09 2024 - 01:16:29 EST


On 9/7/24 01:29, Shuah Khan wrote:
On 9/4/24 23:56, Dev Jain wrote:

On 9/4/24 22:35, Shuah Khan wrote:
On 9/3/24 22:52, Dev Jain wrote:

On 9/4/24 03:14, Shuah Khan wrote:
On 8/30/24 10:29, Dev Jain wrote:

On 8/27/24 17:16, Dev Jain wrote:

On 8/27/24 17:14, Shuah Khan wrote:
On 8/22/24 06:14, Dev Jain wrote:
Rename sigaltstack to generic signal directory, to allow adding more
signal tests in the future.

Sorry - I think I mentioned I don't like this test renamed. Why are you sending
this rename still included in the patch series?

I am not renaming the test, just the directory. The directory name
is changed to signal, and I have retained the name of the test -
sas.c.

Gentle ping: I guess there was a misunderstanding; in v5, I was
also changing the name of the test, to which you objected, and
I agreed. But, we need to change the name of the directory since
the new test has no relation to the current directory name,
"sigaltstack". The patch description explains that the directory
should be generically named.


Right. You are no longer changing the test name. You are still
changing the directory name. The problem I mentioned stays the
same. Any fixes to the existing tests in this directory can no
longer auto applied to stables releases.

I understand your point, but commit baa489fabd01 (selftests/vm: rename
selftests/vm to selftests/mm) is also present. That was a lot bigger change;
sigaltstack contains just one test currently, whose fixes possibly would have
to be backported, so I guess it should not be that much of a big problem?



So who does the backports whenevenr something changes? You are adding
work where as the automated process would just work without this
change. It doesn't matter if there is another test that changed
the name.

Other than the desire to rename the directory to generic, what
other value does this change bring?

Do you have an alternative suggestion as to where I should put my new test then;
I do not see what is the value of creating another directory to just include
my test. This will unnecessarily clutter the selftests/ directory with
directories containing single tests. And, putting this in "sigaltstack" is just
wrong since this test has no relation with sigaltstack.


If this new test has no relation to sigaltstack, then why are you changing
and renaming the sigaltstack directory?

Because the functionality I am testing is of signals, and signals are a superset
of sigaltstack. Still, I can think of a compromise, if semantically you want to
consider the new test as not testing signals, but a specific syscall "sigaction"
and its interaction with blocking of signals, how about naming the new directory "sigaction"?
Adding a new directory is much better
than going down a path that is more confusing and adding backport overhead.


Okay - they are related except that you view signalstack as a subset
of signals. I saw Mark's response as well saying sigaction isn't
a good name for this.

Rename usually wipe out git history as well based on what have seen
in the past.

My main concern is backports. Considering sigstack hasn't changed
2021 (as Mark's email), let's rename it.

I am reluctantly agreeing to the rename as it seems to make sense
in this case.

Thanks! I guess there is no update required from my side, and you can
pull this series?

thanks,
-- Shuah