Skip to contentSkip to conversation

Define minimal supported kernel and (g)libc version

RFC Petr Vorel 3 participants 8 messages Mar 13, 2020 – Mar 21, 2020

lore
Share & tools
lore

Petr Vorel

author
#0 lore raw

Hi,

I'm sorry, I've raised this question in the past, but it got lost. I remember we talked about 2.6 something.

It'd be good to state publicly the oldest kernel and glibc (or even other libc versions) we support. This would allow us to remove some legacy code or force support for legacy code.

This shouldn't require test to be functional (e.g. for some cases like module drivers it could be hard), but LTP to be compiled and when difficult/impossible to achieve this functionality, it could resulted in TCONF (skipped test).

I created github issue for this [1], where I put link to this thread, so we don't loose it again.

Kind regards,
Petr

[1] https://github.com/linux-test-project/ltp/issues/657

#1 lore raw

Hi Petr,

This is a good topic, thanks for kicking off this initiative!

Quoted 5 lines · Petr Vorel

Hi,

I'm sorry, I've raised this question in the past, but it got lost. I remember we talked about 2.6 something.

Yes, the past discussion is still valuable to us. see:
http://lists.linux.it/pipermail/ltp/2019-May/011990.html

Quoted 6 lines

It'd be good to state publicly the oldest kernel and glibc (or even other libc
versions) we support. This would allow us to remove some legacy code or force
support for legacy code.

Maybe we could also state the oldest GCC version too? Though I haven't seen any conflict or supporting issue from my side, it helps avoid some potential error in cross-compilation I guess.

    i.e.  kernel-3.10.0 / glibc-2.17 / gcc-4.8.0
Quoted 7 lines

This shouldn't require test to be functional (e.g. for some cases like module
drivers it could be hard), but LTP to be compiled and when difficult/impossible
to achieve this functionality, it could resulted in TCONF (skipped test).

+1

Quoted 10 lines

I created github issue for this [1], where I put link to this thread, so we don't loose it again.

Kind regards,
Petr

[1] https://github.com/linux-test-project/ltp/issues/657

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200314/6475dc29/attachment.htm>
author
#2 lore raw

Hi Li,

Quoted 1 line

This is a good topic, thanks for kicking off this initiative!

Thanks for your input.

Quoted 5 lines
Quoted 2 lines

I'm sorry, I've raised this question in the past, but it got lost. I remember we talked about 2.6 something.

Yes, the past discussion is still valuable to us. see:
http://lists.linux.it/pipermail/ltp/2019-May/011990.html

Great, thanks!

Quoted 10 lines
Quoted 5 lines

It'd be good to state publicly the oldest kernel and glibc (or even other libc
versions) we support. This would allow us to remove some legacy code or force
support for legacy code.

Maybe we could also state the oldest GCC version too? Though I haven't seen any conflict or supporting issue from my side, it helps avoid some potential error in cross-compilation I guess.

+1
Not sure if we want to specify also clang.

Quoted 1 line
    i.e.  kernel-3.10.0 / glibc-2.17 / gcc-4.8.0

This is for RHEL7 I guess.

The oldest system in travis we have CentOS 6: kernel-2.6.32 / glibc-2.12 / gcc-4.4.7 (clang-3.4.2, but we don't test it with clang). I'm ok to have this older dependency, just to make sure it builds. But code would be cleaner for sure if we drop it.

BTW I also occasionally test build on SLES 11-SP3 (kernel 3.0 / glibc-2.11.3 / gcc-4.3.4 - older glibc and gcc), but this is not even in travis. But for testing these distros we use older releases (the same mentioned Jan [1]). I wonder if there is really somebody using 2.6.x or 3.x < 3.10 on master. If not, we can drop some lapi files which mention 2.6.

Kind regards,
Petr

[1] http://lists.linux.it/pipermail/ltp/2019-May/011991.html

#3 lore raw

Hi Petr,

Quoted 21 lines · Petr Vorel

Hi Li,

Quoted 1 line

This is a good topic, thanks for kicking off this initiative!

Thanks for your input.

Quoted 5 lines
Quoted 2 lines

I'm sorry, I've raised this question in the past, but it got lost. I remember we talked about 2.6 something.

Yes, the past discussion is still valuable to us. see:
http://lists.linux.it/pipermail/ltp/2019-May/011990.html

Great, thanks!

Quoted 1 line
Quoted 1 line

It'd be good to state publicly the oldest kernel and glibc (or even

other

Quoted 2 lines
Quoted 2 lines

libc
versions) we support. This would allow us to remove some legacy code

or

Quoted 2 lines
Quoted 2 lines

force
support for legacy code.

Quoted 7 lines
Quoted 1 line

Maybe we could also state the oldest GCC version too? Though I haven't

seen

Quoted 2 lines

any conflict or supporting issue from my side, it helps avoid some potential error in cross-compilation I guess.

+1
Not sure if we want to specify also clang.

I do not use clang often, so I hope others can advise more here.

Quoted 4 lines
Quoted 2 lines
    i.e.  kernel-3.10.0 / glibc-2.17 / gcc-4.8.0

This is for RHEL7 I guess.

Correct.

Quoted 8 lines

The oldest system in travis we have CentOS 6: kernel-2.6.32 / glibc-2.12 / gcc-4.4.7 (clang-3.4.2, but we don't test it with clang). I'm ok to have this
older dependency, just to make sure it builds. But code would be cleaner for
sure if we drop it.

+1
Sounds good to me.

Quoted 7 lines

BTW I also occasionally test build on SLES 11-SP3 (kernel 3.0 / glibc-2.11.3 /
gcc-4.3.4 - older glibc and gcc), but this is not even in travis. But for testing these distros we use older releases (the same mentioned Jan [1]).

Agreed, we can explicitly declare that(in some place of Doc) from a specific LTP(e.g ltp-full-20200120) version, we don't provide code supporting for the older kernel/glibc/gcc package anymore. If people who are going to test old distros, they can just pick up an old released LTP version and hack it by themself. The latest branch of LTP doesn't accept that patch for old things.

Quoted 9 lines

I wonder if there is really somebody using 2.6.x or 3.x < 3.10 on master. If not, we can drop some lapi files which mention 2.6.

Kind regards,
Petr

[1] http://lists.linux.it/pipermail/ltp/2019-May/011991.html

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200317/568e8c22/attachment.htm>
author
#4 lore raw

Hi Li,

Quoted 9 lines
Quoted 6 lines

The oldest system in travis we have CentOS 6: kernel-2.6.32 / glibc-2.12 / gcc-4.4.7 (clang-3.4.2, but we don't test it with clang). I'm ok to have this
older dependency, just to make sure it builds. But code would be cleaner for
sure if we drop it.

+1
Sounds good to me.

I'm sorry, are you for removing CentOS 6 from Travis or not?

Quoted 7 lines
Quoted 5 lines

BTW I also occasionally test build on SLES 11-SP3 (kernel 3.0 / glibc-2.11.3 /
gcc-4.3.4 - older glibc and gcc), but this is not even in travis. But for testing these distros we use older releases (the same mentioned Jan [1]).

Quoted 6 lines

Agreed, we can explicitly declare that(in some place of Doc) from a specific LTP(e.g ltp-full-20200120) version, we don't provide code supporting for the older kernel/glibc/gcc package anymore. If people who are going to test old distros, they can just pick up an old released LTP version and hack it by themself. The latest branch of LTP doesn't accept that patch for old things.

+1

Kind regards,
Petr

#5 lore raw

Hi Petr,

Quoted 8 lines · Petr Vorel

Hi Li,

Quoted 1 line
Quoted 1 line

The oldest system in travis we have CentOS 6: kernel-2.6.32 /

glibc-2.12 /

Quoted 1 line
Quoted 1 line

gcc-4.4.7 (clang-3.4.2, but we don't test it with clang). I'm ok to

have

Quoted 2 lines
Quoted 2 lines

this
older dependency, just to make sure it builds. But code would be

Quoted 8 lines

cleaner

Quoted 5 lines
Quoted 2 lines

for
sure if we drop it.

+1
Sounds good to me.

I'm sorry, are you for removing CentOS 6 from Travis or not?

I previously think these words mean that only make sure LTP build passes on CentOS6(but not remove it).

But removing is also Ok to me, we can do that after we reaching an agreement on the supporting package-version and write it down in Docs. That probably occurring in the next release if everything goes well:).

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200317/ecf3daff/attachment.htm>
#6 lore raw
In reply to Petr Vorel

----- Original Message -----

Quoted 19 lines

Hi Li,

Quoted 1 line

This is a good topic, thanks for kicking off this initiative!

Thanks for your input.

Quoted 5 lines
Quoted 2 lines

I'm sorry, I've raised this question in the past, but it got lost. I remember we talked about 2.6 something.

Yes, the past discussion is still valuable to us. see:
http://lists.linux.it/pipermail/ltp/2019-May/011990.html

Great, thanks!

Quoted 5 lines
Quoted 5 lines

It'd be good to state publicly the oldest kernel and glibc (or even other libc
versions) we support. This would allow us to remove some legacy code or force
support for legacy code.

Quoted 8 lines
Quoted 3 lines

Maybe we could also state the oldest GCC version too? Though I haven't seen any conflict or supporting issue from my side, it helps avoid some potential error in cross-compilation I guess.

+1
Not sure if we want to specify also clang.

Quoted 1 line
    i.e.  kernel-3.10.0 / glibc-2.17 / gcc-4.8.0

This is for RHEL7 I guess.

Correct, it's still active (though in less extent than RHEL8). But I still see value of running/supporting LTP here.

As I said in previous thread, if we want to draw a line somewhere, e.g. say anything older 10 years is too old, RHEL6/Centos6 would fall in that. For regression tests it should be OK to use older stable release.

Quoted 6 lines

The oldest system in travis we have CentOS 6: kernel-2.6.32 / glibc-2.12 / gcc-4.4.7 (clang-3.4.2, but we don't test it with clang). I'm ok to have this older dependency, just to make sure it builds. But code would be cleaner for sure if we drop it.

Quoted 7 lines

BTW I also occasionally test build on SLES 11-SP3 (kernel 3.0 / glibc-2.11.3 /
gcc-4.3.4 - older glibc and gcc), but this is not even in travis. But for testing these distros we use older releases (the same mentioned Jan [1]).
I wonder if there is really somebody using 2.6.x or 3.x < 3.10 on master. If not, we can drop some lapi files which mention 2.6.

There are some, since LTP didn't reject such patches yet. But updates to those old kernels are few and far between, so it might be not be worth the trouble from LTP point of view.

author
#7 lore raw

Hi,

Quoted 7 lines

Correct, it's still active (though in less extent than RHEL8). But I still see value of running/supporting LTP here.

As I said in previous thread, if we want to draw a line somewhere, e.g. say anything older 10 years is too old, RHEL6/Centos6 would fall in that. For regression tests it should be OK to use older stable release.

OK. We can remove CentOS 6 (or probably replace with CentOS 7) now or after next release. I don't have a strong opinion about it.
After it's removal all that 2.6.x fixes in m4/ should be deleted.

I'd be for putting supported versions in README.md.

Quoted 13 lines
Quoted 5 lines

The oldest system in travis we have CentOS 6: kernel-2.6.32 / glibc-2.12 / gcc-4.4.7 (clang-3.4.2, but we don't test it with clang). I'm ok to have this older dependency, just to make sure it builds. But code would be cleaner for sure if we drop it.

Quoted 7 lines

BTW I also occasionally test build on SLES 11-SP3 (kernel 3.0 / glibc-2.11.3 /
gcc-4.3.4 - older glibc and gcc), but this is not even in travis. But for testing these distros we use older releases (the same mentioned Jan [1]).
I wonder if there is really somebody using 2.6.x or 3.x < 3.10 on master. If not, we can drop some lapi files which mention 2.6.

Quoted 3 lines

There are some, since LTP didn't reject such patches yet. But updates to those old kernels are few and far between, so it might be not be worth the trouble from LTP point of view.

No strong opinion. I wouldn't be against asking these people directly (which would lead to postpone deleting legacy support after next release).

Kind regards,
Petr