Proposal of changing generating version of mdadm

Proposal of changing generating version of mdadm

[Tanska, Kinga]

Hi,

recently we diagnosed few issues with 'mdadm -version' output.
Main problem is that end output varies on few conditions. We come with
simplified proposal. First let's describe current schema:

mdadm - version - date - extraversion
(example: mdadm - v4.2-rc2 - 2021-08-02 - extraversion)

or

mdadm - version - date
(example: mdadm - v4.2-rc2 - 2021-08-02).

VERSION could be taken from code (see ReadMe.c:31), but when git is
installed and .git directory is available in mdadm workspace, version
is replaced with output from # git describe HEAD command. It is assumed
that git command should return last tag from repo, which should contain
information about last release. This might not be true, especially if user
uses tags to mark internal milestones or custom mdadm spins.

The second problem is DATE, which corresponds to date of last release.
When few patches are picked onto HEAD date is not reliable. In my opinion
DATE is not needed. Usually, packages do not contain this element, e.g.
-	# git --version
		git version 2.27.0
-	# gcc --version
		gcc (GCC) 8.3.1 20191121 (Red Hat 8.3.1-5)
-	# yum --version
		4.2.23

To make it const and reliable, I propose removing DATE and always
use VERSION from code. VERSION shall keep general release information.
I would like to move the changeable elements into EXTRAVERSION. This
field will respect following conditions:
-	user definition first
       	(by respecting EXTRAVERSION=xxx during compilation)
-	if not defined by user, result of # git describe HEAD
-	else empty.

Example output:
mdadm - version - extraversion (example: mdadm - v4.2-rc2 - extraversion).
Thanks for any opinion about this proposition.

Regards,
Kinga Tanska

Re: Proposal of changing generating version of mdadm

[Jes Sorensen]

On 10/20/21 8:43 AM, Tanska, Kinga wrote:
> Hi,
> 
> recently we diagnosed few issues with 'mdadm -version' output.
> Main problem is that end output varies on few conditions. We come with
> simplified proposal. First let's describe current schema:
> 
> mdadm - version - date - extraversion
> (example: mdadm - v4.2-rc2 - 2021-08-02 - extraversion)
> 
> or
> 
> mdadm - version - date
> (example: mdadm - v4.2-rc2 - 2021-08-02).
> 
> VERSION could be taken from code (see ReadMe.c:31), but when git is
> installed and .git directory is available in mdadm workspace, version
> is replaced with output from # git describe HEAD command. It is assumed
> that git command should return last tag from repo, which should contain
> information about last release. This might not be true, especially if user
> uses tags to mark internal milestones or custom mdadm spins.
> 
> The second problem is DATE, which corresponds to date of last release.
> When few patches are picked onto HEAD date is not reliable. In my opinion
> DATE is not needed. Usually, packages do not contain this element, e.g.
> -	# git --version
> 		git version 2.27.0
> -	# gcc --version
> 		gcc (GCC) 8.3.1 20191121 (Red Hat 8.3.1-5)
> -	# yum --version
> 		4.2.23
> 
> To make it const and reliable, I propose removing DATE and always
> use VERSION from code. VERSION shall keep general release information.
> I would like to move the changeable elements into EXTRAVERSION. This
> field will respect following conditions:
> -	user definition first
>        	(by respecting EXTRAVERSION=xxx during compilation)
> -	if not defined by user, result of # git describe HEAD
> -	else empty.
> 
> Example output:
> mdadm - version - extraversion (example: mdadm - v4.2-rc2 - extraversion).
> Thanks for any opinion about this proposition.

Hi Kinga,

I am not against changing the format, however I worry that doing so may
break scripts and tools in the field that nobody is maintaining or have
thought of. If we are to change the output, I suggest making a new flag
that provides the details you want and we can deprecate the old one, but
leave it in place.

Thoughts?

Jes

Re: Proposal of changing generating version of mdadm

[Tanska, Kinga]

> On 10/20/21 8:43 AM, Tanska, Kinga wrote:
>> Hi,
>>
>> recently we diagnosed few issues with 'mdadm -version' output.
>> Main problem is that end output varies on few conditions. We come with
>> simplified proposal. First let's describe current schema:
>>
>> mdadm - version - date - extraversion
>> (example: mdadm - v4.2-rc2 - 2021-08-02 - extraversion)
>>
>> or
>>
>> mdadm - version - date
>> (example: mdadm - v4.2-rc2 - 2021-08-02).
>>
>> VERSION could be taken from code (see ReadMe.c:31), but when git is
>> installed and .git directory is available in mdadm workspace, version
>> is replaced with output from # git describe HEAD command. It is assumed
>> that git command should return last tag from repo, which should contain
>> information about last release. This might not be true, especially if user
>> uses tags to mark internal milestones or custom mdadm spins.
>>
>> The second problem is DATE, which corresponds to date of last release.
>> When few patches are picked onto HEAD date is not reliable. In my opinion
>> DATE is not needed. Usually, packages do not contain this element, e.g.
>> -	# git --version
>> 		git version 2.27.0
>> -	# gcc --version
>> 		gcc (GCC) 8.3.1 20191121 (Red Hat 8.3.1-5)
>> -	# yum --version
>> 		4.2.23
>>
>> To make it const and reliable, I propose removing DATE and always
>> use VERSION from code. VERSION shall keep general release information.
>> I would like to move the changeable elements into EXTRAVERSION. This
>> field will respect following conditions:
>> -	user definition first
>>         	(by respecting EXTRAVERSION=xxx during compilation)
>> -	if not defined by user, result of # git describe HEAD
>> -	else empty.
>>
>> Example output:
>> mdadm - version - extraversion (example: mdadm - v4.2-rc2 - extraversion).
>> Thanks for any opinion about this proposition.
> 
> Hi Kinga,
> 
> I am not against changing the format, however I worry that doing so may
> break scripts and tools in the field that nobody is maintaining or have
> thought of. If we are to change the output, I suggest making a new flag
> that provides the details you want and we can deprecate the old one, but
> leave it in place.
> 
> Thoughts?
> 
> Jes
> 

Thanks for your response.
I'm wondering if this change will be used if not will be set as default. 
I suggest providing a new flag with old, deprecated version, for any of 
uses where it couldn't be changed. I can add a description which clarify 
the change and describe flag, for anybody who will meet errors with 
scripts after this change.

Regards,
Kinga Tanska

Re: Proposal of changing generating version of mdadm

[Tanska, Kinga]

> On 10/20/21 8:43 AM, Tanska, Kinga wrote:
>> Hi,
>>
>> recently we diagnosed few issues with 'mdadm -version' output.
>> Main problem is that end output varies on few conditions. We come with
>> simplified proposal. First let's describe current schema:
>>
>> mdadm - version - date - extraversion
>> (example: mdadm - v4.2-rc2 - 2021-08-02 - extraversion)
>>
>> or
>>
>> mdadm - version - date
>> (example: mdadm - v4.2-rc2 - 2021-08-02).
>>
>> VERSION could be taken from code (see ReadMe.c:31), but when git is
>> installed and .git directory is available in mdadm workspace, version
>> is replaced with output from # git describe HEAD command. It is assumed
>> that git command should return last tag from repo, which should contain
>> information about last release. This might not be true, especially if 
>> user
>> uses tags to mark internal milestones or custom mdadm spins.
>>
>> The second problem is DATE, which corresponds to date of last release.
>> When few patches are picked onto HEAD date is not reliable. In my opinion
>> DATE is not needed. Usually, packages do not contain this element, e.g.
>> - # git --version
>> git version 2.27.0
>> - # gcc --version
>> gcc (GCC) 8.3.1 20191121 (Red Hat 8.3.1-5)
>> - # yum --version
>> 4.2.23
>>
>> To make it const and reliable, I propose removing DATE and always
>> use VERSION from code. VERSION shall keep general release information.
>> I would like to move the changeable elements into EXTRAVERSION. This
>> field will respect following conditions:
>> - user definition first
>> (by respecting EXTRAVERSION=xxx during compilation)
>> - if not defined by user, result of # git describe HEAD
>> - else empty.
>>
>> Example output:
>> mdadm - version - extraversion (example: mdadm - v4.2-rc2 - 
>> extraversion).
>> Thanks for any opinion about this proposition.
>
> Hi Kinga,
>
> I am not against changing the format, however I worry that doing so may
> break scripts and tools in the field that nobody is maintaining or have
> thought of. If we are to change the output, I suggest making a new flag
> that provides the details you want and we can deprecate the old one, but
> leave it in place.
>
> Thoughts?
>
> Jes
>

Thanks for your response.
I'm wondering if this change will be used if not will be set as default. 
I suggest
providing a new flag with old, deprecated version, for any of uses where it
couldn't be changed. I can add a description which clarify the change 
and describe
flag, for anybody who will meet errors with scripts after this change.

Regards,
Kinga Tanska