Re: [LTP] [PATCH] Add simple Containerfile

[Richard Palethorpe <rpalethorpe@suse.de>]

Hello,

Petr Vorel <pvorel@suse.cz> writes:

> Hi Richie,
>
>> >> RFC comments:
>> >>     * Add git clean -fdX which should remove any build artifacts
>> >>       this is different from the suggestion of just doing a check. I just
>> >>       found it easier to remove the build files.
>> > FYI what we do in release scripts, is to do a local clone to a different
>> > directory [2]:
>> > git clone ltp ltp-full-YYYYMMDD
>
>> > Not sure what is faster.
>
>> I guess that is something a script could also do then 'git clean'
>> becomes a null op. git clean -X will only remove untracked files so
>> pending changes should get picked up. Which is probably what people want
> "remove untracked files" - if you develop a new test, forget to add it with 'git
> add' and run the container, you will get disappointed :).

Possibly it is only files that are not tracked due to .gitignore. At
least that is how I interpret the Git help. Either way it's not such a
concern due to the below.

>
>> during development. Doing a fresh checkout is probably more like a hard
>> reset and clean.
>
> The benefit is that you have not only a clean git repo for the container,
> but also not touching your working copy directory. But unless nobody else
> raise any concern, I'm ok with your current proposal.

Ah I think I see your concern now. It does not touch the directory
outside the container. It copies everything in first (IDK if it actually
copies the data if you are doing a local build).

Note that I tried using a .dockerignore to stop build artifacts being
copied, but it's not as rich as the .gitignore(s) we have sprinkled
throughout the LTP.

>
>> AFAICT git clean is very quick, far faster than 'make distclean'.
>
>
>> >>     * Added seperate alpine and tumbleweed runtime scripts. Again it's
>> >>       different from the suggestion just because it's easier to add
>> >>       seperate scripts than adding a switch
>> > +1
>
>> > But maybe put it into container directory, because it's not used in
>> > GitHub CI?
>
>> I was thinking it could be used in CI. All we need is a CI that runs VMs
>> and we can do some testing. (e.g. srchut).
>
> Makes sense. Also, having scripts on two directories can lead to confusion,
> let's keep it in ci directory.
>
> ...
>> >> +#!/bin/sh -eux
>> > nit: out of curiosity, why -u (fail unset variables and parameters)?
>
>> I find it finds errors in shell scripts or when using them. E.g. typo's
>> in env variable names. I just include it wherever possible.
>
> +1, maybe we should add it to the current ci scripts as well (+ use params
> instead of setting it via set command, it should work in dash and busybox shell
> as well).

Yeah, I think it should be the default.

>
> Kind regards,
> Petr


-- 
Thank you,
Richard.

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp