Re: reproducible tar archives

[Simon Josefsson]

Bruno Haible <bruno@clisp.org> writes:

> Simon Josefsson wrote:
>> 1) Some of the suggested flags probably requires a newer GNU tar.  ...
>> running 'make distcheck' on old
>> systems is useful for ironing out platform-dependent problems.
>
> I always run
>   - 'make distcheck' only on my development machine,
>   - 'make check' on various platforms.
> What is the kind of bugs I could detect by running 'make distcheck'
> on Alpine Linux or Android?

If you try it you will find out, if you don't you won't :)

The problems that I recall are about cleaning up files which are only
needed on some platforms, and bugs in code that leave temporary files
around which is only triggered on exotic platforms.  Neither is likely
to cause users any real problem, and I can't recall any serious issue
that could be found this way.

Alpine generally tend to trigger assumptions about features of various
tools, and for some of my projects 'make distcheck' in a git checkout
will build a manual which doesn't otherwise happen, and having that work
on Alpine is a good test.

However 'distcheck' and even 'dist' seems to work for fewer and fewer
systems, and work mostly on modern GNU/Linux systems, so I'm gradually
giving up on making them work everywhere...

>> 2) Some choices may be opinionated, but I'm not sure which ones.  Maybe
>> PAX archives as in --format=posix?  I recall seeing some people
>> recommend --format=ustar for greater compatibility
>
> I use --format=ustar because I want the oldest (= most compatible) format
> that fits my needs.
>   - The default is --format=v7 (with Automake).
>   - But with it I got errors such as
>     tar: 
> gettext-0.19.8.1.74-72e49-dirty/gettext-tools/gnulib-tests/test-term-ostream-xterm-linux-mandriva.out:
>  file name is too long (max 99); not dumped
>     tar: 
> gettext-0.19.8.1.74-72e49-dirty/gettext-runtime/intl-csharp/doc/GNU_Gettext_GettextResourceManager.html:
>  file name is too long (max 99); not dumped
>
> Looking at [1], --format=posix brings only features that most GNU package
> tarballs will never need.

Interesting.  I see the long filename complaints too and they are really
annoying, although I don't recall any real problem because of them.

Hmm, it seems --format=posix is actually a bit misleading: it seems that
if no features from --format=posix is necessary (like sub-second
timestamps) then it will be the same as ustar format.  The tar manual
suggest --format=ustar is subset of --format=posix, although not
perfectly clear: "However, any tar implementation able to read ‘ustar’
archives should be able to read most ‘posix’ archives as well".

/Simon

signature.asc
Description: PGP signature