Re: [RFC] cygport pm for managing package releases

[Brian Inglis <Brian.Inglis@SystematicSw.ab.ca>]

On 2020-09-11 10:07, Andrew Schulman via Cygwin-apps wrote:
> cygport has automated a lot of the work of building and maintaining
> packages for Cygwin. But one area where it doesn't help yet is in managing
> the available releases of a package. For me as the maintainer of a dozen or
> so packages, there are routine tasks that I still find to be painful:
> 
> * Finding out which versions of a package are currently available in the
> Cygwin repositories, and which if any are marked as "test".
> * Marking or unmarking a version as "test".
> * Removing versions from the repositories.
> * Marking a package as obsolete.
> 
> All of these are still manual tasks. Most require digging through
> documentation (though that's also much improved in the last few years),
> manually editing .hint files or creating dummy package files, and manually
> uploading files to the right places in sftp://cygwin.com. It's not fun, and
> so I don't keep up with it as well as I should.
> 
> To alleviate that, I think cygport should add a set of "pm" commands to
> automate package management. For example:
> 
> * cygport pm list - list versions available in the Cygwin repositories.
> * cygport pm test - set/clear "test" status for a version.
> * cygport pm del - remove a package version from the repositories.
> * cygport pm obsolete - mark a package as obsolete.
> 
> And probably others. I think this would make maintainer's lives easier, and
> make these management tasks more reliable.
> 
> I can spend some time planning and developing this, and if others want to
> collaborate on it, so much the better. But before I start on that, I want
> to get people's comments here about whether:
> 
> * It's worth doing;
> * (More to the point) It'd be likely to be accepted upstream, assuming the
> implementation is satisfactory; and 
> * There may be problems in implementing it that I haven't thought of. 
> 
> I can think of a few problems or objections:
> 
> 1. The "pm" commands will bake into cygport logic that's specific to how
> the package repositories and upset currently work. So if those change,
> cygport will have to change to match them. That's true, but not just for
> cygport pm - other parts of cygport, such as cygport up, are basically
> clients for upset. And at least it'll centralize the changes in one place,
> so maintainers won't have to worry about them.
> 
> 2. "pm list" will require finding and parsing an appropriate setup.ini
> file, unlike the other "pm" commands which will manipulate
> sftp://cygwin.com. 
> 
> I think these are surmountable, but I want to know if there's a general
> agreement that it's worth doing.
> 
> BTW a successful example like this one is the "cygport up" command, which
> we added a few years ago to automate uploading packages to cygwin.com. I
> think it's working well.

Also cygport commands package-test and all-test added more recently should make
test package management easier.

I and others have forked and hack updates to apt-cyg (static) on github, and
push changes to our forks when they are stable enough, and find that allows me
to easily do a lot with setup.ini and installed.db.

Perhaps we should first outline the maintainer packaging workflow, including
required functions such as creating a package directory and contents, checking
for other repos with a package, sending ITPs/ITAs and SSH public keys, checking
licensing, checking for new upstream releases, and less common functions such as
those you mention, sending upstream emails and submitting patches or PRs, and
others, with a summary like the cygport --help output, description as in the
cygport HTML help, specification of what is needed and why, and example commands
to execute, if known, to implement the function.

Perhaps the additional functions and commands could be guided by the functions
provided for Gentoo portage tools, Debian dh, Fedora fedpkg, and OpenSuSE osc
packaging tools.

Other functions could be like OpenSuSE osc submitrequest, requestmaintainership
to send standardized ITPs/ITAs, Debian equivs to create trivial package files,
lintian/fedpkg lint to check package metadata, etc.

Once we have a list of maintainer functions, we should consider what maintainers
consider pain points to add as cygport commands, plus quick and easy wins to
help maintainers contribute while understanding cygport development and docs.

I like the single word commands in cygport and other tools e.g. your pm list is
like apt show, pm test and pm obsolete remind me of apt mark, and possibly also
pm del, depending on whether you expect those commands to work on .hint files or
upload directories or both, and distinctions and expectations like those need to
be explained.

Should we work with patches, PRs, a dev repo (on sourceware? or github) against
https://cygwin.com/git/?p=cygwin-apps/cygport.git, or other storage space(s), to
maintain lists of workflows, suggested functions, maintainer pain points,
possible commands, command summaries, outlines, help descriptions, spec details,
and commands to execute in an implementation.

-- 
Take care. Thanks, Brian Inglis, Calgary, Alberta, Canada

This email may be disturbing to some readers as it contains
too much technical detail. Reader discretion is advised.
[Data in IEC units and prefixes, physical quantities in SI.]