[Libguestfs] fixes for building guestfs-tools and virt-v2v against a fresh libguestfs

[Richard W.M. Jones]

On Mon, Sep 06, 2021 at 12:57:56PM +0200, Laszlo Ersek wrote:
> Hi,
> 
> I'll post two small patch sets as followups to this "higher order" cover
> letter. The goal is to build guestfs-tools and virt-v2v against a
> freshly built libguestfs (git checkout), using libguestfs's "run"
> script. The modifications affect two projects, libguestfs-common and
> guestfs-tools, hence the two (upcoming) patch series.
> 
> libguestfs-common is tricky because it is consumed as a submodule by
> three git superprojects (libguestfs itself, then guestfs-tools and
> virt-v2v). Yet more trickily, the three superprojects don't all seem to
> consume libguestfs-common at the same submodule git commit, at the
> moment. (libguestfs and guestfs-tools consume the submodule at older
> commit 74bc5c5c5cb4, while virt-v2v consumes the submodule at current
> HEAD commit 6d26b6eac9da.)

This is because I ad-hoc update the common submodule when I need to.

In theory all the projects should work against the latest common
submodule, and a commit against any of them which just bumps the
common submodule to the latest version should always be valid.

> For verifying the libguestfs-common updates, I first attempted various
> git-pushes into the submodule checkouts of the superprojects. That
> proved super unwieldy, as it would require a whole lot of git massaging
> just to (incrementally) rebuild all three superprojects after each
> libguestfs-common update. Instead, in each superproject, I created a
> development branch, and as first commit on that branch, I removed the
> submodule altogether, and replaced it with a directory tree of symbolic
> links into my stand-alone libguestfs-common worktree -- refer to the
> "lndir" command. I didn't expect creating new files in
> libguestfs-common, so once established, the symlink set could be
> considered final. Furthermore, "lndir" was required for symlinking
> individual regular files: symlinking subdirectories from
> libguestfs-common into the superprojects' "common" directories doesn't
> work, as the "common" content refers back to the superproject, via
> relative pathnames such as "../blah". For such pathnames to work, the
> deep "common" directory structure actually needs to exist within each
> superproject, only the leaves (the regular files) can be replaced with
> symlinks. Thankfully, "lndir" implements just that.
> 
> This effectively bumped libguestfs and guestfs-tools to
> libguestfs-common HEAD commit 6d26b6eac9da, as a basis for the needed
> libguestfs-common fixes.
> 
> With the "libguestfs-common" patch set applied (on top of commit
> 6d26b6eac9da), libguestfs can be built and checked without regressions.
> Furthermore, virt-v2v too can be built and checked against the
> just-built libguestfs (using the latter's "run" script) without regressions.
> 
> The same applies to "guestfs-tools", assuming the other patch set is
> applied to it (on top of commit 9ba463545fa0).
> 
> Once these patch sets are up-stream, the three superprojects should
> advance their submodule references to the new libguestfs-common head commit.

Yup, submodules are horrible.  But a bit better than code duplication.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines.  Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v